Some of the brightest minds in tech are leaving cozy, high-paying jobs at places like Google, Meta, and OpenAI to start their own AI companies.
And investors are throwing staggering amounts of money at them, sometimes within months of launch.
This is a contrast to the ideal life that has been painted. They get resources, talent, and the best hardware money can buy.
But there’s a caveat. The pressure to ship products fast is intense, and big labs are trying to beat each other to market.
That means researchers have less time for bold, risky experiments. Less time to chase ideas that might not pay off for years.
“Inside the large foundational labs, the pressure to deliver benchmark performance and maintain rapid release cycles leaves limited room for genuinely exploratory research,” Alexander Joël-Carbonell, a partner at VC firm HV Capital, told CNBC.
So talented people leave to explore and build something new. And right now, investors are ready to back them.
Eye-Popping Numbers
David Silver, a former researcher at Google DeepMind, announced on Monday that his new company, Ineffable Intelligence, had raised $1.1 billion.
It’s a seed round, and his company is just months old. Tim Rocktäschel, another ex-DeepMind researcher, is reportedly raising up to $1 billion for his startup, Recursive Superintelligence.
Then there’s Yann LeCun. He left his role as Meta’s chief AI scientist earlier this year. His new company, AMI Labs, raised $1 billion in March.
In total, VCs have already poured $18.8 billion into AI startups founded since the start of 2025, just in 2026 alone, according to data firm Dealroom.
That’s on pace to beat last year’s $27.9 billion for companies launched since 2024.
Also read: Meta Projects $1.4T in Generative AI Revenue by 2035
Unrealized Opportunities
So what are all these well-funded startups working on? Some are going after problems the big labs have quietly set aside.
Ineffable Intelligence, for example, will focus on reinforcement learning. That’s a method where AI learns from experience rather than from huge amounts of written text scraped off the internet.
It’s a different bet on how to build smarter AI. AMI Labs wants to build AI that can learn from real-world data, continuously, over time. \
The company says current AI is great at generating content but still struggles with things like understanding cause and effect or behaving reliably in the real world.
“As AI moves beyond screens into industry, robotics, healthcare, and other physical environments, those limitations become increasingly important,” an AMI Labs spokesperson said.
Chip Company
One of the more fascinating new players is Ricursive Intelligence.
Founded by Anna Goldie and Azalia Mirhoseini, both former Anthropic and Google DeepMind employees, the company is using AI to help design computer chips.
They’re building on work they did at DeepMind on a project called AlphaChip, which aimed to automate parts of the chip design process.
They raised $335 million across two funding rounds in December and January, just months after starting the company in September.
Goldie explained why being independent actually helps them win customers. Chipmakers are protective of their most valuable secrets.
They wouldn’t feel comfortable sharing that data with Google, a massive competitor. But a neutral startup? That’s a different story.
“For chipmakers to trust us with their most valuable IP, we have to be Switzerland,” she told CNBC. “And that wouldn’t be possible if we were at Google.”
Her team also pulled together former colleagues from AlphaChip. Other hires came from Google, Anthropic, Nvidia, Apple, and xAI.
Competitive Areas
The biggest AI labs are so focused on winning the race to the top that they’re ignoring entire fields of research.
New AI architectures and agent-based systems. Interpretability, which is the study of how AI actually makes decisions. Vertical AI tools designed for specific industries.
All of it is getting deprioritized at the big labs. Elise Stern, managing director at French VC firm Eurazeo, which backed AMI Labs, put it plainly.
“When you’re in a race, you narrow focus,” she told CNBC. “That creates a vacuum.” That vacuum is exactly where the new startups are moving.
Founders who came from frontier labs bring something unique with them. They know what works at scale. They also know what their old employers decided not to pursue, and why.
“That’s where the opportunity lies,” Stern said.
Hiring Preferences
There’s another pattern worth noticing. These new startups aren’t just founded by ex-Big Tech researchers. They’re staffed by them, too.
When Goldie and Mirhoseini launched Ricursive Intelligence, they went back to their old network. They rebuilt the AlphaChip team, at least in part.
This is happening across the board. Humans&, a San Francisco startup launched in October by former Anthropic and xAI employees, raised $480 million in January.
Periodic Labs, founded by ex-OpenAI and DeepMind staff, pulled in $300 million in September. Both companies leaned heavily on their founders’ former networks to staff up quickly.
