In this week’s edition of Coconut Capitalists, we’re diving into:

• Krutrim & The State of India’s AI Race

• Korea’s Kakao & OpenAI Partnering Up

• Quick-Fire Startup News from Around Asia

Let’s get into it.

Krutrim & The State of India’s AI Race

The Scoop

Bhavish Aggarwal, the founder of Ola Group - an India-based conglomerate that recently peaked at a $7.3 billion valuation - has invested $230 million of his own money in Krutrim, an AI lab that he started in 2023 as a local rival to OpenAI. This recent move by Krutrim provides a window into India's rapidly evolving AI landscape, and its impact on a country with 1.45 billion people.

India is remarkably diverse, with 22 official languages and 19,500 dialects, making it the most multilingual country in the world. Yet only 1% of global internet data that’s used to train large language models consists of Indic languages.

This gap represents a massive opportunity: a company that invests in the creation of a new Indic language dataset will gain a significant advantage in developing India’s top-performing foundation model. This new dataset would allow the model to tap into local linguistic nuances that are often missed by western AI firms. Additionally, India offers an opportunity to serve a market with over 820 million monthly internet users - and internet usage is growing at 8% per year. Simply put, there’s no other market in the world like India.

Where the AI Race Currently Stands with local players

According to Crunchbase data, over a dozen startups have raised more than $10 million to build foundation models in India. And here are three of the most well-known:

1. Krutrim AI is founded by Bhavish Aggarwal (as noted above), who has a net worth of $2.2 billion. The company is building Full-Stack AI that includes: foundation models (like OpenAI & Anthropic), infrastructure-as-a-service (like Replicate & Together AI), and custom training/inference chips (like Nvidia & AMD).

2. Sarvam AI is founded by a former Microsoft AI Researcher & a Carnegie Mellon University PhD. The company has raised $41 million in funding from firms like Peak XV (fka. Sequoia Capital SEA), Lightspeed, and Khosla Ventures. The firm builds world-class Indic language models and text-to-speech services. They’ve also developed autonomous agents that can operate telephones and WhatsApp.

3. Ema AI is founded by the former Chief Product Officer at Coinbase, and has raised $61 million led by Accel. The company builds universal AI employees for enterprises across India, automating tasks like customer support, marketing, and legal assistance. The firm is also reportedly developing Indic foundation models.

Where the AI Race Currently Stands with international players

Now, let’s shift our focus to the international players targeting India’s 1.45 billion population. And in doing so, first let’s analyze AI consumer app usage data from Semrush.com for December 2024 in India:

1. OpenAI with chatgpt.com had over 60 million unique visitors

2. Google with gemini.google.com had over 9.7 million unique visitors

3. Perplexity with perplexity.ai had over 2.1 million unique visitors

4. Anthropic with claude.ai had over 1.6 million unique visitors

5. Meta with meta.ai had over 582 thousand unique visitors

We can clearly see that among consumer AI adoption, OpenAI has the lead. However, recent evidence has shown that any single startup has the potential to catch up in India’s AI race. For example, if we transition from website usage to mobile app usage, the Chinese company DeepSeek now sits at #2 on the iOS App Store and #9 on the Android App Store in India. DeepSeek’s ascension to top 10 status comes just one month after it wasn’t even ranked in either app store’s top 100.

Moreover, it’s worth noting that the statistics above focus solely on consumer AI adoption, while India is also home to over 47 million active businesses as of 2024. Given this scale, AI likely won’t represent a winner-take-all opportunity (at least not in the short term). Instead, a winner-take-most outcome is more likely.

Why it Matters

As China continues to block U.S.-based AI companies from operating in the country, and as Chinese-based AI companies face growing scrutiny in the U.S., both U.S. and China - the current leaders in the AI race - are now vying to capture India as their second largest market.

India, with its massive population of over 1.45 billion, is the world’s most important consumer market after the U.S. and China. However, the question remains: can a U.S. company like OpenAI or Perplexity become the sustained market leader across India, or will a Chinese competitor such as DeepSeek, Alibaba, or Tencent make a splash?

And let's not overlook homegrown startups like Krutrim, Sarvam, and Ema. Although still in the early stages, these companies have an incredible talent density filled with India’s best and brightest - many of whom are IIT graduates operating as AI researchers in their labs. And who knows? - maybe when selling to enterprises, an India-native startup with its domestic expertise could prove to be the winning strategy.

Korea’s Kakao & OpenAI Partner Up

The Scoop

OpenAI & Kakao announced a strategic partnership during a joint press conference that included Sam Altman from OpenAI and Shina Chung from Kakao. The event was held at the Plaza Hotel in Jung-gu, Seoul.

If you’re new to Kakao, it’s South Korea’s most popular messaging service with 94% of Koreans over the age of 13 using the app every day. It’s functionally similar to WhatsApp or Telegram, but the look & feel of Kakao is extremely unique compared to the others. For one, the app is incredibly playful, filled with virtual characters called “Kakao Friends”, that pop up throughout the in-app experience. And second, it doubles as a super app - offering integrated services like mobile payments, ride-hailing, and e-commerce.

But getting back to the press conference, here’s the three major announcements:

1. Kanana, a personalized AI assistant, native to the Korean language. It’ll initially operate as a standalone app and be powered by OpenAI models. And reportedly, at a later date, the Kanana AI assistant will also be integrated into KakaoTalk.

2. Kakao’s B2B Integration with OpenAI’s APIs. Kakao has a number of B2B products for advertising, customer support, and in-app communication to help Korean businesses grow. OpenAI’s leading models are expected to be integrated into Kakao’s various offerings, while the specific details are yet to be announced.

3. Kakao & ChatGPT Enterprise. The ChatGPT Enterprise product will soon be adopted across Kakao’s employee base. It’s worth noting that the enterprise version of ChatGPT reportedly costs on average $60 per user, per month.

Why it Matters

South Korea has a thriving AI startup ecosystem. For instance, there’s the AI chip startup Rebellions, which was last valued at $1.5 billion after its merger with SK’s Sapeon. Additionally, there’s AI infrastructure-as-a-service startups like Friendli AI and Vessl AI with millions in funding. And finally, there’s foundation model startups like Upstage AI and Twelve Labs that are overflowing with world-class AI researchers.

It’ll be fascinating to watch how the international AI firms both collaborate and compete with the domestic AI startups over the next few years in Korea.

🇸🇬 Singapore News

Steven Hoi, a leading AI scientist in Singapore, has joined Alibaba’s AI team as vice president of the AI Consumer Apps Division. Previously, he led Salesforce Research Asia as managing director and served as vice president of AI Research at Salesforce.

Hoi was also a tenured professor at Nanyang Technological University and Singapore Management University. He has published over 300 AI papers and was named an IEEE Fellow in 2019, the highest honor from the world famous Institute.

ByteDance has unveiled OmniHuman-1, an AI model that generates realistic videos of people speaking, singing, and moving. It requires only an MP3 file and a single reference image to produce a hyper-realistic AI video (aka a deepfake).

The model is comparable to Sora and most similar to companies like HeyGen and Synthesia. Its most striking feature is the pixel-perfect video quality and nearly flawless physics engine powering the model. However, it’s not open source.

🇨🇳 China News

DeepSeek has surpassed ByteDance’s Duobao in daily active users across China. In January, DeepSeek’s AI chat app recorded 22.2 million daily active users compared to Duobao’s 17 million, according to data I verified on China’s CNZZ.

Moreover, on the first workday after Lunar New Year, a surge of jobseekers gathered outside DeepSeek’s headquarters, resumes in hand, seeking employment. One candidate, surnamed Shen, drove four days from southwestern Sichuan to Hangzhou to apply. He called DeepSeek “the nation’s pride” for its AI achievements and said he would accept any role - ”whether as a cleaner or a driver”.

The Trump administration in the U.S. has shut down the de minimis rule; a loophole that companies like Shein and Temu used to import Chinese goods valued under $800 duty-free and without full customs checks. Now, by executive order, Chinese goods - even if valued under $800 - must pay a 10% tariff and undergo full customs clearance procedures. Notably, the de minimis rule remains in effect for imports from other countries, only China is being targeted.

According to Wall Street analysts, a company like Shein - which generated over $48 billion in revenue and likely $1–2 billion in net profit in 2024 - should expect a 15–20% drop in revenue in 2025 under current economic conditions.

Physical Intelligence, a Silicon Valley–based startup that builds general-purpose AI for robotics, has released Pi0, an open-source foundation model that translates natural language commands into physical actions. The company has raised $400 million from investors including OpenAI, Jeff Bezos, and Sequoia Capital.

While Chinese firms have earned widespread praise in the West for open-sourcing many state-of-the-art language models, the core AI models powering robotics companies like Unitree and UBtech remain proprietary.

Rokid, known for its AI-enabled AR glasses in China - comparable to Meta’s Ray-Ban Smart Glasses or Snap’s Spectacles - has integrated DeepSeek R1 into its product. The glasses look like conventional eyewear but deliver advanced features such as real-time translation, navigation, instant messaging, and photo capture capabilities. Please note that DeepSeek R1 runs server side, so an internet connection is required.

Tencent Cloud has launched an AI API that combines DeepSeek’s R1 with Real-Time Web Search. It’s a similar offering to Perplexity’s Sonar API Service which provides developers with the R1 language model in combination with search capabilities.

🇰🇷 Korea News

Cinamon has raised $7.61 million for its AI-powered 3D video creation platform. Its product, CINEV, allows you to auto-populate a 3D world and then use generative AI models to customize every aspect of the experience. It’s incredibly cool—it feels like an in-between of Pika Labs and the Unreal Engine from Epic Games.

In this week’s edition of Coconut Capitalists, we’re diving into:

• How China is recreating Silicon Valley’s Magic

• Perplexity, & a possible Merger with TikTok U.S.

• Quick-Fire Startup News from Around Asia

Let’s get into it.

How China is recreating Silicon Valley’s Magic

The Scoop

China is massive, home to over 1.4 billion people and 120 cities containing at least 1 million people. Yet, the two AI labs with the current state-of-the-art, open-source models – DeepSeek (known for R1) and Alibaba’s DAMO (known for Qwen) – are within walking distance of each other.

The city of Hangzhou, about 45 minutes south of Shanghai by bullet train, is home to one of the world's highest concentrations of AI researchers. In fact, you don't have to visit to believe it - the Chinese Government has officially named the city's main engineering district "AI Town."

If you’re reading this newsletter from the U.S., maybe you’re not familiar with Hangzhou but rather with Shenzhen, known in the western world as the “Silicon Valley of China.” However, this statement is incorrect. Shenzhen, which is located just 14 minutes north of Hong Kong by bullet train, is not like today’s Silicon Valley but rather resembles the hardware-driven Silicon Valley of the past (back in the days of Shockley & Fairchild semiconductors). Today’s Silicon Valley is now software-driven - bits instead of atoms, while in contrast, Shenzhen is loved by hardware-driven companies like BYD (vehicles), Huawei (phones), and DJI (drones). The reason being that there’s nowhere else in the world where you can go from design to production in a matter of hours. For many Chinese companies, their entire supply chain sits within a 10-mile radius of where they operate.

In contrast, Hangzhou - specifically the "AI Town" district - resembles today's software-driven Silicon Valley in the U.S. And while DeepSeek and Alibaba's DAMO lead the charge, 12 other AI labs, each with over $10 million in VC funding, are located in "AI Town" as well (according to a local Hangzhou reporter I talked with on WeChat).

A Deep-Dive into AI Town

Spanning over 2.13 miles (3.43 km), the district is like a massive college campus. But instead of psychology, math, and social studies students scurrying around, it’s filled with AI engineers, researchers, and founders. Here are just a few highlights of AI Town and why many of the best & brightest in China move there:

• First, individual researchers are often given free "computing vouchers" from the Chinese Government & Alibaba Cloud worth between $500k USD and $4 million USD to fund their initial model development.

• Second, there are over 17 academic research institutions with labs in the town, creating a cross-pollination of the best research ideas.

• Third, the government supplies AI Town with one of the world’s fastest internet infrastructures, delivering peak speeds of 10Gbps. For comparison, in the U.S. the average download speed is 186Mbps.

• Fourth, Luckin Coffee, which makes a kickass Coconut Latte, will reportedly deliver your morning beverage via drone delivery. This way an AI researcher can stay close to their best friend - the H800 GPU - at all times.

It's worth noting that in Hangzhou, next to AI Town - where both DeepSeek and Alibaba's DAMO are located - there's also Dream Town. Yes, Dream Town. I know…this is starting to sound more like something you'd find at Shanghai Disneyland.

For our friends in the U.S., Dream Town is China’s version of Sand Hill Road in Palo Alto or South Park in San Francisco - a playground of sorts for Chinese venture capitalists. But Dream Town is home to more than just VCs; it’s divided into two areas: the “Internet Village” (solely for entrepreneurs, with over 5,210 startups working from WeWork-like complexes) and “Angel Village” (which houses over 180 VC firms).

Why it Matters

It’s worth noting that the city of Beijing arguably has more raw AI firepower, with unicorn-status startups like Baichuan ($1.4 Billion raised), Zhipu ($1.17 Billion raised), Moonshot ($1.0 Billion raised), and 01.AI ($400 million raised), all having their primary research labs in the city. However, if we talk about global impact, it’s DeepSeek with R1 and Alibaba’s DAMO with Qwen2.5-72B-Instruct that have really stolen the spotlight as of late (this is because open-source models often gain rapid adoption online within engineering communities, and subsequently spread like wildfire on places like Twitter & Reddit).

As Y Combinator’s Paul Graham stated in a 2006 essay, all you need to recreate Silicon Valley in a new location is two things: First, lots of money (e.g., Angel Village) and Second, lots of nerds (or in other terms, researchers & engineers) (e.g., AI Town & Internet Village). Many technology hubs like Tel Aviv (Israel), Bengaluru (India), and even cities in the U.S. like Austin & Miami have set out on a quest to recreate the magic of Silicon Valley, but Hangzhou seems to have the most pixie dust so far.

Perplexity, & a possible Merger with TikTok U.S.

The Scoop

On January 20th, 2025, newly elected U.S. President Donald Trump signed an executive order to delay the "TikTok ban" for 75 days. Ideally, this would allow the new administration time to negotiate a four-way deal between ByteDance, the Chinese Government, itself (the U.S. Government), and a third-party U.S.-based M&A candidate (acquirer) that would take over domestic operations in the U.S.

Among the many companies that have thrown their hats in the ring, Perplexity's proposed merger may be the most compelling. For context, Perplexity is the $9‑billion-dollar U.S.-based AI search startup that wants to merge with TikTok U.S.

A Quick Recap of the Events

Just to catch you up to speed; here's a timeline of recent events regarding TikTok U.S. and how it relates to Perplexity.

• First, on January 18th, 2025 - U.S.-based Perplexity submitted a confidential bid to ByteDance, proposing a merger between itself and TikTok U.S.

• Second, on January 19, 2025, the TikTok Ban went into effect in the U.S. This created a cascade of events; the app was pulled from the Apple and Google App stores, TikTok was shut down for 14 hours, then finally, Donald Trump, on his first day in office, signed an executive order allowing for a 75-day extension on the bill that initially banned TikTok.

• Third, on January 27, 2025, Perplexity updated its bid, in an attempt to create a more compelling offer for the Trump Administration - now proposing a U.S.-based holding company called "NewCo" that would combine TikTok U.S. (excluding its core recommendation algorithm) and Perplexity AI.

This updated bid from Perplexity, at least on paper, is quite compelling for the Trump Administration - let me explain - first, I need to give some background context; in a recent press conference, President Trump said, and I quote: ”Without U.S. approval, there is no TikTok. With our approval, it is worth hundreds of billions of dollars - maybe trillions. Therefore, my initial thought is a joint venture between the current owners and/or new owners whereby the U.S. gets a 50% ownership in a joint venture set up between the U.S. and whichever purchaser we so choose.”

The updated Perplexity offer is structured in such a way that would fulfill President Trump’s request; the federal government would have the option to acquire up to 50% of "NewCo" at a future date, aligning the incentives. Essentially, ByteDance (the Chinese parent company) would transfer TikTok U.S. assets to NewCo in exchange for equity (without any voting rights), while Perplexity’s investors would receive shares in the merged entity.

Even though the chances of a deal happening between Perplexity & TikTok U.S. are completely unknown - and that's not an indictment against Perplexity, but simply a testament to how difficult it is to get a multi-billion-dollar M&A deal across the finish line - it's still worth investigating why this deal actually makes so much sense.

Antitrust Standpoint

When we consider American companies buying the TikTok U.S. asset, the big 4: Meta, Microsoft, Amazon, and Google are probably off the table. Even though they have the financial resources to acquire TikTok U.S. and the engineering capabilities to rip out & replace the “algorithm”, there's a very slim chance that the Federal Trade Commission (FTC) will let a supposed "monopoly power" acquire TikTok U.S. And next up on the list, there's Walmart and Oracle. The leadership of both companies have close ties to ByteDance executives. As evidence, back in 2020, both Walmart and Oracle even tentatively agreed to acquire a combined 20% stake in TikTok before the Biden Administration blew up the deal. However, as of today, Walmart has a market cap of $788.56 billion and Oracle has a market cap of $475.65 billion - these are massive companies! And given their sheer size, they would probably face the same antitrust scrutiny as the big 4.

Perplexity, on the other hand, is worth $9 billion - and is currently seen as the underdog in the AI Search war between Google-Gemini and Microsoft-OpenAI. Given this, Perplexity would pass the antitrust test.

Technology Standpoint

The next question is whether Perplexity can cleanly rip out and replace the TikTok "algorithm" with a new one at a similar level of performance and quality without the average user in the U.S. ever noticing.

I use "algorithm" in air-quotes as that word makes the TikTok recommendation system sound overly simplistic - almost like hot-swapping the engine of a car - but instead the "algorithm" is an incredibly complex system, probably containing hundreds of custom AI models, graph databases built specifically for TikTok's multi-modal data structure, dynamic recommenders running billions of A/B tests in parallel, and so many more inventions that make up the core intelligence of this "algorithm".

But, the Perplexity team has some of the best engineering talent in the world, and they're battle-tested in scaling a platform to tens of millions of users. Given this, one could argue that they would pass the technology test.

Product Standpoint

Moreover, Perplexity's AI Search product is already incredibly useful today, with its ability to instantaneously answer even the most complex of questions, and often in less than a second. However, the answer quality would reach a new height if Perplexity could tap into the real-time inflow of user-generated data from TikTok.

A great case study is X.AI's Grok (Elon Musk's company). Grok itself is an incredible language model, but in isolation, the model doesn’t feel like a step-change from anything Anthropic, OpenAI, or another frontier AI lab has created. However, when Grok is connecting to the firehose of real-time X/Twitter data, suddenly it shape-shifts into a truly unique product.

