The AI talent war is producing a surprising byproduct: a startup boom. Top researchers from Meta, Google, and OpenAI are leaving in droves to found their own companies — and investors are throwing billions at them.
According to CNBC, a wave of senior scientists are quitting big labs to start new ventures, with VCs betting they'll crack open new frontiers in model architecture, reinforcement learning, agents, and interpretability.
Take David Silver, the former Google DeepMind researcher behind Ineffable Intelligence. The company raised $1.1 billion in seed funding just months after founding — one of the largest seed rounds in European history. Another ex-DeepMinder, Tim Rocktäschel, is reportedly raising up to $1 billion for his startup Recursive Superintelligence.
AMI Labs, founded by Yann LeCun, announced a $1 billion raise in March. Its goal: build AI systems that learn continuously from real-world data.
Over the past year, alumni from OpenAI, Anthropic, xAI, and DeepMind have also pushed companies like Periodic Labs, Ricursive Intelligence, and Humans& into funding rounds. These startups not only raise money — they often poach talent back from their founders' former employers, creating a new magnetic field for AI brains.
Elise Stern, investment director at French VC firm Eurazeo, says the intense competition among big labs actually creates space for nimbler startups. "In a race, large companies narrow their research focus," she told CNBC. "New architectures, agents, interpretability, vertical models — those areas get deprioritized. That's exactly where startups can thrive."
The numbers back her up. According to Dealroom, AI startups founded since 2025 have already pulled in $18.8 billion in venture funding in 2026 alone, on track to beat last year's total.
What investors are buying isn't just star power — it's a bet on where the next generation of AI will come from. Many of these startups are actively sidestepping the dominant large language model playbook.
Ricursive Intelligence, for example, is building AI chip design tools and pitching itself as a "neutral partner" that customers trust more easily. Periodic Labs is working on autonomous labs. Ineffable Intelligence is doubling down on reinforcement learning. Humans& is also taking an experience-driven learning approach.
The common thread: a growing skepticism that simply scaling up LLMs will get us to the next level of AI.
Alexander Joël-Carbonell, a partner at HV Capital, argues that the pressure on big labs to optimize for benchmarks and rapid release cycles squeezes out exploratory research — pushing more scientists to go it alone.
In some ways, this startup wave is both a continuation of the AI investment frenzy and a side effect of big tech's internal competition. As the big labs become more focused on commercialization and scale, some of the most promising frontier innovations may be migrating outside their walls. Investors are betting that these "spillover innovations" will become the next generation of AI platforms.
One unusual investor in this space: a Japanese TV shopping company that placed early bets on Anthropic, xAI, and OpenAI, and recently quadrupled its VC fund to $200 million.