Thore Graepel, one of the researchers behind AlphaGo who left Alphabet earlier this summer, is raising tens of millions of dollars for a new startup called Metis Reasoning, according to a Bloomberg report published this week. The raise, from a small initial group of backers, is the latest in a rapidly expanding wave of ventures founded by DeepMind alumni who are betting that the next leap in AI will come not from larger language models but from systems that can plan, act, and learn from their own experience.
Bloomberg reports that the structure and size of the funding could still change, and that Metis Reasoning could raise a further tranche of hundreds of millions of dollars at a higher valuation in a later round. For more context on this story, see our ongoing latest AI developments.
What Metis Reasoning is building
Details about the company remain sparse, but the outlines come from the startup's own materials as cited by Bloomberg. According to its website, Metis Reasoning is working on artificial intelligence that can respond to unfamiliar problems and choose an action, with applications in robotics, science, and engineering. Graepel's own project description describes the ambition as bringing AlphaGo-style reasoning to frontier AI, so that machines can plan and act under uncertainty.
The distinction from mainstream AI systems is fundamental. A large language model predicts the next word from human-written text. The systems Metis Reasoning and its peers are chasing learn from their own actions and weigh what to do before doing it — a paradigm rooted in reinforcement learning, the method in which AI improves through trial and error rather than by imitating human examples.
Graepel's credentials give the bet weight. He was one of the researchers behind AlphaGo, DeepMind's landmark system that mastered the game of Go through self-play and reinforcement learning, and he spent more than a decade at Google DeepMind before departing this summer to found the company.
The DeepMind diaspora is raising billions
Metis Reasoning enters a market segment that has seen extraordinary capital flows in recent months, much of it from people with the same London origin story.
David Silver, who led DeepMind's reinforcement learning team, raised $1.1 billion at a $5.1 billion valuation in April for his startup Ineffable Intelligence, according to TechCrunch, with Sequoia and Lightspeed co-leading the round and Nvidia and Google among the participants, CNBC reported. Yann LeCun, departing Meta, raised $1.03 billion at a $3.5 billion pre-money valuation in March for AMI Labs, which is building world models — systems that predict how an environment will respond to an action, letting a robot compare options before it moves, as TechCrunch and a World Economic Forum article described the concept.
Closer to home, Emulate, founded in August by three DeepMind veterans, is in talks to raise up to $700 million at a $3.7 billion valuation, PYMNTS reported last week. That would make it the third lab to spin out of DeepMind's London offices this year with hundreds of millions in financing behind it.
Why investors are looking past LLMs
The common thread across these ventures is a thesis that the scaling laws that produced today's chatbots will not, on their own, produce machines that can reliably act in the physical and scientific world. Robotics, drug discovery, engineering design, and scientific research all require systems that can handle unfamiliar situations, form plans, and learn from the consequences of their actions — capabilities that AlphaGo demonstrated in a narrow domain nearly a decade ago and that startups like Metis Reasoning now aim to generalize.
Bloomberg also notes a financing innovation accompanying the talent wave: tranched rounds, in which capital is raised in steps, are spreading among AI startups. New labs use a high headline valuation to signal confidence and attract talent, while splitting the round lets early backers buy in at a lower price before later investors come in higher. Metis Reasoning's structure — a modest initial raise with a potential follow-on of hundreds of millions — fits that pattern exactly.
The risk side of the ledger is equally clear. None of these companies has yet shipped a product that justifies its valuation, and the gap between demonstrating planning capabilities in benchmarks and deploying them in robotics or science remains wide. As PYMNTS put it in its coverage, a valuation set on a name carries no anchor until a product earns one.
Still, with Graepel's raise, the migration of DeepMind's reinforcement learning leadership into independent labs is now effectively complete, and investors have signaled they are willing to fund a second act for the paradigm that started the modern AI boom. Whether Metis Reasoning can convert AlphaGo's proven playbook into commercial products will determine whether that confidence was foresight or froth. For ongoing coverage of the funding race reshaping AI, follow our AI startup news section.
---
Stay Ahead of AIGet the latest AI news, analysis, and breakthroughs — all in one place.
Read more AI news →