I’ll give an example: if you wanted to learn about the technical architecture behind DeepSeek’s new R1 Model, Grok can actually pull tweets posted by the DeepSeek research team on X/Twitter and use that text data (that cannot be found anywhere else on the internet) as input context for giving the highest quality answer possible. For some specific questions (not all questions), this firehose of user-generated content is much more useful than information coming from a publication like Business Insider or The Verge (that’s instead reporting on DeepSeek). Instead, the information comes straight from the source.

Similar to X.AI's Grok, for Perplexity, integrating its AI search product with TikTok’s firehose of user-generated content would make the answers that much better. TikTok is one of the best repositories of real-time news and events on the internet (along with X/Twitter, Reddit, YouTube…etc), and represents a treasure trove of proprietary data. Given the potential of this data integration between Perplexity & TikTok U.S., the deal would definitely pass the product test.

Why it Matters

TikTok U.S. will need to work out a deal in the coming months; there’s a deadline of April 5, 2025. We can assume that there are lots of conversations happening behind closed doors, but as far as public M&A offers that have been made so far, Perplexity’s is the only one that passes the antitrust, technology, and product tests.

Also, to my friends at ByteDance - things are probably a little hectic right now. I hope you’re doing OK & hanging in there!

🇯🇵 Japan News

Sakana AI, Japan’s leading foundation model startup company, founded by two ex-Google DeepMind researchers and the former COO of Stability AI, recently made the cover of Forbes Japan. The team has been busy at work, from creating their very own Autonomous AI Scientist to creating a framework for The Merging of AI Models to creating Shape-Shifting LLMs with Transformer² where the AI models dynamically adjust their weights to solve various tasks.

Most recently, the company released a 1.5-Billion-parameter Japanese Small Language Model with a novel distillation method on HuggingFace - called TAID. The TAID process transfers knowledge from a huge, high-performing model to a much smaller one. The significance is that the 1.5 billion parameter student model has strikingly similar intelligence to its much larger (& more expensive) teacher model.

SoftBank’s Vision Fund 2, which has $56 billion in committed capital, is reportedly planning to invest $500 million into the robotics startup Skild AI, at a $4 billion valuation. The startup, which is only 2 years old, raised a previous funding round of $300 million at a $1.5 billion valuation last July from investors, including Jeff Bezos, Lightspeed Venture Partners, and Coatue Management. The company's main headquarters are in Bengaluru, India, Pittsburgh, USA, and San Francisco, USA.

Rakuten Group, the parent company of Japan's largest e-commerce company, has announced it will not proceed with the planned initial public offering of its securities arm, Rakuten Securities Holdings Inc., on the Tokyo Stock Exchange. The larger story here is that Japan's TSE is in need of a boost - while 130 IPOs in 2024 is commendable - the Japanese financial sector needs to regain its momentum in 2025 with a big year, as competitive foreign markets, have recently been soaring.

🇲🇾 Malaysia News

Sam Altman’s company, WorldCoin (he is a co-founder), which is most well-known for creating a blockchain-based “proof of human” verification system (similar to an advanced version of Apple’s Face-ID), has secured an MoU (type of contract) with the Malaysian government to integrate WorldCoin’s iris-scanning technology into Malaysia’s digital infrastructure. The technology behind WorldCoin, called World ID, works by scanning a person’s eyeball for verification - it’s a similar concept to a fingerprint scanner used for identification systems at airports.

🇸🇬 Singapore News

The Economic Development Board (EDB), a government-backed agency in Singapore, has a Corporate Venture Launchpad program that recently had an incredibly successful spinout. 129Knots, a fintech startup based in Singapore, has launched with US$10 million. The startup is targeting the intersection of blockchain and financial services, focusing on asset tokenization, audit trails (similar to Chainalysis), and real-time trade credits.

🇮🇳 India News

Slice Bank, a Bengaluru-based fintech company & hybrid bank (hybrid meaning a digital bank with a few physical branches as well), is seeking to raise between US$250 million and US$300 million from financial investors and family offices. The company, which is valued at between US$1.3 billion and US$1.5 billion, has an annual revenue run rate of US$138.94 million. In India, getting a banking license is incredibly difficult, with the central bank rejecting most applications in recent years; however, through a recent merger with North East Small Finance Bank, the startup was able to “acquire” a banking license and is now off to the races.

🇨🇳 China News

YuE (乐), which translates to “music” or “happiness” in Chinese, was recently released as a state-of-the-art, open-source music generation model developed by researchers at the Hong Kong University of Science and Technology (HKUST). The music generation model YuE is able to generate entire songs from scratch and is in a similar category to U.S.-based startups like Suno (US$125 million raised) or Udo (US$10 million raised).

The Bank of China announced that it will invest 1 trillion yuan (about US$137 billion) over the next five years into the Chinese AI ecosystem to counter the US. This announcement comes on the heels of the recent unveiling of Stargate; the $500-billion dollar joint-venture, between OpenAI, SoftBank, and Oracle. The Bank of China, which has total assets of US$3.09 trillion, plans to offer both debt financing and equity investments into various startups & infrastructure projects.

In this week’s edition of Coconut Capitalists, we’re diving into:

• ByteDance’s AI-Powered Code Editor, the Cursor of Asia

• America’s new AI chip export controls on China & Singapore

• Quick-Fire Startup News from Around Asia

Let’s get into it.

ByteDance’s AI-Powered Code Editor, the Cursor of Asia

The Scoop

ByteDance has launched a competitor to Cursor’s wildly popular AI-powered code editor. Cursor, a San Francisco-based startup that’s worth over $2.6 billion, has over 30,000 enterprises as customers and has grown annual revenue from $4 million to $100 million in just eight months.

If you've never tried Cursor before, it's essentially like having ChatGPT built directly into your code editor. Cursor acts as an intelligent pair-programming assistant, analyzing your entire codebase & making improvements to your code's quality in real-time as you type. It's worth noting, that on average customers of Cursor have reported a 37% increase in software development speed & 22% fewer bugs.

However, while Cursor has taken off in Western markets (particularly in the US & Europe), many of the developers across Asia have never heard of the product - this gives ByteDance a massive opportunity. And they've taken it by introducing Trae AI.

AI-Native Products

On paper, ByteDance is the perfect company to build & sell this product in Asia:

• First, they have Doubao-1.5 Pro, their in-house, lightning-fast language model that can power Trae AI. While Cursor lets you plug-and-play with different AI models like Claude-3.5-sonnet & OpenAI’s GPT-4o, ByteDance can take a different approach - reducing variable costs significantly by instead building a full-stack, vertically integrated, Cursor alternative with Doubao-1.5 Pro.

• Second, ByteDance already has significant enterprise distribution. Their product Lark (which is very similar to Google Workspace or Microsoft 365) has over $400 million in annual revenue and is incredibly popular across Chinese, Japanese, and Singaporean enterprises. Packaging Trae with Lark would be a major upsell opportunity for ByteDance and would create a new revenue stream with their existing customer base.

But why can’t Cursor simply go after Asian enterprises? Why does it have to be ByteDance? Obviously, the Cursor team is very good - capturing $100 million in annual revenue & now with over 30,000 enterprises as customers - so the answer is yes - they definitely can capture this market too! However, for many enterprises operating across Asia - the issue is they may be hesitant to give direct codebase access (their most important intellectual property) to a “relatively new” American startup - and one that doesn’t have a local presence in the region. Because of this, ByteDance may have an advantage. Now, I do expect Cursor to eventually partner with a trusted technology vendor like Microsoft Azure, AWS, or GCP which all have extensive local operations across Asia, in order to gain access into these enterprises.

Why it Matters

ByteDance is not a new company. Founded in 2012, it already has a substantial portfolio of products from social media (TikTok, Douyin, etc.) to gaming (Mobile Legends) to enterprise software (Lark). Yet, even with their significant size, their product velocity is exceptional - in being able to ship dozens of new products every year. They move much faster than the other tech giants: Tencent, Alibaba, & Baidu. And their talent density rivals Western companies like OpenAI, Anthropic, & DeepMind. Cursor has the lead, but Trae has a real shot to be a worthy competitor.

America’s new AI chip export controls on China & Singapore

The Scoop

In Joe Biden's final week as US president, his administration signed a new piece of export control legislation, limiting access of Nvidia’s AI GPUs to non-American or non-allied nations. The idea behind this legislation is that it will help American AI companies & close allies of the US, to stay ahead of the competition by giving them access to the world's largest, leading-edge AI computing clusters.

The legislation (which is to go into effect in 100 days) places foreign countries—and companies based in those locations—into one of three tiers:

• Tier 1, has almost no export restrictions. These are the US's 18 closest allies like Canada, Korea, Japan, and the majority of the European Union.

• Tier 2, has limited access, allowing the equivalent of 50,000 Nvidia H100 GPUs over the next two years to each of the countries on the list (and the companies that are based there). This includes neutral countries (that seemingly weren’t considered allies or adversaries by the Biden Administration) such as Singapore, India, and Mexico.

• Tier 3, has no access - it’s a complete ban. This includes leading-edge Nvidia AI chips such as the B100s & H100s, and surprisingly even 5-year-old & 8-year-old chips like the A100s & V100s. A few of the countries are China, Russia, and Iran.

It was expected by many that Donald Trump, who recently took office as the 47th US President, would rescind the Biden Administration's legislation on AI chip exports - in order to ease the very hot global trade tensions, particularly with China. However, this did not happen - The Trump Administration has so far kept the legislation (even though the new administration did repeal a few of Biden’s AI Safety Laws). It’s worth noting that Nvidia disavowed the administration’s decision with an extremely blunt press release, where their executive team called the legislation “misguided”.

Singapore & the Issue for Tier 2 Countries

We all know you can't have more than one best friend, right? Many of us learned this on the playground as kids. And for a country like Singapore, it's even more difficult to maintain close friendships with two countries that are economic enemies. Singapore's attempt to stay neutral (as an ally of both China & the US) is showing a few cracks.

While I’m not a foreign relations expert, it’s clear to anyone living in Singapore that over the past 70 years, the city-state has done a remarkable job of staying neutral on the world stage—collaborating with everyone from Taiwan to China, from North Korea to the US—and miraculously avoiding making enemies. However, for the first time, the US is signaling concern about Singapore’s technology ties with the Chinese government. In simpler terms, the US doesn’t like that Singapore has “two best friends”, and seems to be nudging them towards an ultimatum.

Why it Matters

It’s worth noting that the 168 page piece of legislation has one loophole that might just quickly turn it into Swiss Cheese. And that’s that there’s nothing really stopping an entity from a tier-2 or tier-3 based country from spinning up a US based Delaware C corp and renting leading-edge chips - essentially running a large model training process remotely (via ssh) but on US soil.

Now, for the three major international cloud providers (Microsoft Azure, AWS, or GCP) they have the ability to KYB (know your business/background check) every company that wants to run a large training job on their GPU clusters. And they have the 1000+ person legal/compliance departments to spot obfuscated connections between a US shell company & a tier-1 or tier-2 based entity. However, for newer AI infrastructure Together AI, Lambda Labs, and the San Francisco Compute Company this could create a massive headache (they have smart folks - so they’ll probably be fine!).

However, let's say that I happened to be a Chinese National - and I had a Beijing-based company called "Colin's Super Secret AI Lab" (sounds cool, right?). I could simply create a Delaware C Corporation (as a Chinese National), and rent massive clusters of Nvidia GPUs for training through one of these (IAAS) providers (without them probably ever noticing). And if this strategy doesn’t work out, there’s even a very popular “craigslist for GPUs” service called gpulist.ai where anyone can rent access to leading-edge Nvidia chips on-demand.

🇸🇬 Singapore News

Mistral has announced its plan to open a Singaporean headquarters in 2025. At the World Economic Forum event in Davos, Switzerland, CEO Arthur Mensch stated that the company is quickly expanding its presence in APAC. For many enterprises across Asia, they require custom or on-premise AI solutions, and Mistral's hands-on deployment approach has the chance to catch fire among this customer base.

🇰🇷 Korea News

Friendli AI, one of Korea's leading AI startups, which is founded by the famous Seoul National University Professor, Byung-Gon Chun, and provides “GPU infrastructure as a service” for AI model inference & fine-tuning workloads, has settled a lawsuit with Brooklyn-based Hugging Face over patent infringement. In the suit, filed nearly two years ago, Friendli AI alleged that Hugging Face violated its AI inference patent on "batching with iteration-level scheduling."

It's worth noting that the resolution, that occurred after the settlement was complete, ended up being incredibly positive for both companies. The next day, the Friendli AI CEO announced a strategic partnership with Hugging Face, now allowing one-click deployments of almost any Hugging Face model directly onto Friendli AI's infrastructure. This is incredibly valuable for both companies - Friendli AI gets western eyeballs experiencing their products for the first time, and Hugging Face gets introduced to a market (South Korea) where they're still relatively unknown.

Upstage AI has released a new & improved AI model for converting complex charts and graphs from PDF documents into raw HTML. This unlocks an incredibly important use-case in being able to feed scattered visual information into a language model. And the product has industry-leading processing time, of only 0.84 seconds on average, per PDF document. If you're new to Upstage AI, it is Korea's leading foundation model company & they raised $72 million in their last round of funding.

SK Hynix reported $49.6 billion in revenue for 2024, more than doubling its 2023 revenue of $22.81 billion. For context, SK Hynix produces high-bandwidth memory (HBM) chips, which are critical components for leading-edge GPUs from the likes of Nvidia, AMD, Intel, and Broadcom. It's worth noting that their flagship memory chips like HBM3 and HBM3E are reportedly already sold out for 2025.

🇯🇵 Japan News

OpenAI & SoftBank have formed Stargate - a joint venture between OpenAI, the Japanese conglomerate SoftBank, Larry Ellison’s Oracle, and the Middle Eastern sovereign wealth fund MGX. The group aims to spend up to $500 billion on the financing of AI data centers - based in the state of Texas (USA) - over the next four years. It's been widely reported that SoftBank and OpenAI will each commit $19 billion of funding to the joint venture.

The scale of this data center project is unprecedented. On Wednesday, OpenAI's CEO likened Stargate to a venture fund, with OpenAI and SoftBank being two of the general partners - where the two other general partners are Oracle and MGX. Essentially the GPs will commit $45 billion in total to the project (implying that $7 billion would come from Oracle and MGX combined). Moreover, the remaining $455 billion investment into Stargate will come at a later date from investors categorized as “LPs”, in addition to several types of debt & equity financing, as told by OpenAI’s CEO.

🇨🇳 China News

Researchers from the Shenzhen Institute of Advanced Technology & Nanjing University have introduced a new AI Model called VideoChat-Flash, which aims to increase processing speeds for video understanding by 100x. The team, which released an open-source model & research paper, states that they discovered a new compression technique (Pied Piper?!?) which shrinks long videos into shorter ones through a process called "HiCo" (Hierarchical visual token Compression). The model achieved a 99.1% accuracy on the NIAH (Needle-In-A-video-Haystack) benchmark (which is a test that challenges AI models to find specific details in very long videos - similar to finding a needle in a haystack).

An example use-case for the technology could be catching bad guys via CCTV cameras. Currently, security teams around the world spend countless hours manually reviewing CCTV footage to identify criminal activities. And still, far too often criminals slip through the cracks due to the sheer volume of footage that needs to be analyzed by humans. The team behind VideoChat-Flash (along with some great startups called Twelve Labs & Anduril) has the potential to finally solve this problem.

Elon Musk met with China's Vice President Han Zheng during the week of the US Presidential Inauguration. During the meeting, Vice President Han stated that he "welcomed US companies including Tesla to seize opportunities, share the fruits of China's development, and make new and greater contributions, with ever-closer economic & trade ties between China and the United States." Elon later recapped the conversation via a post on his platform X (fka Twitter) stating that: "I have been against a TikTok ban for a long time, because it goes against freedom of speech. That said, the current situation where TikTok is allowed to operate in America, but X is not allowed to operate in China is unbalanced. Something needs to change."

🇮🇩 Indonesia News

Apple and the Indonesian Government are nearing a deal to lift a 4-month ban on iPhone 16 sales. Indonesia enforces strict requirements for tech products, mandating that 40% of components be sourced or manufactured domestically.

Apple has struggled to comply with the law, leading to an iPhone 16 sales ban that came shortly after its global launch. Negotiations now center on Apple establishing local manufacturing facilities, likely in partnership with Indonesian firms. As Apple's international growth slows, access to Indonesia's 270 million-strong consumer market (with a 96.3% smartphone penetration rate over 16 years old) - and an ever-increasing middle class - will be critical for Apple's future.

BTW, congrats to the DeepSeek AI team (a few of you read this newsletter - thank you!) on achieving Number #1 in the U.S. App Store - it’s an incredible feat 💯

In this week’s edition of Coconut Capitalists, we’re diving into:

• DeepSeek versus China’s Frontier AI Labs

• Sakana AI creates a shape-shifting AI Model

• Quick-Fire Startup News from Around Asia

Let’s get into it.

DeepSeek versus China’s Frontier AI Labs

The Scoop

DeepSeek, an AI lab based in Hangzhou, China, just fully open-sourced a competitive model to OpenAI's o1. DeepSeek's model, named R1, nearly matches the performance across almost every benchmark with OpenAI's best-in-class o1 closed-source model. This includes advanced math, language understanding, and the most complex multi-step reasoning questions.

Just to put the model's capabilities in perspective, on MMLU - a benchmark that tests an AI model's knowledge across 57 subjects including law, mathematics, ethics, and medicine - DeepSeek's open-source R1 model achieved a 90.8% accuracy rate, nearly matching OpenAI's o1 model's score of 91.8%.

In addition to DeepSeek giving away this model as a Chinese New Year present to software developers, it's worth noting that these developers can also access a hosted version of the model via DeepSeek's website. And R1 comes at a price point 90% less expensive than using OpenAI's o1 API.

Similarities between AI Firms in China & The US

It's important to note that DeepSeek is quite a different company (in a good way) compared to its Chinese competitors. In China, virtually all of the frontier labs are funded by the same two multi-hundred billion dollar valuation tech giants: Alibaba & Tencent.

In the US, maybe you'd have venture firms like Sequoia, A16Z, and Founders Fund competing to fund a new Frontier Lab, in addition to the obvious players like Microsoft and Amazon - but in China, nope - it's mostly just Alibaba & Tencent.

Just to give context, these are arguably the top 5 AI Startup Labs in China:

• Baichuan, which has raised $1.4 Billion

• Zhipu, which has raised $1.17 Billion

• Moonshot, which has raised $1 Billion

• MiniMax, which has raised $600 million

• DeepSeek, I actually don’t know

DeepSeek is the only one of the 5 not funded by Alibaba & Tencent. And it's the only one that doesn't publicly disclose its funding total.

