How one VC burns through hundreds of millions of tokens a day to find the next unicorn
The investor behind a $75-million AI-focused fund explains why reading research papers and testing frontier models has become part of his job.
Pratyush Choudhury, co-founder of Activate AI, India's inaugural VC fund exclusively focused on AI, is redefining the venture capital landscape. To maintain a competitive edge, he has cultivated a global network of researchers and developers, dedicating substantial time to comprehending artificial intelligence technology as he evaluates startups.
The $75 million fund, established in December with co-founder Aakrit Vaish, played a pivotal role in catapulting Sarvam, a prominent Indian AI startup, to unicorn status with a valuation surpassing $1 billion.
Choudhury's mastery of AI is evident in his daily consumption of 300 to 500 million tokens, a practice that provides him with invaluable, firsthand insight into the technology's capabilities. He utilizes frontier models extensively for various tasks, from conceptualizing events and guest lists to researching startups and implementing research papers on GitHub.
OpenAI's Codex and Anthropic's Claude account for approximately 90% to 95% of his AI usage, with Gemini and Grok also playing crucial roles. Other tools, such as Manus and Granola, are employed across the organization, with the paid version of Cursor being another significant resource.
Choudhury emphasizes that understanding AI technology has become a prerequisite for building successful AI companies. He stresses the importance of identifying founders who possess technical acumen and are addressing durable customer problems rather than temporary capability gaps. Understanding the second-order effects of AI developments across different regions is also crucial.
The conversation remains focused on the challenges and opportunities that AI presents for the venture capital industry and the broader technological ecosystem.
Written by urgent.news from Rest of World Tech's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
