Top researchers are jumping ship from Big Tech firms like this one Meta again Google launching startups and raising large funding rounds in the process, as investors bet heavily on the commercial capabilities of early AI labs.
Among the big spending on AI, a lot these new startups are raising hundreds of millions in a few months it is established.
Yesterday, former Google DeepMind researcher David Silver announced that he will record 1.1 billion seeds for his startup Ineffable Intelligence. Tim Rocktäschel, another former DeepMind employee, is reportedly raising up to $1 billion for his Recursive Superintelligence startup. Rocktäschel did not respond to CNBC’s request for comment.
AMI Labs announced a multibillion-dollar raise in March, months after its founder, Yann LeCun, said he was leaving his role as Meta’s head of AI. It develops AI systems that can learn from continuous real-world data.
In the past year, former employees of OpenAI, DeepMind, Anthropic and xAI have also raised hundreds of millions from investors in a months-long venture, including AI Periodic Labs, Recursive Intelligence and Humans&.
Many of these companies themselves have hired heavily from former employers of inventors, and other AI giants, as investors have provided them with the funds needed to tempt top Big Tech researchers.
The race for AI dominance among the biggest AI labs has created an opening for smaller, leaner companies, Elise Stern, managing director at French VC Eurazeo, which backed AMI Labs, told CNBC.
“When you’re in a race, you focus less,” he added. “That creates a vacuum. All areas of research, such as new structures, agents, interpretation and precise models, are neglected, not because they are not important, but because they are not winning the fastest race.”
Getting beaten up
Investors are rushing to pour money into AI labs founded by leading researchers who previously worked for leading technology companies.
In 2026, VCs have invested 18.8 billion dollars in AI startups founded since the beginning of 2025, according to Dealroom. That’s on track to surpass the $27.9 billion raised last year by companies that have launched since the start of 2024.
The inventors who worked in frontier labs have a “different” understanding, says Eurazeo’s Stern.
“They know what works for the standard, and they know exactly what is left on the table internally,” he said. “That’s where the opportunity is.”
The increasingly sharp focus on commercial goals – as large AI labs look to justify astronomical valuations – reduces the freedom of top researchers, Alexander Joël-Carbonell, a partner at HV Capital, which also invested in AMI Labs, told CNBC.
“Within large basic labs, the pressure to deliver limited performance and maintain fast release cycles leaves limited room for original experimental research, especially outside the dominant LLM paradigm,” he added.
Finding vacancies
Recursive Intelligence, which raised $335 million in two rounds in December and January after being founded in September, builds AI tools to help with chip design. Founders Anna Goldie and Azalia Mirhoseini both previously worked for Anthropic, and Google DeepMind, where they contributed to the AlphaChip project, which looked like automating chip design.
Goldie told CNBC that potential customers are more likely to consider the new company a neutral partner instead of a competitor.
“For chip makers to trust us with their most valuable IP, we have to be Swiss, and that wouldn’t be possible if we were Google,” he added.
The company also turned to its old partners. “We put together the core team for AlphaChip, and that involved hiring some of our old partners,” Goldie said. Some team members previously worked at Google, Anthropic, Nvidia, an apple and xAI, he added.
Periodic Labs, founded by former OpenAI and DeepMind employees, raised $300 million in September, months after launching. It is looking to develop private labs.
Joël-Carbonell of HV Capital told CNBC that a growing number of AI researchers are questioning whether scaling the current LLM approach will be sufficient to reach the next level of AI expertise.
AMI Labs, founded by former Meta AI CEO LaCun, said it was “an opportune time because AI has made great strides in content creation, but still struggles with foundational, causal, and reliable behavior in real-world settings.”
Yann LeCun, chief AI scientist at Meta, speaks at the Viva Tech conference in Paris, June 13, 2023.
Chesnot | Getty Images News | Getty Images
“As AI moves beyond screens and into industrial, robotics, healthcare and other physical areas, those limitations become more important,” a company spokesperson told CNBC.
Ineffable Intelligence will focus on reinforcement learning, which is where AI models learn from experience as opposed to human data — compared to many leading AI models trained on Internet text — a source familiar with the company told CNBC.
It’s an approach also used by San Francisco-based Humans&, which was launched in October by former Anthropic and xAI employees and raised $480 million in January.
Google, Meta, Anthropic, and OpenAI did not respond to a request for comment.