DeepSeek is instead funded as a spinout startup by a quantitative hedge fund called High-Flyer Capital. High-Flyer Capital is a large firm, with $8 billion in assets under management as of 2025, and with over 160 employees.

So as of today, in China we have:

• The "Big 4" Tech Companies, each with their own AI Labs: Tencent (Hunyuan), Alibaba (Qwen), Baidu (Ernie), and ByteDance (Doubao)

• The "5 Dragons" of AI Startups, each with their own AI Labs: Zhipu, MiniMax, Moonshot, Baichuan, and DeepSeek

And to be fair, the Alibaba/Tencent strategy of running a corporate AI Lab while simultaneously funding competitive startup AI Labs isn't that different from the US. Microsoft has its own internal AI Lab with Phi & is backing OpenAI. Similarly, Amazon has its own internal AI Lab with Nova & is backing Anthropic.

Why it Matters

The difference between China's tech giants and the ones in the US is that Alibaba & Tencent are supposed to be competitors, but instead, they're co-investing partners - better known as BFFs. Unlike the US tech giants Microsoft & Amazon that are primarily backing a single horse in the race, the Chinese tech giants are attempting to back every horse to "guarantee" a winner.

The open-sourcing of frontier AI models in China, the US, and across the world will continue driving prices down for all software developers. It seems that when a frontier lab with a closed-source model like DeepSeek is slightly behind the leader like OpenAI, they will opt to open-source their model. This drives developer adoption, builds brand credibility, and makes it easier to hire world-class AI scientists. But is this strategy sustainable? Or will it continue driving the price of intelligence closer and closer to zero.

Sakana AI creates a shape-shifting AI Model

The Scoop

Are you familiar with the Pokémon Ditto? When it goes into battle, it can shape-shift into another Pokémon to give itself an advantage. For example, if facing a Fire-type Pokémon, it can instantly transform into a Water-type. This demonstrates adaptive learning – Ditto assesses its environment and the challenge of defeating its opponent, then transforms into the most effective form for the situation.

What if artificial intelligence could do the same, dynamically adapting to unfamiliar situations by restructuring its own architecture in real-time? This is the vision behind Transformer², a groundbreaking research paper from Sakana AI. As a quick recap, Sakana AI is Japan's leading foundation model company, having raised over $344 million from the likes of Nvidia, Lux Capital, and Khosla Ventures.

Diving into the Tech

The journey toward truly adaptive AI has seen various approaches. Traditional methods rely on human feedback through reinforcement learning (RLHF) or generating synthetic data through model interactions. Moreover, at the architecture level, the most prominent approach has been the Mixture of Experts (MoE), which works like a sophisticated switchboard - maintaining 8-32 expert networks and routing inputs between them while keeping individual parameters static. It's like having multiple specialized Dittos, each trained for a specific type of battle, rather than one that can truly transform.

Transformer² takes a fundamentally different approach, inspired by nature's adaptability. During inference, in a matter of milliseconds, it modifies its parameter values using SVD decomposition. Think of it as Ditto not just mimicking surface features but actually reorganizing its entire genetic makeup in real-time. The system can amplify or dampen specific neural pathways - for instance, potentially boosting mathematical reasoning pathways by 80% while reducing language generation capabilities by 30% when tackling a complex math problem. This mirrors how biological systems adapt: not by switching between different pre-set configurations but by dynamically modifying their entire structure to meet new challenges.

Why it Matters

Just as Ditto transforms its entire structure to match its opponent, Transformer² introduces real-time weight transformation in LLMs through SVD decomposition. While traditional approaches like LoRA add new parameters or rely on prompting - equivalent to giving Ditto new moves or teaching it strategies - Transformer² actually reshapes its core parameters during inference, similar to how Ditto morphs its entire biological structure.

But why should you care? For a company running customer service AI, this means replacing 10 different fine-tuned models (taking up terabytes of space) with a single adaptable model and a few megabytes of transformation vectors. This isn't just cost-efficient - it's a glimpse into a future where AI systems can seamlessly evolve to handle any task thrown at them.

🇰🇷 Korea News

Starlink is reportedly coming to South Korea within the next two months. While the satellite internet service makes little sense for a city like Seoul, it's a game-changer for the 10+ million Koreans scattered across rural towns. Two products from Starlink are expected to hit the market: a residential router you can mount on your house for high-speed internet, and the far more interesting product, a "direct-to-cell" eSIM service that beams satellite internet straight to your iPhone or Galaxy device.

Public reports suggest Starlink is in talks with SK Networks, KT, and LG UPlus for a partnership. The reason is simple: Starlink works great in rural areas but has trouble in cities. Partnering with local telcos isn't a new strategy for Starlink. In the US, they have a partnership with T-Mobile, and in Japan with KDDI. For Starlink, the formula is quite clear: own the rural market and team up with local carriers to handle cities.

SK Telecom has launched a new GPU-as-a-Service platform in collaboration with Lambda Labs (a US startup with $1.2 billion in funding). The service, called “SKT GPUaaS,” appears to solely target large enterprises and various departments inside of the Korean Government. This is evidenced by the lack of a public-facing website.

It’s worth noting, that the GPU-as-a-service market is heating up globally. In the US, companies like Together AI ($228 million) and SF Compute Co ($15 million) provide short-term GPU compute for training, while Baseten ($60 million), Replicate ($58 million), and Modal ($23 million) focus on inference services. Also, Korea's own Friendli ($12 million) is growing very fast - offering both inference and fine-tuning services to companies from small startups to large enterprises.

Samsung and OpenAI are reportedly developing a smart TV that integrates ChatGPT directly into the TV’s UI Layer. The AI enables natural voice conversations about what you want to watch with ChatGPT, which will in parallel analyze your viewing history across Netflix, HBO, Disney+ and other streaming services to make recommendations. When you find something you like, ChatGPT then takes control of the TV’s interface to open the right app and start playing your content automatically.

🇯🇵 Japan News

Nintendo has announced the Switch 2 will launch in 2025, marking eight years since the original Switch's 2017 debut - an unusually long gap between generations. The original Switch was a massive success, generating $84 billion in lifetime revenue and becoming the third best-selling gaming device ever with 148 million units sold, trailing only the Nintendo DS (154 million) and PlayStation 2 (160 million).

🇸🇬 Singapore News

Singtel and Perplexity have announced a partnership targeting the Singaporean population. The AI Search startup, last valued at $9 Billion, is offering a free year of Perplexity Pro to all Singtel customers. Perplexity combines OpenAI or Claude's less expensive models with real-time data, while its Pro version unlocks access to more sophisticated and expensive AI models. This move by Singtel puts Perplexity in direct competition with OpenAI's SearchGPT and ByteDance's Doubao, which are already popular AI search products in Singapore.

🇮🇩 Indonesia News

Grab and BYD have announced a major partnership, with Grab planning to purchase up to 50,000 electric vehicles to expand its Southeast Asian fleet, particularly in Indonesia. The deal showcases BYD's aggressive regional expansion - for context, the Chinese automaker is investing $1 billion in an Indonesian plant capable of producing 150,000 EVs annually by late 2025.

🇨🇳 China News

Tsinghua, Fudan, and Stanford students jointly released an AI agent development framework called "Eko". The Eko framework can take over users' computers and browsers to perform various tasks on behalf of humans. Developers can potentially use the framework as a drop-in replacement for bare-bones web automation tooling like Playwright or Puppeteer (or in conjunction with). The framework could be best described as a competitor to Langchain or Browser-use.

The open-source TypeScript framework was released under Fellou AI's GitHub organization, which appears to be a startup possibly based in China. Additionally, it seems that the core team of researchers are simultaneously attending university - that's badass!

The US Government has blacklisted 25 Singapore and China-based AI companies for allegedly collaborating with the Chinese military. A notable company on the blacklist is Zhipu AI, one of China's leading foundation model startups. This blacklist is worth noting, as the companies on the list will face significant challenges accessing Nvidia training chips, and connecting to US-based software and tools will become virtually inaccessible. For instance, even basic resources like pip or npm packages created by US-based entities would theoretically be restricted.

In this week’s edition of Coconut Capitalists:

• SK Telecom unveils “Aster”, a personal AI Assistant inspired by “Her”

• A deep-dive into ByteDance, the eight-headed monster & a future without TikTok

• Quick-Fire Startup News from Around Asia

Let’s get into it.

SK Telecom unveils “Aster”, a personal AI Assistant inspired by “Her”

The Scoop

At the CES Conference in Las Vegas, SK Telecom - South Korea's largest telco - showcased its secretive AI agent "Aster." This personal AI assistant will soon be available to the masses in March of 2025 through an iOS and Android app.

SK Telecom's pitch is that products like ChatGPT are only capable of helping people with the planning phase of their daily tasks. For instance, take vacation planning: Aster's vision is to have a flexible AI agent that could instantly transform itself into your personalized travel agent - one that not only plans your next trip, but also navigates the internet (on its own) and books your end-to-end vacation itinerary.

Aster would start by autonomously booking your hotel rooms, flights, and even taxi rides based on historical knowledge about your travel preferences. Aster can also handle dinner reservations, coordinate daily travel excursions, and even magically take care of niche tasks like finding nearby ATMs to pick up local currency.

However, this concept of having AI agents complete end-to-end tasks isn't exactly new - many of the largest AI firms have spun up 50+ person teams over the last few years dedicated to building fully autonomous agents that can interact with the web & do all sorts of things on your behalf. For instance, Google DeepMind has a project that, funny enough, is called "Astra" - with a similar promise to SK’s "Aster." OpenAI has a project codenamed "Operator," that’s to be announced in the coming weeks, and even Anthropic recently introduced a product called "Computer Use" that allows their LLM "Claude" to control your computer screen and take actions on your behalf.

Breaking Down Aster

But getting back to Aster, the main difference between SK Telecom’s AI Agent and the rest of the pack is their vision of turning "Aster" into a consumer brand. If you take a quick peek at Aster’s website: www.asterapp.ai, you can see that the brand's UI/UX is incredibly warm, inviting, & consumer-friendly. And I give kudos to the product managers at Aster for thinking through the minor details of the experience, like credit card uploads - so the AI agents can quickly access your financial details when making online purchases - as well as timely push notifications to keep you on schedule, and personalized recommendations that are based on historical data of your past actions.

Also, it’s worth pointing out that Aster's branding reminds me a lot of www.pi.ai. For my non-American friends who might not be familiar with Pi - this was the very hyped, consumer chatbot product built by the folks at Inflection AI in Palo Alto. If you’re new to Inflection, the company’s founders are Mustafa Suleyman (creator of DeepMind) and Reid Hoffman (creator of LinkedIn). However, Pi ended up being a massive flop in the end. Inflection AI raised over $1.3 billion, couldn't find product/market fit in the face of competition from ChatGPT & Gemini, and fast-forward to now, most of the original Inflection engineering team has been acqui-hired by Microsoft.

But just because the branding of Aster & Pi are strikingly similar, I’m not saying that Aster will suffer the same fate as Pi. Aster, for one, has a much bolder vision: integrating not just the planning capabilities of LLMs, but turning them into autonomous agents that can complete the most complex of tasks, on your behalf. If Aster can actually pull this off, their total addressable market is literally billions of people - it's an app that gives everyone more of their most precious resource; time.

An elephant in the room, the internet hates is not a fan of Bots

However, for a product like Aster to work, it will need to interact with travel websites like Airbnb, Expedia, & Trip.com. But these companies hate bots. As a quick stat, over 87% of the top 1,000 travel websites employ bot-detection - essentially JavaScript code - in their website, in an attempt to block any non-human interactions. Based on this one - very blunt - stat, we can probably assume travel website’s won’t welcome our AI friends (like Aster) with open arms anytime soon….

But why is this? And is there really a difference (for example) between a human using a credit card to book a room on Airbnb versus the human’s AI assistant booking the room instead? The answer as it turns out is YES. Let’s break down why:

Consider what happens if we shift to a world where the majority of a travel service’s website traffic comes from AI agents instead of humans. In such a scenario, the very user experience moats and brand goodwill that each of these travel sites has built up over many years would completely disappear. Overnight, their products would transform from content-rich applications where hundreds of designers have put in blood, sweat, and tears (I’m being dramatic, I know) to tweak every little button color, making the experience perfect for human interaction - but for AI Agents, they interact with nothing more than an ocean of backend APIs.

This would cause the unique value proposition that differentiates sites like Airbnb, VRBO, and Hotels.com to drop near zero. To an AI agent, these are all just commodity services. Take Airbnb: while humans appreciate its beautiful user interface, stellar customer support (unless I happen to be the customer...), and key features like the interactive map showing available properties - these are all elements that currently make the product superior to its competitors - but to an AI agent, this carefully crafted user experience becomes irrelevant, reduced to simple API endpoints.

Given that there’s currently an incentive misalignment between many websites and the creators of these AI agents, certain startups see an opportunity to strike. Take www.browserbase.com which helps AI agents look more human and avoid bot detection. They’ve raised over $28 million in funding on the premise that your agents can bypass captcha restrictions, IP bans, Playwright headless browser detection, and everything in between. And even though there's no way sites like Airbnb are fans of these startups, they do represent legitimate workarounds at the moment.

To better understand the technical complexities at play, let's dig slightly deeper into the details (apologies ahead of time if you're not a Computer Science nerd like myself). Today, there’s two primary ways that AI agents interact with websites. Using Aster as our example, let's say we want to have Aster book a Grab Taxi on our behalf. Aster is (most likely) presented with two distinct approaches:

• The first option is Aster navigates to Grab.com with a browser automation tool like Playwright or Puppeteer. From here, the AI agent develops a sequence of website actions (like button clicks or filling out input boxes) that eventually lead to a taxi getting booked. In this example, Aster may use a language model API like OpenAI's 01, with a vision encoder. Aster would send OpenAI’s endpoint two critical pieces of information:

  • Screenshots of the grab.com website (showcasing all the different parts of the website’s page that the AI agent could interact with)

  • The raw HTML of the website's codebase (With this, OpenAI's API could respond with the exact HTML snippet that needs to be clicked on)

• The second option is Aster could reverse engineer the sequence of API calls on the Grab.com website, which leads to booking a taxi. Making these API requests versus taking the first option of multi-step DOM node interactions, saves a ton of money on compute costs (as you don't have to load entire web pages), but for sites with top-tier detection systems, this approach usually screams "I'M A BOT."

However, there might be a third, more collaborative path forward that's a little less "move fast & break things." SK Telecom is in a unique position to leverage its brand credibility to create partnerships with these travel service providers. While industry leaders like Airbnb, for example, might hesitate initially, their competitors may see an opportunity. For example, Airbnb's 2nd or 3rd largest competitor may view a partnership with SK Telecom's Aster as a chance to quickly gain market share.

This situation perfectly illustrates the classic innovator's dilemma: market leaders often hesitate to take risks on new user experiences that might completely disrupt their current business model. But if they don't take the risk, a competitor may seize the opportunity instead, and eventually surpass them as the new incumbent.

Why it Matters

While my comparison to the innovator's dilemma might suggest a slow adoption by industry leaders, the hypothesis could actually turn out to be very wrong. Maybe a travel company like Airbnb will hop on the AI Agent Train much sooner than we would expect. It’s hard to know!

We can, however, take a look at an example of an innovator's dilemma that allegedly occurred between The Wall Street Journal (a prominent news publication in the US) and a few Silicon Valley based AI startups. Supposedly, when these startups were small, they could scrape publications like The Wall Street Journal on a daily basis from a single paid subscription account—storing thousands of articles in their internal databases (I read this on Y Combinator’s Hacker News, so I can’t fully confirm). But allegedly, this allowed them to surface bits and pieces of the original articles’ content thousands of times per day—because they no longer needed to scrape the WSJ ad-hoc for each user query in real-time (which would trigger bot detection). Instead, they could simply ping their own database for a copy of the WSJ’s content.

Eventually, The Wall Street Journal discovered this practice and filed multiple lawsuits. However, smaller publications (the WSJ’s competitors) saw this as an opportunity. Places like Fortune, Time, and The Independent sought out official partnerships with these startups, granting legitimate access to their content—while The Wall Street Journal got bogged down in legal disputes. Many would argue this is a textbook example of the innovator’s dilemma.

So does this foreshadow anything specifically related to Aster? I’m not sure. But Aster will need to develop a business model that incentivizes all parties. By the way, what Aster is attempting to build should be applauded, and I’ll be rooting for them from the sidelines—화이팅!

An uncovering of ByteDance, the eight-headed monster + a future without TikTok

The Scoop

For our friends who haven’t been keeping up with America’s quest to ban force a divestiture of TikTok to a US Company, here’s a quick recap of the two main events that led to this “ban or sell” scenario:

1. First, in June of 2022, the House of Representatives (one of the three branches of the US government) began an investigation into TikTok & its relationship with the Chinese Communist Party. This came after multiple software engineers turned whistleblowers alerted the House Intelligence Committee that US citizen user data had allegedly been accessed by ByteDance employees in Beijing. Overseas access to a US citizen's data is not usually a national security concern, but given China's National Intelligence Law stating that "companies operating in China must cooperate with the government and provide all data upon request," the US found this to be concerning in regards to TikTok. Now, one could argue that companies like Temu (PDD), Tencent, and Alibaba are also based in China + handle large amounts of US user data - so why aren't they being targeted? Well, in the US government's report, TikTok user data has been deemed highly sensitive for a number of reasons, including the ability to 1) access real-time location data and 2) listen to a user's audio conversations at any time (allegedly).

2. Second, in April of 2024, after months of hearings that included TikTok’s CEO Shou Zi Chew testifying in front of the US Congress for 4+ hours as well as a few of ByteDance's US investors being subpoenaed to appear on Capitol Hill, the investigation finally concluded. In the end, congress decided to enact the "Protecting Americans from Foreign Adversary Controlled Applications Act," which required ByteDance to divest TikTok's U.S. operations by January 19, 2025, or face a nationwide ban of the app.

But without TikTok, where does this leave ByteDance?

In a very good spot actually. The $300 Billion Dollar parent company of TikTok—ByteDance—will probably end up just fine. Just like the legendary 8-headed Japanese monster known as “Yamata no Orochi” - if you chop off one of ByteDance’s heads - you still have 7 more to go.

Outside of TikTok, ByteDance has 7 other products heads you should pay attention to:

• 1. Douyin, the China-only version of TikTok has 748 million monthly active users where the average users spend almost 2 hours on the app per day.

