Top AI researchers are leaving tech giants like Meta and Google at an unprecedented rate to found startups and secure massive funding rapidly, marking a new acceleration phase in Silicon Valley’s AI talent migration.
On April 28, CNBC reported that David Silver, a former Google DeepMind researcher, announced a $1.1 billion seed round for his newly established startup, Ineffable Intelligence, setting a record.
Another former DeepMind employee, Tim Rocktäschel, is reportedly seeking up to $1 billion in funding for his new company, Recursive Superintelligence. Meanwhile, AMI Labs, founded by Yann LeCun after his departure from Meta AI, completed a $1 billion funding round in March this year.
Reports indicate that investor enthusiasm has provided strong momentum for this wave of departures. According to Dealroom data, venture capital has invested $18.8 billion in AI startups founded since early 2025, on track to surpass last year’s full-year total of $27.9 billion.
Analysts point out that talent from major companies not only takes away technical expertise but also deep insights into industry blind spots—this is precisely the core logic behind investors' bets.
Remarkable funding size: Secured hundreds of millions of dollars within months of establishment
Reports indicate that the funding raised by these departing entrepreneurs has far exceeded the traditional boundaries of early-stage investment.
Recursive Intelligence was co-founded by former Anthropic and Google DeepMind researchers Anna Goldie and Azalia Mirhoseini, focusing on AI tools for chip design.
The company was established in September last year and completed two funding rounds totaling $335 million in December and January this year. Periodic Labs, founded by former employees of OpenAI and DeepMind and focused on developing autonomous laboratories, raised $300 million in September last year, just months after its founding.
Humans&, headquartered in San Francisco, was founded in October last year by former employees of Anthropic and xAI, and raised $480 million in funding this January.
Stern of Eurazeo attributes these founders' competitive advantage to their unique internal perspective:
They know what truly works at scale and are fully aware of which opportunities are being abandoned internally. The opportunities are there.
Technical divergence: Betting on the next-generation paradigm beyond LLMs
According to reports, these emerging companies are not merely copying the paths of established giants but are demonstrating clear differentiation in their technological approaches.
Joël Carbonell of HV Capital noted that an increasing number of AI researchers are questioning whether continuing to scale current large language models is sufficient to break through the next bottleneck in AI capabilities.
AMI Labs is focused on developing AI systems that can learn from continuous real-world data. A company spokesperson said,
AI has made significant progress in content generation, but it still exhibits clear shortcomings in fundamental cognition, causal reasoning, and reliable behavior in real-world environments. As AI moves from screens into physical environments such as industry, robotics, and healthcare, these limitations become increasingly critical.
Ineffable Intelligence will focus on reinforcement learning—enabling AI models to learn from experience rather than relying on human-labeled data, contrasting with the current mainstream approach of training on internet text. According to a source familiar with the matter, this is also the technological path adopted by Humans&.
Goldie from Recursive Intelligence emphasized the strategic value of independent identity:
Chip manufacturers must trust us with their most core intellectual property; we must remain neutral, which is impossible within Google.
Notably, the report indicates that these startups, after securing ample funding, are reaching back to large corporations, creating a second wave of talent mobility.
Goldie from Recursive Intelligence revealed that the company has reassembled the core AlphaChip team, "which involves recruiting some of our former colleagues." The current team members come from backgrounds at Google, Anthropic, NVIDIA, Apple, and xAI.
This model is common among many emerging companies—founders leverage their personal reputation and ample funding from investors to attract top researchers from their former employers and other AI giants, intensifying the talent competition between startups and large corporations.
Behind the mass exodus: Big tech's internal competition creates a window of opportunity for entrepreneurship
The arms race among large AI labs is inadvertently creating opportunities for smaller, more agile companies.
As one of the investors in AMI Labs, Elise Stern, Managing Director at French venture capital firm Eurazeo, said,
When you're in a race, your focus narrows dramatically. This creates a vacuum—entire research areas like new architectures, agents, interpretability, and vertical models are being deprioritized, not because they're unimportant, but because they can't win the current race.
HV Capital partner Alexander Joël-Carbonell also noted that as major AI labs face pressure to justify their sky-high valuations, business objectives are taking higher priority, significantly reducing the exploration space for top researchers.
Inside large foundational model labs, the pressure to deliver benchmark performance and maintain a rapid release schedule leaves almost no room for truly exploratory research, especially in directions outside the mainstream large language model paradigm.
Analysis suggests that this structural contradiction is leading top talent seeking cutting-edge but non-mainstream research directions to increasingly opt for leaving and going independent.