OpenAI Co-Founder Andrej Karpathy Joins Anthropic

Andrej Karpathy is not a typical senior hire. He was in the room when OpenAI was founded in 2015, led Tesla's self-driving and computer vision programs for years, and spent the time since leaving OpenAI in early 2024 becoming one of the most-watched public educators in AI. His YouTube courses on how large language models work have been watched by millions. When someone with that background picks a company, people pay attention. He has joined Anthropic's pre-training team, the group that runs the large-scale training jobs that give Claude its core knowledge and capabilities. That phase is also one of the most expensive parts of building a frontier AI model. Training a leading model can cost hundreds of millions of dollars per run, and the companies that do it better, faster, or more efficiently have a real edge. Karpathy's specific job is to start a team focused on using Claude to accelerate pre-training research itself. That is a meaningful bet. The idea is that AI can help researchers design better experiments, spot patterns in training data, and iterate faster than a purely human team could. If it works, the compounding effect could be significant: better research methods lead to better models, which lead to even better research tools. The broader context matters here. Anthropic is no longer the underdog. The company raised $30 billion in February 2026 at a $380 billion valuation, and by May 2026 is reportedly in early discussions for another round at a valuation near $900 billion. Its annualized revenue grew from roughly $9 billion at the end of 2025 to over $44 billion by May 2026. Enterprise clients, including large global companies, account for over 80% of that revenue. More than 1,000 business customers now spend over $1 million per year on Claude. At the same time, the competition for elite AI researchers has turned extraordinary. Top researchers at OpenAI can earn over $10 million per year. Google DeepMind has offered $20 million annual packages to retain key people, and has enforced six-to-twelve-month non-compete periods with full salary paid. Meta offered a compensation package reportedly worth $1.5 billion over six years to one researcher alone. In this environment, a single hire is also a signal to every other researcher considering where to go next. And Anthropic has been winning that competition. Engineers from OpenAI were eight times more likely to leave for Anthropic than the reverse. From DeepMind, that ratio was nearly eleven-to-one in Anthropic's favor. The Karpathy hire extends a streak that also included Jan Leike and John Schulman from OpenAI in 2024. For people who use Claude in their business, or are considering it, the story here is simple: Anthropic is pulling in research talent and financial capital at a pace that would have seemed implausible two years ago. The models its customers rely on are likely to keep improving at a fast rate. The risk, as always with AI tools, is that the pace of change makes any integration decision feel temporary. What works well in Claude today may be significantly different in twelve months. For OpenAI, this is awkward. Karpathy was a co-founder. He returned for a year in 2023 and then left again. He has now joined its most direct competitor. No single hire decides a market, but the direction of travel is hard to ignore.