Lilian Weng told Thinking Machines Lab last week that she was stepping away as a co-founder, saying the job had taken a toll on her health and she wanted a more focused role. Days later, The Information reported that she would rejoin OpenAI to work on recursive self-improvement, the effort to build AI systems that help design the next generation of themselves. She is the fourth Thinking Machines co-founder to leave within the past year, according to Axios, which reported the pattern on Aug. 3.
The moves at Thinking Machines, the startup founded by Mira Murati, OpenAI’s former chief technology officer, show that the churn is not confined to the biggest labs. Well-funded young companies with clear missions are losing founding talent too, often to the same handful of rivals.
Google has been the most visible loser. In June, Noam Shazeer, a co-author of the 2017 paper “Attention Is All You Need” that introduced the Transformer architecture and a co-lead of the Gemini models, left Google DeepMind for OpenAI. Two days later, John Jumper, who shared the 2024 Nobel Prize in Chemistry for AlphaFold, announced he was joining Anthropic after nearly nine years at DeepMind. Alphabet shares fell as much as 7.2% intraday on the news, their steepest drop since February, Bloomberg reported.
Meta has had the opposite problem. The company spent heavily to recruit leading researchers into Alexandr Wang’s superintelligence operation, only to watch several prized hires quickly decamp, some for OpenAI, according to Axios. The pattern predates this year: Safe Superintelligence, the startup founded by former OpenAI chief scientist Ilya Sutskever, lost its chief executive to Meta last year.
The defections are driven by more than pay. Top researchers can now command compensation packages that include equity in Anthropic or OpenAI, both of which are moving toward public listings, giving early employees a shot at liquidity that incumbents like Alphabet cannot easily match. Access to compute is part of the calculus, along with each lab’s perceived position in the race to build more capable systems.
Culture matters too. A source familiar with the matter told Axios that Anthropic Chief Executive Dario Amodei has expressed concern that some new hires are joining for the money rather than the mission. The comment reflects a broader worry inside the frontier labs: that the flow of people has become so financialized that it will erode the research culture each company is trying to protect.
Some researchers say they see a narrowing window in which their expertise commands a premium. As AI systems automate more of the work of improving and developing new models, their own bargaining power could shrink, they told Axios. Moving while the market is hot is a rational calculation, and the most sought-after names can pick among labs within days.
The industry’s structure makes the churn hard to stop. The leading labs invest in one another, buy one another’s services, rely on the same cloud providers and recruit from the same small pool of researchers. Frontier AI, Axios notes, now resembles a single, tightly connected ecosystem rather than a set of isolated competitors. An engineer who leaves Google is as likely to end up working with former colleagues at OpenAI as anywhere else.
The retention problem has become a competitive variable in its own right. Demis Hassabis, Google DeepMind’s chief executive, has publicly expressed confidence in the company’s ability to hold onto talent, telling Semafor in June that the departures did not shake his view of the research organization. Investors are watching whether Google responds with structural changes, to equity, research governance or the autonomy of senior scientists, rather than pay alone.
History suggests money buys time, not loyalty. Google paid roughly $2.7 billion to bring Shazeer back from Character.AI less than two years before he left again. Recruiting raids in the summer of 2025 involved nine-figure signing packages and founder-led pitches, and the pace has not slowed since. The question for every lab is no longer whether it can hire the best people, but whether it can keep them long enough to ship.
For the labs that can, the payoff compounds. Model development is measured in months, and a team that holds its core researchers can iterate faster than a rival that is rebuilding after each departure. That is why retention has moved up the priority list of chief executives across the industry, alongside compute and data.
The churn is likely to continue as long as the pool of researchers who can build frontier models stays small. Until then, the most valuable asset in the AI industry, the people who understand how these systems work, will keep moving between the companies that want them most. The labs that figure out how to keep them will define how fast the field advances.