• 2. Capcut, an AI powered video editing tool for iOS, Android, and Desktop. Think of the tool as an AI-ified version of Adobe Premiere Pro or Apple Final Cut Pro, but with a massive user base of over 320 million monthly actives

• 3. Toutiao, the information-sharing platform built for the Chinese population (similar to Google or Apple News) has over 292 million monthly active users

• 4. Doubao, the ChatGPT competitor that has seen rapid growth in China & throughout South East Asia, has over 74 million monthly active users

• 5. Lark, the enterprise collaboration suite (strikingly similar to Google Workspace or Microsoft M365) has over $400 million in annual recurring revenue

• 6. Lemon8, a social media app focused on beauty, fashion, & food that’s targeted at Southeast Asia and Japan, has over 16 million monthly active users

• 7. Melolo, is a newly launched mobile app that offers short movie dramas from 2 minutes to 10 minutes long, where the content is produced exclusively by ByteDance’s in-house creative studio. It's a similar concept to Netflix Originals or more-so Quibi… if you remember what that is haha).

Why it Matters

If TikTok, the viral social media app that 170 million Americans use each month, shuts down in the US, then obviously it's bad for its parent company ByteDance, but the question remains: how bad is it really? This is why highlighting ByteDance's 7 other heads is critical. Additionally, we haven’t even mentioned ByteDance’s world-class AI Research team that rivals OpenAI & DeepMind - so maybe the 8 heads do remain…

And if TikTok gets banned in the US, one could argue that it's losing only half a head, not a full one. TikTok still has over 190 million people in Indonesia on the platform, 120 million in Brazil, and another 100 million scattered across Europe.

It's worth noting, however, that the bulk of spending power by TikTok users is concentrated in the US, where TikTok brought in over $6 billion in revenue for 2023 by serving ads to American consumers. It doesn't matter that ByteDance is worth $300 billion - $6 billion disappearing overnight will sting.

🇯🇵 Japan News

Toyota has announced that Phase 1 of construction is complete for a new tech-driven city called "Woven" at the base of Mount Fuji. While founders and CEOs constructing new cities isn't an entirely new concept, Toyota's project represents Japan's first venture into this space. In the USA, entrepreneurs like Marc Andreessen (A16Z) and Patrick Collison (Stripe) are involved in the construction of a new city just north of San Francisco, while Marc Lore (Walmart) is creating Telosa in rural Nevada (Telosa is expected to have a sci-fi look, reminiscent of something you'd find in the year 4500).

Toyota’s Woven is a $10 billion smart city that will serve as a testbed for autonomous cars, drone delivery, and smart homes. It's worth noting that the city of Woven is at a significantly smaller scale than the American projects I mentioned - this first version of Woven V1 is designed to accommodate just 2,000 residents.

FujiSoft finds itself in the middle of a high-stakes takeover battle between American firms KKR & Bain Capital. Both firms are attempting to acquire the company at roughly a $4 billion USD price point. If you're not familiar with FujiSoft, it's a technology consulting firm (similar to Accenture in the US) whose job is to find innovative technology in the world and implement it into the workflows & products of their clients (large Japanese enterprises). It's important to highlight that the proposed acquisition of FujiSoft comes at a time when technology consulting firms are booming - for example, Accenture just reported $3 billion in new AI consulting revenue for the year of 2024.

To explain the knife-fight between the two firms: initially, KKR made the first offer of $4 billion USD to FujiSoft, but Bain quickly came in with a competing offer of $4.30 billion USD. In response, KKR made a counter offer of $4.31 billion USD. Now the most recent offer from Bain Capital is a "counter-counter" offer of $4.36 billion USD. However, the latest offer from Bain Capital has become highly controversial as Bain is attempting to buy out the majority of shareholders without board consent (there's nothing inherently wrong with this approach, but it's often called a hostile takeover).

🇰🇷 Korea News

Samsung is launching what they call an "AI Subscription Club". Users can pay a monthly fee to rent a number of AI devices from the company. The concept is similar to leasing a car, like a Model 3, but instead of receiving a Tesla, you get access to Galaxy phones, Galaxy laptops, plus dozens of other product options with AI at the core. There are even rumors that Samsung will include access to AI robot dogs (being built by their subsidiary Rainbow Robotics).

It's worth noting that we don't yet have a price on the AI Subscription Club, even though the planned launch is in February of 2025. There are only a handful of companies capable of offering such an AI subscription service: Maybe just Apple, Google, and Samsung that have enough of a product mix of AI software and hardware for mass appeal.

🇮🇳 India News

Accel, a global venture capital firm that's best known for its early investments in Facebook, Spotify, and Flipkart, announced a $650 million early-stage fund for founders in India.

If we zoom out, we can see that there's been an uptick over the last five years with many tier-1 venture firms including Lightspeed, Peak XV, & Accel - continuing to back Indian founders building for their home country, at the earliest of stages. Traveling west to pursue your entrepreneurial dreams is becoming less and less the “only” path for achieving startup success.

🇨🇳 China News

MiniMax, the China-based AI startup company backed by Alibaba—and most recently valued at over $3 billion—had the 4th most downloaded app in the USA for the first half of 2024 with their product Talkie AI, according to analytics from Sensor Tower. If you're not familiar with Talkie, it’s similar to the US’s Character AI (Character was founded by Noam Shazeer, one of the co-authors of the Transformer LLM Whitepaper) Both the Talkie & Character AI apps look strikingly similar - they're both packed full of virtual characters from prominent anime figures to Super Mario Bros. - where you can strike up a fictitious conversation instantaneously.

However, the app has now completely disappeared from the iOS App Store. It's currently unknown what the underlying reason was for the app being taken down. But if we put our tin-foil conspiracy hat on for a quick second, this does come during a time in which Chinese social media apps in US markets are under heavy scrutiny.

China's government (specifically Beijing's Municipal Science and Technology Commission) announced plans for a nationwide elderly care service network that will come into effect in 2029 - an elderly care plan with humanoid robots at the center. As stated in the document, these are "robots that will provide companionship, monitor health for the elderly, and help them with household services."

By 2029, approximately 20% of China's population will be over the age of 65. This creates an opportunity for robotic startups to take a break from AI dogs (cough cough Unitree) and instead build critical infrastructure for their home nation.

Zixuan Huang, an engineer from Stability AI who received his bachelor's degree at the University of Science and Technology of China, has open sourced a new AI model called SPAR3D (Stable Point Aware3D).

SPAR3D is able to generate complete 3D objects from a single image in less than a second. This is incredibly helpful to game designers who spend a significant amount of time hand-crafting in-game assets. Games like Fortnite, Skyrim, and GTA 5 have tens of thousands of these assets, each taking many hours to create manually. It's worth noting that SPAR3D is not the first AI model to generate 3D objects from single images, but it offers best-in-class quality while being open source. The model weights can be easily downloaded via Hugging Face.

Alibaba's Qwen vision LLM is now being integrated directly into AR smart glasses. The system uses a tiny version of the Qwen base model that physically sits on the hardware, while a much larger & more intelligent version runs server-side, streaming in for complex tasks.

This approach mirrors Apple Intelligence, which also uses a small model on-device, while concurrently running a multi-trillion parameter model in its data centers. It's worth noting that the company creating the AR glasses is not Alibaba, but a startup called Thunderbird Innovation. The glasses come with a Qualcomm Snapdragon chip, weigh 39 grams, have a 7-hour battery life, and a 1.3-second response time for streaming in AI content.

In this week’s edition of Coconut Capitalists, we’re diving into:

• Alibaba’s #1 LLM Team Poached by ByteDance

• Sakana AI’s Self-Improving Artificial Intelligence

• Quick-Fire Startup News from Around Asia

Let’s get into it.

Alibaba’s #1 LLM Team Poached by ByteDance

The Scoop

This week, it was reported that Zhou Chang, one of the top AI Scientists behind Alibaba's Qwen Family of Language Models, which is often regarded as China's Best-In-Class Model Family, has been poached by rival ByteDance.

And ByteDance didn't just acquire Zhou – they also brought over his entire team of direct reports from Alibaba's DAMO AI division. The group joined ByteDance to help develop the next generation of Doubao, the company's family of language models that competes with offerings like OpenAI's GPTs and Anthropic's Claudes.

But why would Zhou leave Alibaba at such a pivotal time in China's AI race? An AI race that is hot hot hot with competition from billion dollar startups like Moonshot AI, Zhipu, and 01.AI. And let's not forget about the Tech Giants such as Tencent with Hunyuan and Baidu with Ernie, that have also built top 1% AI Research Teams.

Reportedly, ByteDance offered him a pay package worth $10 million USD cash compensation + a significant amount of stock to join ByteDance's Doubao team.

Now, it's important to note, pertaining to Zhou's departure from Alibaba, he allegedly told executives he was leaving to start a company. But within two months of his exit, Zhou was discovered to have not only joined ByteDance, but had also started hiring away his previous Qwen Research Staff from Alibaba. Due to this situation, Alibaba has accused Zhou of violating both his non-compete & non-solicitation agreements with the company, and is pursuing arbitration in the coming months.

Poaching Often Goes Very Wrong

If we look at recent history, there are two great examples of engineering teams leaving market leaders (from a technology perspective) to join underdogs.

1. The first example is Tesla’s vehicle engineers leaving to join Apple's electric car division (maybe some of you remember the codename "Project Titan"). It’s worth highlighting that in just one week, Apple poached 40 of Tesla's top engineers, including Tesla's Vice President of Vehicle Engineering (their top guy!). Apple's electric car project ended up being in development for over 9 years & cost the company a reported $12 Billion USD before Tim Cook lost his patience and shut it down entirely.

2. The second example is when Uber poached 8 of Waymo's top self-driving researchers to kickstart a brand new division called Uber ATG (Advanced Technologies Group). Similar to the Apple example, this also flopped. Over the span of 5 years, the division burned through a reported $1.4 Billion USD. Oh, and get this - the Waymo team lead who joined Uber later found himself in prison for "borrowing" Waymo's source code. But that's a story for another day...

It's worth noting that I'm probably being overly pessimistic & simply highlighting examples where these situations did not work out (& I'm not sure why, maybe it's the Singapore rain making me sad). But anyways, there's actually a more closely related comparison to ByteDance hiring Alibaba's Top Researchers - with OpenAI.

In 2021, a group of seven AI researchers left OpenAI in an act of rebellion over what they referred to as a "lack of concern for AI Safety" to co-found a competing AI firm called Anthropic. Fast forward to today - the company is now worth a reported $40 Billion USD and has annual revenue of over $1 Billion from customers like Jane Street, Notion, & Zoom. Even though the Anthropic example wasn't a "poach" per se, it's an incredible success story of a group of talented engineers leaving a company to start something new.

Why it Matters

It's difficult to explain just how world-class Zhou Chang & Alibaba's Qwen Team of AI Researchers have historically been. Before we dive in, it's worth noting that the Qwen team is reportedly over 200 people strong, so a small team leaving DAMO is for sure a brain-drain, but by no means is it catastrophic to Alibaba.

In terms of numbers, over 90,000 enterprises currently access the Qwen Models (like Qwen-Plus and Qwen-Turbo) through Alibaba's Cloud Platform. And that's not counting the tens of thousands of additional enterprises accessing open-source versions of the models through providers like AWS, GCP, and Azure. Alibaba loves to brag about how companies like Xiaomi, Sephora, & LVMH (yes, even the handbag company is excited about AI) are building sophisticated customer experiences on top of Qwen.

Moreover, in China, Alibaba Cloud owns a whopping 39% of the market, with Huawei Cloud & Tencent Cloud coming in at 19% and 16%, respectively. These 39% of Chinese cloud customers need a model family to build upon, and the Alibaba Cloud product is so much more sticky when offering best-in-class proprietary models that run only on this 1 cloud provider. It’s the same strategy that the team over at Microsoft Azure is taking with their exclusive OpenAI deal.

Sakana AI’s Self-Improving Artificial Intelligence

The Scoop

On December 3rd 2024, Japanese artificial intelligence startup Sakana AI released a blog post titled "Population-based Model Merging via Quality Diversity," highlighting what this world-class team of Ex-Google AI researchers has been pursuing. One of the many research directions they're tackling is developing a framework called CycleQD that can spin up large networks of AI Agents, capable of autonomous self-improvement to complete even the most complex of tasks.

The framework is still very new and has yet to be fully realized (right now it's much more focused on model merging and SVD mutations). But, in a future version of CycleQD, the system could run much more like a traditional organization; where instead of people, AI agents are completing tasks, and the CycleQD system could run as a manager of sorts, completely autonomous & with zero human intervention.

Now, let's take a quick step back and provide a recap on Sakana AI for our friends who are less familiar with the company: Sakana AI is the brain-child of co-founder Llion Jones (one of eight co-authors on the world's most important AI research paper, which introduced the Transformer Architecture; OpenAI's GPT Series and many other modern language models are built upon this).

The co-founders include:

• David Ha (CEO): Previously a Managing Director at Goldman Sachs Japan; also spent 6 years as an ML researcher at Google Brain.

• Ren Ito (COO): Former member of the Japanese Prime Minister's cabinet, ex-CEO of Japanese unicorn startup Mercari, and former COO of Stability AI.

Additionally, the company is currently worth $1.5 billion USD, backed by Nvidia, Khosla Ventures, and Japan's three largest banks. They've got a multi-million dollar supercomputing grant from the Japanese government, boast one of the top AI teams in Japan, and have raised a total of $344 million in funding - all in less than two years since incorporation.

Breaking down CycleQD

CycleQD is not an AI model per se, but instead a framework (Complex Python code) that could evolve into how large swarms of AIs will eventually collaborate and solve complex tasks. Multi-agent frameworks aren't new: there's LangChain, Auto-GPT, BabyAGI, and many more frameworks you can find with a quick GitHub search. But what's novel about this research direction is that each AI model will have the ability to recursively & rapidly self-improve.

Currently, for an AI model to get better, one of our fellow friendly humans would have to take a pre-trained model off the shelf (like Llama 3), manually collect fine-tuning data, fire up a GPU, write the Python architecture code, kick-start the training job while painfully debugging hardware failures, and then finally redeploy the improved model so the world can access it. This is work!!

Here's a speculative example of what this research direction, if successful, could turn into as an end-state (I must note, this is my own hypothetical - not one provided by Sakana AI). Imagine a system that can create a Hollywood-production-level movie without a single human in the loop. It's as compelling as anything you'd find in a cinema and is completely generated by a self-improving artificial intelligence. Sakana AI is far away from this outcome, but this highlights how the research could eventually be turned into a product.

As a very forward looking example (just for fun), let's say Sakana AI is partnering with Netflix to develop “AI Generated Original Movies”.

The way the CycleQD system would work is:

1. First, the CycleQD framework would be given an initial group of 100 identical Llama 3 LLMs, each with 8 billion parameters as a starting point.

2. Second, the system is then told to accomplish a very specific task: "create a movie that earns a 95% thumbs-up approval rate from 100 randomly selected Netflix users". Luckily, today we have all the AI tools to accomplish this:

   • 4K Quality Synthetic Video Generation (Like Pika Labs or OpenAI Sora)

   • Voice Cloning Technology (Like Eleven Labs or PlayHT)

   • AI Music Composition (like Udio or Suno)

   • CycleQD could eventually be the missing piece to seamlessly integrate and coordinate this entire process—much like how Apple's Final Cut Pro or Adobe Premiere Pro brings together various elements to assemble a finished movie.

3. Third, the system randomly selects one Llama 3 model as the leader. The “lead AI” now breaks down the movie creation process into 99 distinct steps for the 99 other Llama 3 models to complete. For example:

   • One model handles actor selection by searching the web in real-time via the Bing API for the top 10 actors of 2024

   • A second model takes an input from the first model to then create the synthetic AI Avatars for all 10 actors via API calls to HeyGen

   • Now, a third model generates an entire catalog of background music and soundtracks for the movie via API calls to OpenAI’s Jukebox

   • And so on...

4. Fourth, once the initial version of the movie is complete, CycleQD presents it to the 100 Netflix viewers. But, the results are disappointing—only 16 people gave it a thumbs up. The system then takes this feedback data, improves itself via fine-tuning jobs, model-merging processes, training additional models from scratch, firing underperforming models, and other techniques. And now CycleQD's second movie attempt gets 21 likes. After 21 likes, the third attempt yields 34 likes, and the cycle of self-improvement continues.

Why it Matters

Just as humans figured out how to make "rocks" (phones) communicate across the world, what often seems like hyperbole today—such as "Hollywood-level synthetic movies"—may soon become the norm. It's worth noting that Sakana AI's perspective differs from the idea of a single "god model" (which competitors such as OpenAI seem to be betting on); instead, they envision a future powered by a myriad of small, specialized models collaborating to solve the most complex tasks.

🇻🇳 Vietnam News

Nvidia has signed an agreement with the Vietnamese government to set up an AI R&D center and has committed to a reported investment of $400 million USD in Vietnam over the next 5 years. This announcement came immediately on the heels of another announcement that they would acquire VinBrain, an AI Healthcare startup owned by the Vietnamese conglomerate Vingroup.

If you're not familiar with Vingroup, it has a similar structure to players you'd find in South Korea like Samsung and Hyundai - very successful, family-controlled conglomerates with significant government connections. The difference here is that Vingroup is by far Vietnam's largest private conglomerate, with significant influence across various sectors. Its subsidiaries include Vinhomes (real estate), VinFast (automobiles), Vincom Retail (shopping malls), Vinmec (healthcare), and Vinschool (education). Acquiring VinBrain could be seen as a method for Nvidia to develop closer ties to Vingroup (and I don't say this in a negative light - VinBrain has a cracked AI team that is world-class).

🇰🇷 Korea News

Rebellions & Sapeon, two of South Korea's most well-funded ($225 million & $45 million, respectively) fabless semiconductor design startups, have officially merged. Rebellions was founded in 2020 by a top 1% technical team, led by Sunghyun Park, an Electrical Engineering & Computer Science PhD from the Massachusetts Institute of Technology (MIT, #1 STEM University in the World), who was previously the ASIC design lead for SpaceX's Starlink program. Sapeon was created in 2022 as a spin-out from SK Group.

Both startups have historically built chips solely for AI Inference workloads and will now operate under one roof with the name "Rebellions". The company plans to collaborate closely with SK Telecom & SK Hynix to expand into the global AI data center market.

Ably, South Korea's fastest-growing e-commerce fashion app, has raised $71 million from Alibaba and now has a $2.1 billion valuation - making it Korea's first new unicorn of 2024. It's worth noting, the company has 9 million monthly active users and is rapidly accelerating with a user growth rate exceeding 30% per year.

🇸🇬 Singapore News

ByteDance now has the world's second most used AI chatbot, behind OpenAI's ChatGPT. ByteDance's Doubao has a reported 62 million weekly active users. It's worth noting, this pales in comparison to OpenAI's 300 million weekly active users, but it highlights a continuation of strong momentum from ByteDance's product portfolio, which includes Douyin, CapCut, Lark, Toutiao, and TikTok.

