Hi friends 👋,
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:
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).
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.
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:
OpenAI with chatgpt.com had over 60 million unique visitors
Google with gemini.google.com had over 9.7 million unique visitors
Perplexity with perplexity.ai had over 2.1 million unique visitors
Anthropic with claude.ai had over 1.6 million unique visitors
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:
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.
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.
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 NewsSteven 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 NewsDeepSeek 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 NewsCinamon 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.
