Readers may be familiar with several household names in the AI space, such as OpenAI, Anthropic, and Perplexity. However, it is estimated that tens of thousands of AI companies are battling to reach the same heights of global recognition.
Attracted by the potential of multi-billion-dollar valuations, the AI startup market is not slowing down. Several of these startups are “spin-offs” launched by former employees of the AI titans such as OpenAI, Google, and Meta. These derivatives have no problem attracting funding, but they are struggling to hold down their top talent.
Go deeper with GlobalData
Access deeper industry intelligence
Experience unmatched clarity with a single platform that combines unique data, AI, and human expertise.
Thinking Machines Lab is losing key staff to AI titans
Founded in February 2025, Thinking Machines Lab is unlikely to be well-known to many readers. Despite the $2bn in funding and $12bn valuation it obtained in mid-2025, the startup has launched just one product, Tinker, which assists with the fine-tuning of large language models (LLMs). The product launch in October 2025 deflated expectations following the remarkably high sums it received earlier in the year. Nevertheless, on the back of the product launch, reports emerged that Thinking Machines Lab is seeking a $50bn valuation.
Why was Thinking Machines Lab able to obtain $2bn in funding even before its first product launch? The answer lies in its founding team. The CEO, Mira Murati, is the former CTO of OpenAI. Murati left OpenAI amicably in September 2024, but the timing raised questions about whether her departure was related to disagreements about organisational changes at OpenAI. Murati brought with her approximately 20 former OpenAI employees, including John Schulman as chief scientist and Barret Zoph as CTO.
The harvesting of the same top-tier talent that made OpenAI the half-trillion-dollar company it is today excited investors. However, by October 2025, one co-founder, Andrew Tulloch, had already left to return to Meta. It is rumoured that Tulloch was poached by Meta after Meta’s acquisition bid of Thinking Machines Lab was rejected. In January 2026, Thinking Machines Lab lost two more key staff. Luke Metz, one of the startup’s co-founders, has left to join OpenAI, along with CTO Zoph.
Thinking Machines Lab is no exception
Thinking Machines Lab is not alone in witnessing the exodus of its prized talent. Established in France in 2023 by former Google DeepMind employees, H Company raised $220m in seed funding in 2024. Nevertheless, despite auspicious beginnings, three of the five co-founders have since departed the start-up, citing “operational differences”. It is thought that these departures occurred as a result of too much investor-side control over the company’s direction.
US Tariffs are shifting - will you react or anticipate?
Don’t let policy changes catch you off guard. Stay proactive with real-time data and expert analysis.
By GlobalDataFacing similar co-founder flux, Safe Superintelligence was launched in 2024 by Ilya Sutskever. Sutskever was the former chief scientist at OpenAI and was in favour of removing Sam Altman from the board in 2023. When Altman returned to OpenAI, Sutskever resigned due to disagreement over the company’s commercial direction. Within less than a year of being founded, Safe Superintelligence raised $1bn at a $5bn valuation to develop agentic AI systems that surpass human capabilities and emphasise safety. As of April 2025, the start-up has a $32bn valuation. Impressive considering it has not released any products. In June 2025, co-founder Daniel Gross left the company to join Meta. Similar to Tulloch, the procurement of Gross followed Meta’s unsuccessful attempt to acquire Safe Superintelligence.
From AI titan to AI startup and back again – a recurring cycling?
As evidenced by Murati and Sutskever, many of the better-funded AI startups have been created by employees who themselves defected from the leading AI players. Perplexity was founded by Aravind Srinivas, a former research intern at Google and former research scientist at OpenAI. Srinivas has made no secret of his critique of Google’s modus operandi.
Yann LeCun left his position as chief AI scientist at Meta after 12 years to pursue his own startup, Advanced Machine Intelligence Labs. LeCun has been a notorious proponent of world models, which use multimodal inputs to construct a simulation of the physical world around us. This differs from LLMs that are focused on language processing and statistical pattern matching. Reports state that LeCun could not adequately pursue research into world models within the LLM-focused confines of Meta.
The continuous cycle of AI visionaries fleeing the nest, enticing fellow futurists to join fledgling startups, and losing them within months to more stable incumbents appears to have become the norm. As evidenced by Meta’s “acqui-hiring” of key members of Thinking Machines Lab and Safe Superintelligence, top talent appears to be in short supply, and even with massive valuations, startups are struggling to offer the same compensation packages that the likes of Meta can afford. It remains to be seen how much this flux will impact investor confidence, particularly for newly launched startups and in the wake of increasing concerns over the AI bubble bursting.