Speaking of TikTok, the success of Doubao is even more critical given the significant geopolitical headwinds ByteDance faces from the US. TikTok will soon face a "ban" in the US if the company does not find an American buyer for the app. TikTok is currently used by over 170 million Americans and on the open market could be worth $100 billion. Oracle is best positioned to purchase TikTok - they have a market cap of $525 billion, already run all of TikTok's US cloud infrastructure, and would not face the same regulatory scrutiny as Microsoft, Meta, or Google. If Oracle is not interested, look for Walmart to make a move - they have a market cap of $750 billion.

🇨🇳 China News

Moonshot AI's founder Yang Zhilin (a PhD graduate from Carnegie Mellon University, whom many refer to as the next "Colin Huang") recently responded to two public claims: first, that he created the initial intellectual property for Moonshot AI (now valued at $3.3 billion USD) while working at his previous company, Recurrent AI; and second, that he did not receive proper consent from Recurrent shareholders to start his new venture. In his response, Yang countered these claims, stating that when he left Recurrent AI, he had received signed consent from all board members. It's worth noting that the facts of this situation are difficult to verify, as the claims and counterclaims have turned into a bit of a "he said, she said" situation.

OnlyFans, a US startup valued at over $18 billion dollars and known as a creator subscription platform for adult content, is now fully accessible in China.

The company, which paid out $6.6 billion USD to creators in 2023, has been completely unblocked from the Chinese Internet Firewall. It's worth noting that this $6.6 billion number is more than the NBA paid all of its players, which totaled $4.9 billion. Arguably, OnlyFans is larger than basketball.

This instance highlights the lack of transparency around which businesses are allowed in China. Facebook, YouTube, and Netflix are banned, while American company OnlyFans, known for explicit content, gets the a-okay from party officials.

Tencent has released an open-source synthetic video generation model that competes directly with closed-source rivals such as Pika Labs & OpenAI's Sora. The model, known as HunyuanVideo, packs a whopping 13 billion parameters and is completely free for anyone to use. It's now officially the world's largest open-source video generation model, with Tencent showing off some seriously impressive demos featuring everything from hot air balloons floating through photorealistic skies to surfers navigating complex waves.

🇮🇳 India News

Enterpret, a customer feedback data platform (similar to Segment or Qualtrics but with AI at its core), has raised $20.8 million USD from firms like Peak XV & Kleiner Perkins. The company, which is located in Bengaluru, India with satellite offices in New York City & San Francisco, already has enterprise contracts with customers such as Canva, Vimeo, and Nextdoor.

In this week’s edition of Coconut Capitalists, we’re diving into:

• The Dawn of Japan’s AI Television Reporters

• Alibaba’s Open-Source version of OpenAI’s #1 LLM

• Quick-Fire Startup News from Around Asia

• Startup of the Week with Shein, Singapore’s Top E-Commerce Powerhouse

Let’s get into it.

The Dawn of Japan’s AI Television Reporters

The Scoop

This week, the South Korean startup DeepBrain AI struck a multi-million dollar deal to bring AI Television Reporters to one of Japan's largest weather broadcasting companies - a company that reaches over 1.46 million viewers per week.

The deal, which kicks off in January of 2025, highlights an emergent use-case for AI, which some refer to as “B2B Deepfakes”, “Digital Twins”, and most often “AI Avatars”. To put it simply, platforms like DeepBrain AI, specialize in creating videos of photorealistic, but fake humans, that can talk & move, just like normal people.

Moreover, if we transition into a financial analysis of DeepBrain AI and examine their funding history, we quickly notice that they’ve raised over $44 million in venture capital (coming primarily from Korean & Japanese investors). Yet, they’re far from the only venture-backed startup in this category: In the UK, a strikingly similar startup known as Synthesia has raised over $150 million; In the US, HeyGen (which is backed by Benchmark Capital, Conviction, Thrive Capital, & Elad Gil) has raised $74 million; and in Southeast Asia, Gan AI (backed by Peak XV) has raised $6 million.

But in a sea of AI companies, many of which have higher burn rates than actual revenue, why are companies in the “B2B Deepfakes” category breaking out from the pack? Well, if we take a step back: large enterprises like Hyundai, T-Mobile, & Samsung (DeepBrain AI’s customers), buy from startups for one of two reasons. The first is to increase revenue (think products like Salesforce, Shopify, or Google Adwords that drive new customers), or the second is to save on costs (think AWS, UiPath, or Snowflake that replace expensive infrastructure). It’s really that simple.

So with the above paragraph in mind, if we now jump back into the weather broadcasting example, and further conduct some back-of-the-envelope math, we see why companies took the bait: using AI Avatars provides significant savings on cost. For instance, in Japan, the average salary for a TV weatherman/weatherwoman is $100,000 USD. Meanwhile, producing a 10-minute video using DeepBrain’s software only costs $24. And if we can go a step further with the numbers: let’s say a TV anchor does 5 shows per week, at 10 minutes per show; that's approximately 260 shows per year. 260 shows times $24 USD per show (cost of DeepBrain AI) amounts to $6,240 USD. This comes while the alternative, hiring a real person, costs $100,000 USD, per year. Based on our rough calculations, adopting DeepBrain AI saves this weather broadcasting company 94% in annual expenses.

And in the eyes of the company, this AI broadcaster comes with many perks: they never get sick, never ask for a raise, and there’s the presumption that many Japanese viewers might still believe they're watching a real person...

The Technology

If we get very slightly nerdy for a second, we can breakdown just how these "B2B deepfake" videos are made. Essentially, it's a mix of three different types of AI models/systems of models all working in tandem (language, text-to-speech, & image-to-avatar). Let’s run through the broadcasting example:

1. First, you take an off-the-shelf LLM (like OpenAI GPT-4 or Llama 3) and prompt the model to write a broadcasting script based on today's weather data. This data can be autonomously pulled from the Bing, Google, or Baidu News APIs.

2. Second, we pass this LLM script into a text-to-speech model (like Eleven Labs or Speechify) that takes this script and turns it into a pitch-perfect MP3 audio file.

3. Third, we take the audio file and a reference image of a real person (a full body photo), and make an API inference call against a large "Frankenstein" system of models to create our Virtual AI Avatar. This is essentially a mashup of "photo to animation” models, "lip movement” models (similar to Wav2Lip), and “motion generation” models that are typically based on a reference video (you can find many reference videos on Hugging Face). The final output is a "virtual twin" that looks and sounds just like the actual human.

Why it Matters

So is the weather broadcasting example just a one-off instance of AI replacing human jobs? Or is this real-life example foreshadowing the future of the entertainment industry at large? Will actors, singers, and influencers all just become AIs? Well, if our answer to this question relies solely on cost savings data, the answer is a resounding YES. Every industry wants 94% in cost savings; it’s a no-brainer.

But, if we steel-man the case against generative AI in entertainment, and propose that the broadcasting example is just a one-off, there are many compelling arguments to be made. One argument is that we can split the entertainment industry into two distinct categories: The first category is pure information consumption, like the weather, where the actual entertainment value of the person communicating isn’t incredibly relevant. In this case, the viewer just wants to obtain the information quickly & efficiently, to then move on with their day.

The second category (which many would argue is less likely to be disrupted by AI), is “parasocial” entertainment. This is where the relationship between the audience and the presenter (no! not that kind of relationship) goes beyond just the information being communicated. Here, the value lies in the emotional or personal bond that fans develop with real-life individuals. For example, if you’re a fan of Blackpink, it’s not just about the music—it’s about following their Instagram and TikTok updates, engaging with their personalities, and dreaming of seeing them live. This blend of online interactions and in-person experiences, such as attending a concert, creates a uniquely human connection that is challenging for AI to replicate.

It’s important to note that companies like DeepBrain AI, HeyGen, & Synthesia, are currently making most of their revenue from B2B use-cases, such as product demos & onboarding videos for new employees, and the many consumer-oriented use-cases are yet to be demonstrated to a broad audience.

Alibaba’s Open-Source version of OpenAI’s #1 LLM

The Scoop

A famous Alibaba internal team known as the International Digital Commerce Group, released an open-source competitor to OpenAI's leading model, o1. And you can download this Alibaba model right now on Hugging Face! It's called Marco-o1, and its logo is a strawberry emoji like "🍓". If you followed the OpenAI Twitter 'lore,' 'strawberry' was the internal codename for OpenAI's chain-of-thought-based LLM.

Okay, so what is Marco-o1? It's a fine-tuned LLM within Alibaba's family of models known as Qwen. Think of the Qwen Family similar to how Anthropic has Claudes (Sonnet, Opus, and Haiku) or how OpenAI has GPTs (3.5, 4, 4o, o1, etc.). Each model in the family has various tradeoffs regarding inference speed and performance.

It’s worth noting that Marco-o1 uses a similar chain-of-thought mechanism to OpenAI's o1 model, where it breaks down complex reasoning tasks into smaller sequential steps. Essentially, this is where the model verbalizes the answer in each step back to itself, similar to how humans utilize their inner monologue to think through complicated problems step by step.

Now, it's important to highlight that this 7B Qwen fine-tune doesn’t have the same performance capabilities (as of today) compared to OpenAI’s o1 model. This is primarily due to the fact that it’s a much smaller model with only 7 billion parameters, whereas OpenAI’s o1 model reportedly has 1.7 trillion parameters. Given this, we can assume that Alibaba releasing Marco-o1 was meant to serve as an example, of what could be, if the same fine-tuning technique was applied to a much more sophisticated & much larger Foundation Model.

The Performance

For the ML researchers who read this newsletter (and anyone else that’s “AI curious”), with regards to the actual performance evaluations for Marco-o1, the team was able to increase the accuracy rate from 85% to 92%. And these results came solely from this new in-house chain-of-thought fine-tuning approach. For more context, the performance evaluations were conducted through the MGSM LLM benchmark, which measures mathematical reasoning ability through word problems.

It’s worth noting that the 7% improvement (in isolation) might not feel like a step-change in model capabilities. But given that this was simply a fine-tuning approach, and there was no overhaul of the architecture code, 7% is actually quite significant.

Why it Matters

There’s a healthy debate going on in the AI industry regarding open-source versus closed-source language models - especially as the industry begins to consolidate around a few large players that continue to keep their models quite secretive.

Just look at the US as a case-study, there were previously dozens of large language model companies, with valuations far into the billions: including Adept, Character AI, & Inflection; with each of these companies competing fiercely against the other; now it’s down to X.AI, Anthropic, and OpenAI as the big three. Technically Adept, Character AI, & Inflection aren’t dead (and still have a few 100x researchers at the companies), but they’re operating in a zombie-like state given that their founders have left. If you need a quick refresher on the undead, I’d recommend “Train to Busan”!

Anyways, the point is: as these AI companies become increasingly powerful, it’s crucial to maintain a path for startups & researchers to compete; And by providing an open-source option for these groups, the increased competition will be positive for users in the long run. And publishing findings like Marco-o1 is critical for the industry.

🇯🇵 Japan News

The SoftBank Vision Fund announced its plan to buy $1.5 Billion worth of shares from OpenAI employees by the end of December 2024. This comes just weeks after Masayoshi Son's SoftBank Vision Fund invested $500 million into OpenAI alongside Joshua Kushner's Thrive Capital (the firm led by Donald Trump's son-in-law's brother), Microsoft, Nvidia, and Khosla Ventures.

In other OpenAI news, this investment announcement came just a day after a group of artists leaked OpenAI's secretive video model "Sora". Now when we say leaked, this requires additional context. The actual model weights and architecture source code (aka the crown jewels) were not leaked. Instead, a group of artists who were initially given API access by OpenAI to provide feedback on the product decided to release the endpoints on a public internet forum.

Jack Dorsey, the founder of social media apps Twitter/X & now Bluesky, saw his new platform achieve a 500% increase in Japanese user downloads during the month of November 2024.

It’s worth noting that, although Jack Dorsey was the original founder of Bluesky, he officially cut all ties to the app in May 2024. As a bit of history, Bluesky was developed through a small incubation within Twitter/X's San Francisco HQ before being spun out as a separate company. Moreover, in Japan, Twitter/X currently has 6.7 million DAUs, Threads has 1.5 million DAUs, and Bluesky has 500,000 DAUs.

🇨🇳 China News

ByteDance has filed a $1.1 million lawsuit against a 19-year-old intern who had previously worked as a software engineer at its Beijing HQ. Reportedly, the intern diverted computing resources from 8,000 GPUs, which were originally meant to train ByteDance's latest LLM. Why did a 19-year-old need 8,000 GPUs? Perhaps an AI to slide into his crush's TikTok DMs? Who knows - but hopefully nothing more sinister…

Jack Ma, the founder of Alibaba with a net worth of $25.6 billion, whose wife has been buying up much of the real estate within a 1-mile radius of my apartment in Singapore (thank you, Mrs. Ma...), has returned for a visit to Alibaba's Hangzhou HQ.

It's difficult to know the reason for the visit. Is Jack returning to Alibaba full-time, in a similar fashion as Sergey Brin recently returning to Google as a technical lead (I heard Sergey missed the free breakfast), or did Jack stop by simply to boost company morale—given that Jack is one of China's top entrepreneurial stars?

It's worth noting that his visit comes at a time when the Chinese government is trying to boost confidence in the country's private sector, as the world's second-largest economy faces an onslaught of upcoming challenges: including (many) venture capitalists moving their money out of China, Chinese entrepreneurs leaving to start companies in Japan & Singapore, as well as the recent geopolitical headwinds, in the form of import tariffs, that are being threatened by U.S. President-elect Donald Trump.

🇰🇷 Korea News

Seoul will be hosting two of the world's largest startup events in the upcoming two weeks. The first is "COMEUP 2024," which will be hosted on December 11th-12th and is expected to draw over 70,000 attendees. In the words of the organizers, "The event is a global startup festival to meet early-stage startups as well as the people who are nurturing these early-stage startups towards success, including investors & corporate partners." The second event is "AI SUMMIT SEOUL 2024," which is geared towards AI Startups, Researchers, Professors, and anyone who is "AI Curious". This second, event is taking place on December 10th-11th. Conveniently, both events are occurring at the Iconic Coex Complex near the Gangnam District.

Taking a look at the programming & speakers for both events, it’s quite stellar. For many people, they get a bit icked out at conferences geared towards "networking only", but both of these events stand out by offering a strong focus on technical education for attendees. The speakers include Cristóbal Valenzuela, the CEO of RunwayML ($4 Billion Valuation), Peggy Johnson (an ex-Microsoft EVP & my former boss), as well as Sunghyun Park, the CEO of Rebellions ($700 Million Valuation).

🇹🇼 Taiwan News

TSMC's founder Morris Chang released a memoir on November 29th. The biggest news from the book was that more than 10 years ago, Morris Chang asked Nvidia's Jensen Huang to be the CEO of the Taiwanese chip giant, but Huang turned him down within 10 minutes.

This conversation between Chang and Huang occurred shortly after Chang had returned as the TSMC CEO (after Rick Tsai stepped down). Tsai served as TSMC CEO from 2005 to 2009 and is now the CEO of MediaTek, which is worth $61 billion USD.

🇸🇬 Singapore News

Figma, a $20 billion Silicon Valley darling that creates design software, is now suing a Singaporean design startup, claiming they have stolen large chunks of Figma's proprietary source code.

This week, a closed-door hearing – where matters like witness details, confidentiality orders, and documents to be produced are discussed before a trial – was held within Singapore's Superior Court. In public court documents, Figma claims that Motiff's products don't merely "share similar interfaces and commands," but their products "perform identically" to previous versions of the Figma Platform.

A spokesperson for Motiff said that these accusations were "unfounded" and rejected any suggestions that it had copied Figma's source code. The spokesperson added that "Our code was developed in-house and independently of other organizations."

While it’s not publicly disclosed how much Figma is suing Motiff for in Singapore, looking across the pond to the US, court records show Figma is seeking compensation of at least $75,000. Huh, only $75,000 for a company you claim has stolen your source code? That's the price tag of a one-day corporate retreat for the Figma Engineering Team. Based on this low figure for what is such a serious allegation, it's likely we don’t have the full story, and many more documents are yet to be disclosed.

🇮🇳 India News

Andrew Ng, the world-renowned Stanford AI professor, who (fun fact) also spent much of his childhood living in Hong Kong and Singapore, has made his first venture investment in India via an AI healthcare company. This came in the form of a reported investment of $3 million through his $120 million venture fund, AI Fund II.

Jivi AI creates language models for India, with the goal of expanding into AI-powered co-pilots for doctors and consumers. And for a startup that has only been around a little over a year, their technology is incredibly impressive. For instance, their latest model scored an average of 91.65 on a healthcare leaderboard. In comparison, both OpenAI's GPT-4 and Google's Med-PaLM 2 scored in the low 80s.

Shein, Singapore’s Top E-Commerce Powerhouse

The Scoop

For many of us, especially those in the USA, we've bought more clothes on Shein than we'd ever like to announce publicly. The Shein app is like a digital baby between TikTok and Fast Fashion, resembling companies like H&M, Zara, and Forever 21, but offering even lower prices, styles that are on-trend, and instead of a brick & mortar retail experience, users make purchases via a few clicks on an iPhone app. For the year 2024, the company is on track to hit $40 Billion in revenue - yet for a company of its size, the actual business maintains a low profile. As just one example, Shein’s CEO has no social media & there’s only a single photo of him on the internet.

If we go back in history, Shein was founded in 2008 in Nanjing, China, as a small online retailer that sourced clothing from local wholesale markets in Guangzhou and resold the clothing on its online store at a 20-30% markup.

A turning point came in 2014 when Shein became large enough to acquire a smaller retailer, but with its own sophisticated manufacturing and supply chain capabilities. This acquisition transformed the company from an e-commerce storefront selling "other people's stuff" into a vertically integrated online retailer.

The most drastic shift in Shein’s strategy came when it entered the US market in 2017, recognizing that it was American consumers who most wanted to refresh their style every week with what’s currently trendy (versus other markets, which prefer to reuse the same clothes for years at a time). With their built-up supply chain and manufacturing advantage, they put their foot on the pedal, adding $300 million in US revenue, in just their first year.

Their advantage—being the supply chain—is best illustrated through a real-life example of how the company can go from concept to production, clothing in the hands of consumers, in a matter of days. So imagine the influencer Kim Kardashian attends the Oscars on a Monday night, wearing a trendy outfit that cost around $2,000. Now, just like magic, by Tuesday morning, a designer at Shein has already recreated the entire look - and with a price-tag of $40. And by Tuesday night, the outfit has been pushed out and is available for purchase on the Shein app.

Behind the scenes, Shein’s strategy is to initially launch new products in small batches of 100 to 200 units to gauge consumer interest. And once orders start pouring in, the company can quickly scale up production to meet the surging demand. This allows Shein to go from The Oscars to a Person’s Wardrobe in a matter of days.

