Ex-DeepMind AlphaGo co-creator launches London AI startup Metis Reasoning
Thore Graepel's new venture is raising tens of millions of dollars to pursue a compute-efficient alternative to frontier AI models from OpenAI and Anthropic.
New London AI venture unveiled
A London-based AI company founded by former Google DeepMind research lead Thore Graepel has launched, aiming to pursue what it calls a “compute-efficient” alternative to the frontier AI models built by Anthropic and OpenAI. The company, named Metis Reasoning, is understood to be initially raising tens of millions of dollars.
Graepel, who left Google DeepMind this summer and is a co-creator of AlphaGo, the programme that beat human players at Go, has assembled a 12-person team of researchers, mathematicians, robotics experts and entrepreneurs. The founding team also includes Piers Ingram, a serial technology and biotech entrepreneur who has raised more than $350m across his ventures; Michael Warren, a robotics and autonomous systems expert with a background at Boeing; and David Berman, head of AI at Cambridge Consultants and an adviser to the UK government and NATO on AI.
A different approach to reasoning
Metis, named after the Greek goddess of wisdom, is betting on an architectural approach to AI reasoning rather than the large language model path favoured by most frontier labs. The company argues that while LLMs have made significant progress and can produce fluent, instruction-following answers, they have a fundamental limitation in how they reason, making them unreliable and lacking a clear audit trail. Metis says this is a key barrier to wider AI adoption, particularly in higher-stakes fields such as engineering, science, medicine and law.
Ingram, who met Graepel at a Cambridge barbecue where the pair discussed Turing machines, said: “When you think about higher-stakes cases in engineering or science, or medical or legal reasoning, the arguments that the models come up with basically can’t be trusted. I don’t think that’s a good foundation for what we want our AI to be.”
The company’s thesis draws on lessons from Graepel’s work on AlphaGo, arguing that a similar architectural approach can be applied to AI reasoning more broadly. Metis believes this could prove particularly suited to robotics and engineering applications.
Funding and compute efficiency
Ingram declined to comment directly on reports that Metis is raising tens of millions of dollars, but said: “We have been able to fund the company more than adequately to date.” He said the company’s architecture was “much much more compute-efficient” than frontier models, adding: “We are very much not going in this direction of ever larger models and ever larger training budgets.” Ingram said Metis had achieved “encouraging” results on frontier reasoning problems at a fraction of the cost of current leading models.
Asked why larger rivals such as Anthropic and OpenAI were not pursuing a similar path, Ingram said: “I think they are thinking about it differently.”
Part of a wider London cluster
Metis joins a growing cluster of new AI companies launching out of London, including Ineffable Intelligence, founded by Graepel’s former AlphaGo colleague David Silver, and Recursive Superintelligence, co-founded by computer scientist Richard Socher. Ingram said there was likely little overlap between the three ventures, describing Metis’s approach as “really a very different, almost philosophical approach.”
What happens next
It remains unclear when Metis will publicly launch its first AI model, or who is backing the company’s early funding. The venture adds to signs that London is becoming a hub for AI startups pursuing approaches distinct from the dominant large language model paradigm, a trend likely to draw further investor interest in the coming months.