Singapore Washing

It’s worth noting that in 2022, Shein moved its headquarters from Nanjing to Singapore, in a practice that many have dubbed "Singapore washing". This was meant to distance itself from China amid geopolitical tensions. During this time, Shein faced scrutiny over labor practices, environmental impact, and alleged copyright infringement, leading to calls for boycotts and bans in certain Western markets. By shifting its corporate base to business-friendly Singapore, an ally of both Western countries and China, Shein aimed to sidestep negative sentiment and improve its chances of accessing Western capital markets (via an IPO)…

Furthermore, in 2023, Shein confidentially filed for a US IPO, but SEC regulators intervened, citing alleged unethical labor practices in China. Shein quickly withdrew its IPO paperwork and shifted its sights to the London Stock Exchange, targeting a 2025 IPO timeline.

On another note, it’s important to highlight that in 2025, the Texas Stock Exchange (TSE), which is endorsed by Elon Musk, is planned to launch. The TSE will require less reporting to access capital markets, potentially providing a new method for global companies, including those in China and other markets, to achieve liquidity without the same level of paperwork and regulatory hurdles.

Bull Case

In 2024, Shein is estimated to have a gross revenue of $40 billion USD, with the USA alone representing over 35% of its sales. The company has gone global - with 7% of sales in Germany, 6% in France, and 5% in Japan.

One of Shein’s clear advantages is the company’s employment of a large number of computer science graduates as AI Engineers. These engineers build sophisticated AI sentiment models that scrape the internet for fashion trends with thousands of unique data points, being collected every second.

This allows Shein to develop new SKUs almost instantaneously to meet consumer tastes. While Lululemon has 120-day lead times from design to manufacturing, Nike has 60 days, Zara has 21 days, and Shein averages 4 days or less. Despite many Western shoppers publicly stating their concern for sustainability and the environment on social media, Shein's large revenue percentage in these markets suggests that, for many, this could be a form of virtue signaling.

Bear Case

Shein's U.S. market, contributing 35% of its projected $40 billion in 2024 revenue (~$14 billion), is under threat. Donald Trump's proposed round of tariffs, as high as 60%, could raise costs by billions annually. Additionally, closing the de minimis rule, which allows imports under $800 to avoid duties, could add $100 million or more in taxes annually. These changes would decimate Shein's price advantage, potentially jeopardizing its dominant position in the U.S. market.

Additionally, despite Shein's high revenue, its future profitability could suffer under a new round of tariffs. In 2023, Shein reportedly had gross margins of 35%, which is significantly lower than Zara's 57%, despite operating at a much larger scale. While the company reported $2 billion in net profit in 2023, the proposed 60% tariffs and the closure of the de minimis rule would cause much of this profit to evaporate.

It’s worth noting that although Shein's HQ is in Singapore, the majority of its supply chain is in China and would therefore face pressure from Trump administration tariffs, threatening the company's entire business model.

• China’s second-richest man publicly blasts ByteDance & Pinduoduo

• Anthropic & Snowflake are new friends, and the scoop with Reka AI

• Taking on Mark Zuckerberg is China’s AR startup built on Alibaba’s LLMs

• More Asia Startup News

🔥 China’s second-richest man publicly blasts ByteDance & Pinduoduo

The scoop: China's Zhong Shanshan, whose net worth exceeds $60 billion USD, delivered a scathing critique of two Chinese tech giants during an event near Shenzhen. He first accused Pinduoduo of price manipulation and claimed that they were destroying local mom & pop merchants. Second, he launched a particularly harsh tale of how ByteDance allegedly orchestrated a misinformation campaign against his company.

Zhong went on to demand an apology from ByteDance's founder over what he claims was a coordinated effort by online trolls to create artificial narratives such as his company’s water contains large amounts of lead (which can be deadly to humans) and also that his company had moved their HQ to the Cayman Islands to evade taxes.

The impact has been substantial: Nongfu's shares have declined sharply, erasing more than $20 billion USD in market value since January 2024. This decline cost Zhong his position as China's wealthiest individual in August.

Fun facts:

• If you're not up to speed with Zhong, his fortune derives primarily from two holdings: an 84.5% stake in Nongfu Spring, China's dominant bottled water company, and a 73% stake in Beijing Wantai Biological Pharmacy Enterprise, which produces vaccines and hepatitis test kits.

• Chinese media has long dubbed Zhong the "Lone Wolf" for his independent business approach and tendency to avoid the spotlight. This is a rare occasion for him to speak publicly, and even more so in a negative fashion.

• In addition to the stock price, the controversy has materially affected Nongfu Spring's business - the Hangzhou-based company reported an 18% decline in bottled water revenue to $1.18 billion USD in the first half of 2024.

Why it matters: Public criticism from China's business elite toward the domestic business sector is exceptionally rare. The last notable instance was Jack Ma's critique of Chinese financial regulations (which occurred just days before the $300 billion USD Ant Group anticipated IPO, which was halted in the following week).

This led to his lengthy public absence (during which much of Western media questioned if Jack Ma was even still alive) until his eventual reappearance at a Brooklyn Nets game in Paris. After this critique of regulators, the previously outspoken businessman has maintained a noticeably lower profile.

💪 Anthropic & Snowflake are new friends, and the scoop with Reka AI

The scoop: Snowflake and Anthropic have formed a multi-year partnership to develop AI applications together, integrating Anthropic's Claude 3.5 model with Snowflake's Data Warehouse product. Spokespeople for both parties stated that the collaboration will improve productivity for Snowflake's 10,000+ enterprise customers.

This partnership marks a strategic shift for Snowflake. Just four months ago, the company was reportedly close to acquiring Singapore's Reka AI for $1 billion USD, a startup founded by former Google DeepMind researchers (who are originally from Indonesia, Hong Kong, & Singapore).

The potential deal followed Snowflake's acquisition of AI search company Neeva and suggested a "buy vs build" approach to expanding its AI capabilities. While the reasons why the acquisition didn’t materialize remain unclear, Snowflake has pivoted to a partnership-driven strategy as their backup plan, as shown by the Anthropic deal.

Fun facts:

• Reka AI has built one of the industry's most comprehensive suites of multimodal AI models: Reka Core (67B parameters), Flash (21B), Edge (7B), and Spark (2B). These models can process text, code, images, video, and audio data

• The startup raised $58 million USD in 2023 from investors including DST Global Partners and Radical Ventures

• According to LinkedIn, the company has 22 employees with offices/team members based in California, Singapore, Hong Kong, and even Zurich

Why it matters: Asia-based foundation model companies like Reka AI, Sakana AI, and Sarvam AI face an increasingly competitive landscape for major partnerships, competing directly with established players like OpenAI and Anthropic.

However, so far, they've been able to compete against these heavyweights with great success by focusing on their regional strengths - Reka AI, for instance, has secured strategic partnerships with Telkom Indonesia and Alibaba Cloud. Their ability to continue winning significant deals while competing with Western AI companies will be crucial for their growth, and they're off to a great start!

🧠 Taking on Mark Zuckerberg is China’s AR startup built on Alibaba’s LLMs

The scoop: China-based Rokid recently launched its latest AI Augmented Reality smart glasses, alongside a unique integration with Alibaba's Qwen family of language models. Unlike Meta's recently demonstrated Orion glasses (which if you haven't seen yet - there are tons of YouTube videos), Orion remains years away from a consumer release. Instead, Rokid's product is market-ready & you can immediately purchase on their website. However, from a product design standpoint, there's still room for refinement in the aesthetic department. The look & style is more similar to Snapchat Spectacles or Google Glass versus the Ray-Ban Meta Smart Glasses.

The team:

• The CEO is a computer science PhD from the University of California, Berkeley

• The CTO was previously a VP of Engineering for Alibaba's Research Team

• The Chief Scientist previously led Samsung's Semiconductor Institute in China

Bull case: Rokid's first-mover advantage in integrating LLMs with AR creates a compelling market position. And their flagship Rokid Max delivers 1920x1080 resolution per eye at $439, which is reasonable in comparison to mixed reality competitors such as Apple Vision Pro at $3,499 or the Meta Quest Pro $3,499 (obviously these are quite different products, but it's the best comparison we've got in Augmented Reality land - haha)

Bear case: Rokid faces significant competition from tech giants with vastly superior resources. Meta's Reality Labs division spends $10 billion annually on R&D, while Apple has committed $15 billion yearly to AR/VR development. The latter's Vision Pro, launching in early 2024, sets new standards with 4K resolution per eye and advanced spatial computing capabilities.

Rokid's $20 million funding is insignificant against these investments. The AR market remains niche, with just 2.4 million global shipments in 2022. If market growth disappoints, hardware-dependent companies like Rokid could struggle to survive.

The democratization of AI models and AR development tools could enable well-capitalized Chinese tech companies to enter the market. ByteDance's Pico headsets already control 20% of China's XR market, leveraging TikTok's massive distribution network. This competitive pressure could force Rokid to choose between costly innovation or margin-eroding price cuts.

🇯🇵 Japan News

• The Japanese government is taking on the role of a late-stage venture capital firm. The Department of Economy, Trade, and Industry (METI) is committing $1.28 billion to semiconductor startup Rapidus, to be wired in April 2025.

🇨🇳 China News

• Baidu has achieved a remarkable manufacturing milestone with its latest autonomous vehicle (called Apollo RT6), with production costs down to only $34,000 per unit. This stands in sharp contrast to America's Waymo vehicles, which were estimated to cost $180,000 each as of January 2021.

🇸🇬 Singapore News

• FTX co-founder Gary Wang (who has recently been in hot water with both Singapore & the US) was awarded zero prison time after providing extensive cooperation in the case against Sam Bankman-Fried. Wang's testimony proved crucial in securing Bankman-Fried's conviction and 25-year sentence.

🇮🇱 Israel News

• Lightricks has open-sourced its LTX-Video Generation Model, breaking new ground in real-time video generation. Listen to this: The system can generate 5-second videos in just 4 seconds at 24 FPS and 768x512 resolution, while solving painful challenges like frame inconsistency and flickering. The technology's speed is a breakthrough for interactive media - imagine Fortnite generating adaptive cutscenes based on player actions without pre-rendering.

🇰🇷 Korea News

• Naver Pay partnered with blockchain firm Hashed to implement NFT-based ticketing at its DAN24 technology conference, successfully processing nearly 9,000 transactions for over 20,000 digital tickets. The successful deployment marks a tangible step forward in blockchain ticketing, an often-discussed but rarely implemented use case that Naver appears to be seriously pursuing.

• Bering Lab has achieved 99% accuracy in its AI-powered legal translation product by combining machine learning with a network of 500 lawyers and 800 translators across 30 countries. The company (backed by funds like SBVA & Naver's Startup Factory) has attracted over 300 global clients, including 140+ law firms, while reportedly maintaining a 90%+ solution repurchase rate.

• UAROUND's photo & video AI content creation platform called Genbler (built for face swaps, background edits, image upscaling, etc.) earned Product Hunt's Product of the Day distinction shortly after its launch. If you're into startups, you'll know that Product Hunt is a critical distribution channel for new software.

• Samsung introduced its Gauss2 LLM in three variants - Compact, Balanced, and Supreme. The AI Model supports 14 languages and processes requests 1.5-3x faster than leading open-source alternatives. Interesting news, as there were many rumors of an OpenAI & Samsung partnership. But unlike Apple, it seems that Samsung is taking the path of developing large language models in-house.

• AI Friends, Nvidia’s Jensen Huang & SoftBank’s Masa Son

• Singapore’s Peak XV, Lightspeed, & RazorPay partner up to fund startups

• Korea’s Twelve Labs is the World’s #1 AI Video Understanding Startup

• More Asia Startup News

🔥 AI Friends, Nvidia’s Jensen Huang & SoftBank’s Masa Son

The Scoop: At the NVIDIA AI Summit Japan 2024, held November 12–13 in Tokyo, NVIDIA CEO Jensen Huang and SoftBank CEO Masayoshi Son announced a collaboration to build Japan's most powerful AI supercomputer. The system will utilize NVIDIA's Blackwell platform (the latest & greatest chip designs), featuring a fleet of DGX B200 systems. With the USA having 6 of the world’s top supercomputers, it’s an operating for Japan to catch up.

Fun Facts:

• In addition to the supercomputer announcement, at the event, NVIDIA and SoftBank also successfully piloted a system called “AI-RAN,” the world's first telecom network combining AI and 5G. This breakthrough allows AI models to process real-time data at the edge of the network—such as optimizing signal quality, predicting maintenance needs, or enabling advanced services like personalized user experiences—while simultaneously delivering 5G connectivity. Projections suggest it could generate $5 in revenue for every $1 invested in AI-RAN infrastructure, driven by significant efficiency gains.

• After a slow couple of years, Japan's AI startup ecosystem is finally accelerating, with over 370 AI-driven startups participating in NVIDIA's Inception program at the event, and a significant number of developers from NVIDIA’s community of 250,000 strong participating via an online video stream of the AI Summit.

Why It Matters: The collaboration between NVIDIA and SoftBank marks a significant milestone in Japan's AI development, positioning the country as a leader in AI and telecommunications convergence. The deployment of the AI-RAN network demonstrates AI's potential to enhance 5G networks, offering new revenue streams and improved services that could be lucrative for the two companies.

🚀 Singapore’s Peak XV, Lightspeed, & RazorPay partner up to fund startups

The Scoop: Razorpay, in collaboration with Peak XV Partners and Lightspeed, has launched the 'Razorpay Venture Investment Program' to invest in early-stage B2B startups across sectors including fintech, e-commerce, retail, healthcare, and logistics. The program aims to invest up to $1 million in 10–15 pre-seed to Series A SouthEast Asian & Indian startups on an annual basis.

Fun Facts:

• The program targets early-stage B2B startups, offering not only funding but also access to Razorpay's technology stack, deep-discounts on startup credits, and integration support.

• With three equal partners (Razorpay, Peak XV, and Lightspeed), startups will access a combined expertise of $5 billion+ in funds managed globally, amplifying their growth potential.

• The program's portfolio startups have the opportunity to leverage Razorpay’s tech stack, which supports $100 billion+ in annual payment volume, offering unparalleled scalability for payment solutions.

Why It Matters: This partnership between Razorpay, Peak XV, and Lightspeed represents a strategic move to strengthen India & SouthEast Asia’s B2B startup ecosystem. It’s a rare move with 2 venture firms (Peak XV & Lightspeed) collaborating with a large technology company (RazorPay) to create a new type of fund.

The only recent comparison, I can think of, is with Anthropic (an AI startup) partnering with Menlo Ventures (a US VC Firm) to launch a $100 million fund to invest in startups building products on top of Anthropic’s API infrastructure.

🧠 Korea’s Twelve Labs is the World’s #1 AI Video Understanding Startup

The scoop: Meet Twelve Labs, a Korean startup on the fast track to unicorn status, with the industry-standard for AI Models that understands video contant. Valued at over $250 million and having raised $77 million in funding (including $50 million from Nvidia, Samsung, and Intel), they're tackling the massive challenge of making video content searchable and analyzable.

To put how large this industry is into perspective, more than 500 hours of video are uploaded to YouTube every minute, Zoom has surpassed 300 million daily video-meeting participants, and TikTok users average 58 minutes a day on the app.

Their two flagship models, Marengo and Pegasus, are gaining recognition as industry standards. Marengo creates vector embeddings for text, images, and audio, enabling sophisticated search capabilities, while Pegasus provides detailed video analysis and generates rich metadata. The company has already secured major clients like the NFL, demonstrating real-world applications of their technology.

The team:

• Jae Lee (CEO): Computer Scientist from UC Berkeley (Top 5 STEM globally), former Head of Data Science at Korea's National Defense Ministry

• Dave Chung (Head of Korea Business): Business Administration graduate from Yonsei University, successful blockchain startup founder

• Aiden Lee (CTO): Computer Scientist from Brown University, former Machine Learning Research Lead at Korea's National Defense Ministry

• Soyoung Lee (Head of US Business): Business Administration graduate from Korea University, PwC alumni

• SJ Kim (Head of Engineering): Computer Scientist from Sungkyunkwan University, former Software Engineering Lead at Korea's National Defense Ministry

Bull case: Twelve Labs is positioned perfectly to capitalize on the exponential growth of video content. Their Marengo model could revolutionize streaming services like Netflix, replacing endless scrolling with precise, natural language searches such as "show me K-dramas with love triangles and fight scenes from before 2015." The Pegasus model could dramatically improve CCTV footage analysis, enabling real-time threat detection and sophisticated search capabilities across thousands of cameras.

Their technology has already proven valuable in sports and media, as demonstrated by their NFL partnership, where they can instantly locate specific moments in vast video archives. Their victory at the ICCV VALUE Challenge, beating tech giants like Google and Alibaba, demonstrates their technical prowess, while strong backing from industry leaders like Nvidia, Samsung, and respected VCs suggests significant growth potential.

Bear case: The competition in video AI is intensifying rapidly. Tech giants like Google (with its MUM model), Microsoft, and Amazon are investing heavily in similar technologies, with several advantages: they have access to vast amounts of training data through their existing platforms, superior computing infrastructure, and can offer integrated solutions through their cloud platforms (GCP, Azure, AWS).

I should note that, at the current moment, if I surveyed founders in Silicon Valley who are building AI applications, the majority would pick Twelve Labs as the current “state-of-the-art” for video understanding models.

However, with competition heating up, they'll have to compete like Korean soccer star Son Heung-min to keep a sustainable edge ツ

🇯🇵 Japan News

• FPT, a Vietnamese tech firm, has partnered with Nvidia to launch an AI factory in Japan, acting as a hub for AI development and innovation. Think of this as a commitment between the two companies to fund the ecosystem together.

• SoftBank Vision Fund's Co-CEO Rajeev Misra steps down. Despite this change, the fund reports a strong recovery with $3.96 billion in gains for the fiscal second quarter ending September 30.

🇰🇷 Korea News

• South Korea intensifies the chip race by unveiling a $9.9 billion support package for its semiconductor industry, aiming to maintain global competitiveness. Shortly after, Japan additionally announced a competing $65 billion subsidy plan for its domestic chip sector, intensifying competition between the nations.

• NAVER's group-based social media platform, BAND app, surges to 6 million U.S. users, recording 20% year-over-year growth. The app is gaining traction among school and sports communities for its group communication features.

🇮🇩 Indonesia News

• GoTo & Indosat introduce a language model ecosystem called Sahabat-AI to understand and process local dialects. This could be seen as a friend or foe to other local competitors such as Reka AI or AI Singapore’s SEA-LION.

• The Indonesian government announced a new sovereign wealth fund, with a similar structure to Singapore's highly successful Temasek Holdings (which manages over $300B in state assets and invests in promising global companies).

🇨🇳 China News 

• TikTok considers introducing Quibi-style minidramas to broaden its appeal, following the success of similar content on Douyin (the Chinese version of TikTok). Meanwhile, ByteDance's valuation has reached $300 billion, reflecting strong market confidence despite U.S. ban uncertainties.

🇸🇬 Singapore News 

• FTX Co-founder Gary Wang developed a fraud detection tool for prosecutors while under investigation in Singapore and the U.S. for his involvement in the FTX collapse. This is a fascinating instance of Wang using his computer science technical abilities for good, instead of continuing down the path of darkness.

• Tencent’s Open Source AI Model is now #1, outshining Meta’s Llama 3

• Baidu Unveils AI-Powered Smart Glasses, Challenging Meta's AR Ray-Bans

• The Next Blackpink might just be a girl-gang of virtual avatars

• More Asia Startup News

⚡ Tencent’s Open Source AI Model is now #1, outshining Meta’s Llama 3

The scoop: Tencent's new Hunyuan large language model has claimed the top spot in Chinese & English language processing among open-source models, surpassing Meta's Llama 3.1 405B across many key benchmarks. With 389 billion parameters, this powerful model enables developers to build within Tencent's ecosystem, affecting millions of users across WeChat, QQ, and enterprise applications. It’s worth noting, the model isn’t universally the top performer on all benchmarks, but is #1 on many, and highly competitive on almost all.

Fun facts:

• Hunyuan-Large exceeds Llama3.1 70B's performance on multiple English and Chinese benchmarks. It matches Meta's flagship Llama 3.1-405B in language understanding, coding, mathematics, and logical reasoning.

• The model achieves its edge through innovative techniques, including training on 1.5 trillion tokens of refined synthetic data within its 7 trillion parameter training set. It employs advanced model structure enhancements to optimize memory usage, boost performance, and balance token distribution.

Why it matters: Hunyuan's emergence marks a shift in the LLM landscape, where regional expertise and platform integration have become crucial. By excelling in Mandarin, Tencent gains an advantage over Meta's Llama in this vital market, establishing itself as China's AI leader.

👓 Baidu Unveils AI-Powered Smart Glasses, Challenging Meta's AR Ray-Bans

The scoop: Baidu is launching AI-enabled smart glasses to compete directly with Meta's Ray-Ban Stories. Powered by Baidu's ERNIE large language model, these glasses feature real-time translation and live navigation assistance, targeting China's wearable tech market.

Fun facts:

• The glasses deliver up to 5 hours of active use, exceeding current market standards.

• A high-resolution micro-LED display provides a 23-degree field of view, matching leading competitors while preserving clear vision.

• Priced at 1,500 to 2,000 RMB ($210-280 USD), they compete directly with Meta's Ray-Ban Stories.

• The integrated ERNIE model processes multiple Mandarin dialects and regional accents, addressing China's linguistic diversity.

Why it matters: Baidu's smart eyewear launch marks its strategic entry into consumer AI products. With longer battery life, competitive pricing, and language capabilities tailored for Chinese users, Baidu emerges as a formidable player in Asia's wearable market. This move could transform how Chinese consumers experience augmented reality while strengthening Baidu's position in the expanding consumer tech landscape.

🎤 The Next Blackpink might just be a girl-gang of virtual avatars

The scoop: Metaverse Entertainment is positioning itself as the next-generation version of YG Entertainment, best known for Blackpink, but with a virtual twist. Their digital K-pop group Mave and virtual influencer Rina represent a new frontier in entertainment: AI-powered, hyper-realistic avatars designed for the digital stage.

Since Mave's debut in January 2023, one of their songs, for example, has amassed over 10 million YouTube views, while Rina's online presence has scaled rapidly to 85,000 Instagram followers and k-pop group Mave to 350,000 Instagram followers. Financially, it’s worth noting, that the company raised a Series A in 2022 of $11.4 million - with the money coming from multiple Korean Technology Conglomerates.

The team:

• Their CEO is a seasoned digital entertainment leader, formerly with NCSOFT and Kakao Games, bringing 20+ years of experience in gaming and media.

• Their CTO specializes in AI-driven character design, previously at Naver, with a PhD from KAIST, focusing on advanced virtual modeling.

• Their Chief Creative Director, formerly with SM Entertainment, is a K-pop veteran with 15 years in the industry, expertly blending music production with fan engagement.

Bull case: Mave and Rina, as virtual avatars, are uniquely positioned to capture a global fanbase similar to Blackpink but without the physical constraints of traditional idol groups. Virtual idols can deliver endless performances, meet fans across time zones, and interact with audiences on multiple platforms 24/7, potentially achieving deeper engagement.

Bear case: The novelty of virtual avatars risks waning as more entertainment and tech companies create AI-powered idols. The rapid rise of digital avatars could lead to market saturation, with fans gravitating back to live performances by traditional idols. Moreover, established K-pop agencies like YG and SM Entertainment, backed by their well-oiled marketing machines, may develop their own virtual stars to compete, potentially challenging Metaverse's early lead in the virtual idol space.

🇸🇬 Singapore News

• OpenAI's Singapore-based Developer Day has been announced for November 21, alongside a GovTech X OpenAI Hackathon. Both events are open to the public, but require individuals to apply online for access.

🇻🇳 Vietnam News

• Shein and Temu are likely to be blocked by the Vietnamese Government in January 2025 unless they are approved for additional (& very strict) regulatory requirements. This would stop the service for their 2 million users in the country.

🇮🇩 Indonesia News

• GoTo, Tencent, and Alibaba are investing over $100 million into a joint cloud project in Indonesia. The partnership aims to capture a share of Indonesia's $2.3 billion cloud market, with initial infrastructure expected to launch by Q3 2025. This collaboration highlights a regional play for the Chinese tech giants to serve Indonesia's 270 million population, while relying on a local trojan horse (GoTo).

🇯🇵 Japan News 

• 7-Eleven Japan has launched a trial of robot deliveries targeting Tokyo's aging population. With nearly 30% of Tokyo's residents aged over 65, these robots aim to make around 200 deliveries daily. If successful, 7-Eleven plans to scale to 1,000 delivery robots by 2025, aiming to reduce delivery costs by 20%.

• Japan Airlines has committed $500 million to a new electric taxi project with Soracle and Archer (think like Boom Supersonic in the USA). The venture aims to deploy 100 electric taxis by 2026, reducing transportation emissions by 40% over conventional taxis and helping Japan meet its carbon neutrality target by 2050.

🇰🇷 Korea News 

• JAY-Z has invested in a South Korean music copyright platform that offers fractional ownership, where users can buy into music copyrights starting at just $100. This startup has facilitated over 10,000 transactions in the past year, marking South Korea as a new tech investment hotspot for global celebrities.

• South Korea's "Startup Korea" visa, launched on November 7, 2024, offers a 5-year visa for foreign entrepreneurs and startups, aiming to attract 1,000 new startups by 2026. The program has already received over 200 applications, highlighting South Korea's push to become a startup hub in Asia.

🇮🇳 India News 

• Swiggy, India's $10 billion food delivery giant, has filed for a $1 billion IPO, expected in H1 2025. This IPO will be one of the largest in India's history, testing investor appetite for tech companies after the mixed performance of Paytm's $2.2 billion IPO in 2021.

• Elon Musk & Masa Son + TikTok’s & Uber’s CEOs take a trip to Saudi Arabia

• America’s Character AI, has been dethroned by China’s MiniMax

• Sarvam AI is the newest competitor to OpenAI for India’s 1.4 Billion People

• More Asia Startup News

🔥 Elon Musk & Masa Son + TikTok’s & Uber’s CEOs take a trip to Saudi Arabia

The Scoop: Few conferences can gather a more powerful group of finance and tech moguls than Saudi Arabia's sovereign wealth fund, which sponsors the annual Future Investment Initiative, also known as "Davos in the Desert."

This year's conference, held this week in Riyadh, has featured Elon Musk, SoftBank's Masayoshi Son, Ben Horowitz, TikTok CEO Shou Zi Chew, Uber co-founder Travis Kalanick, and a host of CEOs from major firms including Blackstone, Citadel, Goldman Sachs, and Apollo Global Management.

Fun Facts:

• During the event, SoftBank Group CEO Masa Son alluded to his next big investment in artificial superintelligence—AI that is much smarter than the human brain—predicting its arrival by 2035. "I had to save some money for the big fight for the big opportunity," he said. "Now I'm saving tens of billions of dollars so I can make [my] next big move."

• Also during the event, Elon Musk stated that "By 2040, there will be more humanoid robots than there are people." These robots will retail for about $25,000, he said. "Assuming we are on the good path of AI, I think we will be in a future of abundance."

Why It Matters: The PIF has historically been one of the largest Limited Partners of funds like A16Z, Vision Fund, Craft Ventures, and more. But during the event, Al-Rumayyan, the leader of PIF, announced that the firm would decrease the portion of its international investments to between 18% and 20%, down from 30%.

It’s worth noting, PIF now manages $925 billion, up from $764 billion at the end of 2023. Still, Al-Rumayyan remarked, "There is a big paradigm shift for how PIF is deploying investments" as the firm aims to refocus on its domestic economy.

In summary, we're likely to see PIF continue to invest in the U.S. venture capital funds it's already backed, but the window for VC funds to get major PIF backing for the first time may have closed.

💸 America’s Character AI, has been dethroned by China’s MiniMax

The Scoop: Chinese artificial intelligence unicorn MiniMax, which is funded by Alibaba & Tencent with a war chest of $850 million, has become the mainland's latest social-media star overseas, thanks to its popular Talkie chatbot entertainment app, which lets users have conversations with virtual characters.

Data from market research firm Sensor Tower showed that Talkie – part of the fast-growing "companion AI" market segment – was the fourth most-downloaded AI app in the United States in the first half of 2024, ahead of Google-backed rival Character.ai, which ranked 10th. Character.ai was founded by arguably the world's #1 AI researcher, Noam Shazeer, who now co-leads Google’s DeepMind division.

Fun Facts:

• Globally, Talkie recorded 17 million downloads in just the first eight months of the year, and is growing 220% month over month.

• MiniMax is expected to take in around US$70 million in sales this year, according to a recent report by The Financial Times citing anonymous sources. The bulk of that revenue is expected to come from the start-up's avatar chatbot platform, which targets the US and other overseas markets.

• Companion AI apps are particularly popular among younger users, with those aged 18-35 accounting for more than 70 percent of the user base for top apps like Character.ai, Talkie, Linky AI, and HiWaifu, according to Sensor Tower.

Why It Matters: The first wave of AI apps in the USA were ChatGPT and Character.ai, but China seems to have caught up. Between ByteDance with TikTok (similar to Instagram's Reels), MiniMax with Talkie (similar to Character.ai), and Kuaishou with Kling (similar to OpenAI's Sora), the international competition is on full blast.

🇮🇳 Sarvam AI is the newest competitor to OpenAI for India’s 1.4 Billion People

The scoop: Sarvam AI is an India-based AI competitor to the likes of OpenAI, Sakana AI, and Anthropic. Recently, the company secured $41 million from firms such as Khosla Ventures, Lightspeed, & Peak XV to become India’s Foundation Model leader. Sarvam's platform supports speech-to-text, text-to-speech, and translation across all 22 major Indian languages. The company is positioning its upcoming products as voice first, as it believes this is the user experience paradigm that will ultimately win.

The team:

• The Co-founder and CEO is Vivek Raghavan, who holds a PhD from Carnegie Mellon University, and has been in C Suite Executive Roles for over 12 years.

• The Co-founder and CTO is Pratyush Kumar, who holds a degree from IIT Bombay, and was previously a top 1% researcher at Microsoft India.

Bull case: Sarvam aims to build at levels 1 through 3 of the AI Stack (Chips, Foundation Models, & Infrastructure). The best comparison to Sarvam’s strategy is looking at what Google has undertaken in building Gemini. First, Google has built custom TPU chips or more specifically ASICs for model training and inference. Second, Google has a massive team of engineers & researchers building foundation models, and lastly, third Google has a hosted Models as a Service (MaaS) offering that allows developers to interact with Gemini through simple, production-ready AI APIs, built on GCP infrastructure.

Sarvam AI's market positioning is also compelling where 90% of India’s new internet users prefer voice over text, and only 10% speak English fluently. Their voice-first, multilingual AI platform targeting 92% of India's population through the top 22 native languages addresses a clear market gap.

Bear case: Google and Meta already offer services in 15+ Indian languages, backed by significantly larger compute power ($11.3B for OpenAI vs Sarvam's $41M). Only 17% of Indians have home broadband (even though cellular service coverage is rapidly growing), limiting complex AI deployment, and Indian users historically spend 30% less on digital services versus the global average.

🇯🇵 Japan News

• SoftBank is launching a new startup campus in Nagoya, Japan's industrial hub, to energize the nation's startup ecosystem. The open campus (it’s a similar concept to a WeWork, but at a much larger scale) is the size of 4 football fields.

• Sony has acquired sports motion capture firm KinaTrax for ~$100M. Their technology, already deployed across 75 MLB and NCAA teams, uses video motion technology to improve a hitter’s swing & a pitcher’s windup.

🇹🇼 Taiwan News

• OpenAI is partnering with Broadcom and TSMC to develop its first in-house AI chip. Production begins 2026 at TSMC, while AMD's MI300 chips will bridge the gap for immediate computing needs. Reportedly, the chip is designed for inference only, and OpenAI will continue using Nvidia for AI training workloads.

🇨🇳 China News

• Luckin Coffee, China's largest coffee chain, will expand to the U.S. in 2025, targeting NYC first with $2 drinks to challenge Starbucks. Their initial focus: areas with high Chinese student and tourist presence. This follows their comeback from a 2020 scandal that resulted in Nasdaq delisting and a $180M fine.

🇻🇳 Vietnam News 

• Vingroup, Vietnam's largest private conglomerate, launched a $150M venture fund for tech startup investments, expanding beyond their usual businesses of real-estate, manufacturing, and electric vehicles.

🇦🇪 UAE News 

• MGX, the UAE-based AI investment firm, is exploring investment in Elon Musk's xAI (Grok chatbot developer). xAI is seeking $5B at a $45B valuation.

🇰🇷 Korea News 

• Coupang's Co-founder Chris Koh raised $100M for HRZ Han River, his VC firm focused on Korean tech investments. Coupang is one of Korea’s largest companies - currently serving 18M+ active customers with a $58B market cap.

🇸🇬 Singapore News 

• Singapore’s Universities NUS and NTU, with $50 million in backing from the National Research Foundation, are launching National GRIP, a venture fund & accelerator program aimed at creating startups from local universities and research institutes. The initiative aims to produce 300 startup teams by 2028.

• Peak XV releases new cohort of 14 startups

• Korea’s IMM & LG Group form alliances with UAE investors

• Singapore’s AutoCodeRover is building AI Software Engineers

• More Asia Startup News

🔥 Peak XV releases new cohort of 14 startups

The scoop: The India and Singapore-based venture firm Peak XV (formerly Sequoia Capital SEA) has publicly released profiles of 14 early-stage technology startups from the 10th cohort of its Surge accelerator. The program provides up to $3 million in seed funding and currently runs two batches per year.

Fun facts:

• 1 Beauty and Wellness startup: Amaani (UAE).

• 2 Finance startups: Ambak (India), SalarySe (India).

• 5 Artificial Intelligence startups: Auquan (India), Brainfish (Australia), OrbitShift (India), Parseable (India), Tailcall (India), Wobot (India).

• 2 Content Creation and Media startups: Clout Kitchen (Philippines), Dubbing AI (Singapore).

• 2 Healthcare startups: Dezy (India), The Health Factory (India)

Why it matters: Peak XV is one of Asia's most important venture firms, with roughly $9 billion in assets under management (AUM) and an additional $2 billion yet to be deployed. The Surge accelerator highlights a recent development among VC firms, where many have shifted to becoming "full-stack" - essentially operating at all three layers of the VC ecosystem: accelerators, early-stage venture, and late-stage venture.

💸 Korea’s IMM & LG Group form alliances with UAE investors

The scoop: According to a Korean VC insider, Seoul-based IMM Investment Corp. has partnered with the UAE's Shorooq Partners to launch a $100 million joint fund. The alternative investment firm IMM, which manages $6.5 billion in assets, is the latest Korean player to establish strategic ties with UAE investors.

In a similar move last May, LB Investment Inc., the former venture capital division of LG Group, established a joint venture with UAE-based AIM Global Foundation. The partnership aims to create a venture fund of up to $1 billion, backed by UAE-based limited partners, with LB Investment taking the lead on fund management.

Fun facts:

• In the first half of 2023, South Korean VCs invested 20.7% of their total funding in foreign companies, marking a significant trend towards international engagement.

• About 32% of South Korea's industrial conglomerates now have corporate venture capital arms, in addition to their “fund of fund” strategies

• This year, The South Korean government has allocated $2.6 billion for the startup sector, focusing on loans, grants, and direct technology investments.

Why it matters: This showcases South Korea's focus on creating deep economic ties with other nations through venture capital partnerships. The deals reflect a growing trend of Asian VC firms partnering with Middle Eastern sovereign wealth funds to gain access to larger capital pools.

🧠 Singapore’s AutoCodeRover is building AI Software Engineers

The scoop: Let me introduce you to AutoCodeRover, an autonomous agent software system, that ranks among the top 5 AI coding agents worldwide (according to industry benchmarks). The technology was developed by a group from the National University of Singapore, the project has evolved from a widely-cited white paper into a fully-fledged startup company.

From the information I’ve gathered, the team is currently building "AI Software Engineers" for enterprise use-cases. The closest comparison would be Cognition Labs' Devin (a US-based company backed by Peter Thiel's Founders Fund).

The team:

• The CEO & Co-Founder is a Ph.D from the National University of Singapore, and has numerous publications in journals like ICSE, PLDI, TOSEM, and IEEE CLOUD.

• The Two Co-Scientists & Co-Founders are 3rd year PhD student at National University of Singapore. One of them was previously a software engineering intern at SEA Group & the other is a graduate from Tsinghua University in China.

• The CTO & Co-Founder is a 2nd year PhD student at the National University of Singapore, and a graduate of a Top 2 CS University in the Netherlands

• The Advisor & Co-Founder is an NUS Chair Professor of Computer Science

Bull case: The future of autonomous coding agents is incredibly bright - especially for companies targeting Asian enterprises, where the market isn't yet saturated like in the US, where many enterprises are overwhelmed with pitches from AI startups. Furthermore there are significant productivity gains to be captured across developers & business use-cases, particularly in Korea, Japan, Southeast Asia, and other regions in Asia, and this team could help lead the charge. With their highly technical team and lots of ambition, I'll be rooting them on!

Bear case: When the team first announced their white paper in April, this space was relatively uncrowded, with their solution and Cognition Labs' Devin as major players - AutoCodeRover being the open-source version with arguably superior performance. Now, two Y Combinator companies (Honeycomb and CodeStory) seem to have caught up according to the SWE-bench assessment, which is the industry standard for coding agent evaluations. However, it’s worth noting, that I’m much more bullish than bearish as I see this space being large enough for many outlier startups ツ

🇰🇷 Korea News

• Scatter Labs, a South Korean generative AI startup backed by $11 million from SK Networks, has surpassed 1 million users just six months after launch. Their product, Zeta, lets people create personalized AI characters and engage with highly customized interactions, similar to America's Character.ai.

• SK Group's Hynix Division, the world's second-largest memory chip maker by revenue, reported a quarterly operating profit of $5.1 billion, compared to a loss of $1.24 billion a year ago.

🇸🇬 Singapore News

• NUS, better known as “The National University of Singapore”, entered into a $50 million dollar partnership with Vietnam’s tech conglomerate FPT to develop a new AI research lab at the university

🇯🇵 Japan News

• Tesla has partnered with Japan's largest electronics store chain, Yamada Denki, to sell its Powerwall home battery system. The retailer operates more than 1,000 stores across the country.

• Shohei Ohtani is competing in the MLB World Series with the Dodgers. One of his investments, an early-stage data analytics startup called Rapsodo that focuses on baseball and golf analytics, has seen a 300% increase in internet traffic since the MLB playoffs began.

🇮🇳 India News 

• Neysa, a Mumbai-based startup building cloud infrastructure for AI, has raised $30 million from Matrix Partners India (which recently raised a $525 million India-specific fund).

🇨🇳 China News 

• Deep Blue Aerospace, a startup located near Shanghai, has announced plans to begin taking passengers to space in 2027 for $200,000 a ticket - an experience similar to Virgin Galactic's spaceline or SpaceX's rideshare program.

• Researchers from the Beijing Academy of Artificial Intelligence have released OmniGen, a new image generation model that has catapulted near the top of HuggingFace's model leaderboards. This is reportedly the first AI model capable of handling multiple image generation tasks (text-to-image, image editing, style transfer, etc.) in a single, simplified system, rather than requiring separate specialized models or add-ons for each task.

• Japan’s SoftBank Vision Fund eyes Perplexity at a $9 Billion Valuation

• China's Xiaomi, Alibaba, and Tencent invest $300 million in Baichuan AI

• Digital Twin Startup from the Philippines gets $4M from A16Z and Peak XV

• 10 New Startup Jobs & 6 New Startup Internships

• More Asia Startup News

💰 Japan’s SoftBank Vision Fund eyes Perplexity at a $9 Billion Valuation

The scoop: According to the Silicon Valley Publication, “The Information”: SoftBank’s Vision Fund is reportedly considering investing in Perplexity at a $9 billion valuation. Perplexity has been on a hot streak as of this year - already raising two other funding rounds in 2024, the first coming at a valuation of $500 million from IVP, and then 5 months later at a valuation of $3 billion from Bessemer Venture Partners.

Fun facts:

• Early investors include Jeff Bezos (Ex-Amazon CEO), Yann LeCun (Head of Meta AI), Nat Friedman (Ex-Github CEO), & Tobi Lutke (Shopify CEO).

• The company uses retrieval-augmented search (RAG) on models from OpenAI, Anthropic, Meta, and Mistral, instead of building foundation models from scratch.

• & their number of monthly active users (“MAUs” for my data nerds) is 10 million.

Why it matters: The SoftBank Vision Fund, just a couple of weeks ago, invested $500 million into OpenAI as part of a $150 billion valuation funding round. This investment is particularly noteworthy because OpenAI has a product called SearchGPT in closed beta, which is a direct competitor to Perplexity.

It’s also important to highlight that the Vision Fund has previously invested a small amount in Perplexity, reportedly putting in $10 million during the last funding round. However, this $10 million investment is considered a very small check relative to the fund’s $65 billion in investable capital that has been raised to date.

🔥 China's Xiaomi, Alibaba, and Tencent invest $300 million in Baichuan AI

The scoop: The Chinese AI startup Baichuan raised $300 million in funding, backed by tech giants like Alibaba, Tencent, and Xiaomi. The company is best known as a full-stack AI startup building foundational models for text, image, and video generation.

Fun facts:

• Their in-house LLM referred to as “Baichuan 4” currently ranks #1 on SuperCLUE, an LLM benchmark test that specializes in evaluating Chinese models

• The team has also built a full-stack Conversational Healthcare AI app, using a fine-tuned version of Baichuan 4 that is currently in a large, private beta

• The company is additionally developing an AI "super app" akin to Tencent's WeChat, integrating various functionalities such as messaging, payments, and e-commerce into its platform.

Why it matters: The Chinese AI ecosystem is flourishing - & Baichuan is just one piece of China's "6-Headed Monster," which includes AI startups valued at over $1 billion such as MiniMax, Moonshot, 01.AI, DeepSeek, & Zhipu. These startups are both backed by the tech giants and compete directly with them, as demonstrated by Alibaba with their Qwen family of models, Tencent with Hunyuan, & Baidu with Ernie.

The bigger story in China’s AI ecosystem mirrors a phenomenon occurring in the USA, where many of the most talented engineers at these frontier AI Labs (usually OpenAI) are now leaving to start their own startups. This causes a significant "brain drain" on the organization, as the incentives to "do their own thing" are so lucrative that it's nearly impossible to say no to the candyland awaiting on the other side.

It's the same in China - many of the best and brightest from ByteDance & Alibaba's AI labs have left to create new startups, and then eventually many of the early employees of those startups leave to start their own ventures. As Simba said in The Lion King, "It's the circle of life." But maybe, it’s best explained in non-Disney terms: A startup's advantage isn't just money - it's assembling a team filled with the top 1% of engineers. However, these world-class engineers are being spread thinner than peanut butter across dozens of these AI startups.

🧠 Digital Twin Startup from the Philippines gets $4M from A16Z and Peak XV

The scoop: Clout Kitchen is Asia's cool kid on the block. Coming in with fresh funding from A16Z Games & Peak XV's Surge Accelerator. The startup creates interactive gaming buddies. For example, if you're playing Fortnite, you could potentially have a digital twin of Elon Musk or Mr. Beast watching your screen & playing along with you in real-time - giving you tips, strategic guidance, and being a virtual BFF.

The team:

• The CEO, Justin Gorriceta-Banusing previously founded AcadArena, the #1 campus gaming platform in Southeast Asia and Latin America. He also created the CONQuest Festival, which has become the largest creator event in Southeast Asia, attracting over 80,000 attendees per year.

• The CBO, Marcel Feldkamp, is a former professional League of Legends player who competed in the 2013 World Championship. He also founded a talent agency for gamers that later merged with United Talent Agency (worth over $100 million).

Bull case: Clout Kitchen's initial product, Backseat AI, is an in-game companion for League of Legends players. To help you visualize the experience, imagine a small circular area in the bottom right of your computer screen containing a talking, virtual avatar. And the avatar is present throughout your entire gameplay session. The product has only been out for a couple of months now and has already gained over 45,000 sign-ups. While it shares some similarities with Character AI or Replika-type products, which are mainly chat experiences, Backseat AI's video-first experience could be something special.

Bear case: Gaining usage and attention from gamers is crucial for Clout Kitchen now, but monetization will be a major hurdle in the future. While virtual goods spending hit $171.82 billion in 2023, it's unclear how much would go to AI gaming companions. The largest markets for gaming revenues include the United States ($46.7 billion), China ($44.6 billion), and Japan ($18.4 billion).

There could be opportunities for revenue-sharing plays working in tandem with the game developers, but it will be challenging for a startup. Ultimately, Clout Kitchen faces significant obstacles in turning its user base into a profitable business model.

Pony AI, one of China’s top autonomous driving startups (similar to Cruise & Waymo), has filed for a U.S. IPO to raise $300 million. The company currently operates 250 robotaxis and 190 robotrucks, primary based out of Guangzhou, China.

Robot Era, a Chinese humanoid robot startup, raised $42 million in a pre-Series A funding round to bring these mechanical AI companions to market, aimed at sectors like healthcare, education, and elderly care (very similar to Figure or Tesla’s Optimus).

Taiwan's National University created the world’s smallest quantum computer using a single photon, running at room temperature. Why should you care? Well, this will help scientists create new drugs by simulating the most complex of molecules, which in turn, will speed up treatments for diseases like Breast Cancer & Alzheimer's.

Upgrad secured $60 million from Temasek, at a $2.25 billion valuation. Think of the company similar to Coursera but focused on the Indian market, and offering skill development programs for over 2 million of India’s working professionals every month.

ByteDance has made significant improvements to its web scraping tool "Bytespider". And it’s now capable of scraping internet data 25 times faster than OpenAI’s GPTbot and 3,000 times faster than Anthropic’s ClaudeBot. Some call it the secret sauce behind ByteDance’s AI efforts, but the tool raises ethical concerns as it ignores the vast majority of website scraping rules & avoids internet rules set by the US & EU.

Singtel is joining the GPU as a service party with the launch of an AI cloud service. It competes in the same category as Microsoft’s Azure, Amazon’s AWS, & Google’s GCP (but the offering is a little more related to American startups like BaseTen, Modal, and Together which offer on-demand AI compute for training & inference).

Full-Time Opportunities

🇸🇬 • 🌐 OpenAI, Singapore - Security Engineer

🇸🇬 • 📚 Locofy, Singapore - Product Manager

🇸🇬 • 📊 Shopee, Singapore - Product Manager

🇸🇬 • 🚀 Gemini Crypto, Singapore - Software Engineer

🇸🇬 • 🤖 Palantir, Singapore - Software Engineer

🇸🇬 • 🔬 Snapchat, Singapore - Software Engineer

🇯🇵 • 🧠 Notion, Japan - Customer Success

🇯🇵 • 💼 WorkOS, Japan - Product Designer

🇰🇷 • 🧪 Epic Games, Korea - Software Engineer

🇰🇷 • 📈 Upstage AI, Korea - Software Engineer

Internship Opportunities

🇸🇬 • 💻 Hypotenuse AI, Singapore - Product Manager Intern

🇸🇬 • 📣 Shein, Singapore - Public Relations Intern

🇯🇵 • 🚀 Mercari, Japan - Data Analyst Intern

🇯🇵 • 📊 AWS, Japan - Engineering Architecture Intern

🇰🇷 • 🤖 Sendbird, Korea - Growth Strategy Intern

🇭🇰 • 📈 Jane Street, Hong Kong - Product Manager Intern

• Sakana AI goes fishing for Japan’s Big Banks with an expanded $200M Series A

• Microsoft & KT (Korea Telecom) announce a $1.8B Partnership

• China’s Kling AI is the world’s #1 Rated AI Video Model startup

• 10 New Startup Jobs & 6 New Startup Internships

• More Asia Startup News

🔥 Sakana AI goes fishing for Japan’s Big Banks with an expanded Series A

The scoop: Last month, Sakana AI (which was recently awarded a large supercomputing grant from the Japanese government) has raised a $100 million Series A round led by a few of the top American VC firms: Khosla Ventures, Lux Capital, NEA, as well as small checks from many Japanese technology conglomerates.

But given that Sakana AI is arguably Japan's hottest startup, they decided to up the funding amount, expanding to $200 million thanks in part to Nvidia's Investment Arm. But more importantly, Japan's three largest banking groups are also joining the mix: Mitsubishi UFJ Financial Group, Sumitomo Mitsui Banking Corporation, and Mizuho Financial Group.

Fun facts:

• Sakana AI creates foundation models including image generation, Japanese text generation, and recently a model that autonomously conducts scientific research

• These banking partnerships give Sakana AI access to massive proprietary Japanese financial datasets - & data is still the lifeblood of AI model training

• Not too long ago McKinsey & Company conducted a large-scale survey revealing that 52% of white-collar workers, in America, use ChatGPT every month. However, when compared to Japan, only 7% of white-collar workers said they use the product. Even more startling, 46% of Japanese respondents said they had never even heard of ChatGPT. So the opportunity for Sakana AI is massive

• Sakana’s team is made up of: David Ha (Sakana’s CEO), previously an executive at Goldman Sachs Japan, and formerly a 6-year ML researcher at Google Brain. Ren Ito (Sakana’s COO), previously a member of the Japanese Prime Minister's cabinet, was formerly the CEO of the Japanese unicorn startup Mercari, and then lastly he was previously the COO of Stability AI. As well as Llion Jones (Sakana’s CTO), who was one of the eight co-creators of the Transformer Architecture.

Why it matters: Sakana AI & OpenAI are going head to head in a showdown on the streets of Tokyo. As OpenAI recently poached the head of AWS Japan and moved into this new international market, Sakana quickly responded, securing a war chest of cash. While OpenAI brings global brand recognition, Sakana's partnerships with Japan-based companies (including the major banks) provides critical local infrastructure and data access.

In Japan's relationship-driven business culture, Sakana's domestic roots and conglomerate ties could prove helpful, potentially boxing out OpenAI among particular customers. With Japan's AI market projected to grow from $7.5 billion in 2022 to $23.3 billion by 2027, the stakes are high, and Sakana AI's local connections might just give it the edge in becoming the big fish in Japan's AI pond.

💸 Microsoft & KT (Korea Telecom) announce a $1.8 Billion Partnership

The scoop: The two companies aim to develop a Korean-specific AI model together based on GPT-4o's architecture, which is a multimodal generative pre-trained transformer developed by Microsoft-backed OpenAI. This $1.8 Billion Partnership will go towards data centers, infrastructure, and most importantly foundation models.

Fun facts:

• The CEO of KT recently stated that "Many companies claim they have Korean-style AI but it is all talk," during a recent press conference alongside Microsoft.

• However, Korea does have foundation model startups: Upstage AI has raised well over $100 million + SK Networks, Naver, & Kakao (three of Korea's largest technology companies have dedicated 50+ person, model development teams)

• KT is going on the offense with AI. As just one example, they recently led a $200 million funding round into an AI inference chip startup called Rebellions - alongside Temasek, and the Korea Development Bank.

Why it matters: Korea offers a highly lucrative market with consumer spending power that ranks among the highest in Asia, averaging over $17,000 per capita annually. For Microsoft and OpenAI, as growth slows in the U.S., expanding into Korea provides a strategic pathway to tap into fresh demand and secure a foothold in an influential, tech-savvy region.

🧠 China’s Kling AI is the world’s most popular AI video model

The scoop: Kling is an AI video creation platform that’s a part of a China company called Kuaishou Technology. It’s a $30 billion dollar valuation powerhouse, and their product Kling AI had over 17 million website visits last month alone. Also, recent consumer research from Bain & Co ranks Kling's model quality above Stability AI’s Video Diffusion, Pika 1.5, and Runway Gen3, based on a survey of 4,800 individuals.

Fun facts:

• Kling’s video model generates clips up to 2 minutes long at 1080p resolution

• Their secret sauce is a specialized 3D Variational Autoencoder (VAE) for realistic face and body reconstruction, paired with a 3D spatiotemporal joint attention mechanism to handle complex scenes while maintaining physical accuracy.

• Their training data allegedly came from a combination of 5 major video platforms: Kuaishou, Douyin, YouTube, Bilibil, and Sina Weibo

• The only other real Chinese contender is MiniMax with Hailuo, a $3 billion Chinese unicorn backed by Alibaba & Hongshan (Sequoia China)

Bull case: Kling AI is best positioned to capture a significant share of China’s rapidly growing video content market. And its ability to produce 1080p videos from simple prompts can be interwoven into the parent company’s app Kuaishou, with its 400 million DAUs, who collectively consume over 1.5 billion short videos daily.

Bear case: As AI video generation technology moves from trade secrets to open-source, with diminishing information moats, the barriers to entry are dropping. And this shift could allow well-funded platforms to enter the space rapidly. For example, Canva, with 135 million monthly active users, and CapCut, favored by over 1 billion TikTok users, are well-positioned to develop best-in-class AI video models tailored for their application layer products. These platforms already have significant user bases and can seamlessly integrate AI capabilities, potentially squeezing Kling out.

ByteDance has released $170 dollar AI-equipped wireless earbuds into the Chinese market. The earbuds enable users to talk to the company’s artificial intelligence chatbot, Doubao, without looking at their phones. The product looks similar to Apple’s AirPods but with a twist - now with your very own AI companion, à la the movie "Her".

OpenAI sets its sights on Singapore, announcing a new HQ and plans to bring 20 fresh faces on board by the end of 2024. This includes team members from solutions engineering, customer success, and a head of sales.

Baidu enters the robotaxi race with Apollo Go, in an effort to keep up with Tesla & Waymo’s autonomous ride-sharing services, while also competing against local rivals like Pony AI, AutoX, and DiDi Autonomous Driving in China’s rapidly growing market.

Linear, Korea's rising star in consumer AI search with 10 million users, secures an impressive $20 million of Series A funding to fuel its growth. The cash-pile coming from Samsung Ventures, Industrial Bank of Korea, and many other Seoul-based firms.

Temu hits a roadblock in Indonesia as the government shows it the door, asking Google and Apple to remove the iOS & Android application from their respective app stores in a bid to protect local businesses.

Hyundai revs up innovation in Seoul, partnering with 20 Korean financial firms to launch a startup demo day program that's drawing comparisons to Y Combinator.

Full-Time Opportunities

🇸🇬 • 📚 Reka AI, Singapore - Data Engineer

🇸🇬 • 🤖 OpenAI, Singapore - Customer Success Manager

🇸🇬 • 🌐 StashAway, Singapore - Head of Marketing

🇸🇬 • 💼 Eko, Singapore - Technical Program Manager

🇸🇬 • 🍌 Funding Societies, Singapore - Data Scientist

🇸🇬 • 🧪 Carro, Singapore - Operations Associate

🇯🇵 • 🧠 Sakana AI, Japan - Computing Platform Engineer

🇯🇵 • 📈 Tier IV, Japan - Frontend Engineer

🇰🇷 • 🚀 Upstage AI, Korea - Research Engineer

🇰🇷 • 🔬 Rebellions, Korea - Technical Recruiter

Internship Opportunities

🇸🇬 • 📈 Tencent, Singapore - Product Manager Intern

🇸🇬 • 📝 Carousell Group, Singapore - Media Strategy Intern

🇸🇬 • 🤖 Zenyum, Singapore - Operations Intern

🇸🇬 • 📊 Dyson, Singapore - Design Engineering Intern

🇯🇵 • 📣 Mercari, Japan - Product Manager Intern

🇰🇷 • 💻 Twelve Labs, Korea - Research Intern