David Silver, the computer scientist who led the creation of AlphaGo and AlphaZero at Google DeepMind, is raising $1 billion for a London-based startup called Ineffable Intelligence. Sequoia Capital is leading the round at a $4 billion pre-money valuation. If completed, it would be the largest seed round ever raised by a European startup, dwarfing Mistral AI’s 105 million euro seed in 2023.
The bet: that large language models alone will never produce true superintelligence, and that the path forward runs through reinforcement learning and world models instead.
Who Is David Silver
Silver spent more than a decade at Google DeepMind, joining full-time in 2013 after consulting since the lab’s founding. He led the AlphaGo project that defeated world champion Lee Sedol at Go in 2016, then built AlphaZero, which taught itself Go, chess, and shogi from scratch without any human training data. He won the 2019 ACM Prize in Computing for these contributions and holds a professorship at University College London.
He left DeepMind late last year to found Ineffable Intelligence. The company has no product yet.
The Technical Thesis
Silver’s argument against LLMs is specific: training on human text gives you a ceiling of human-level performance. To go beyond that, AI systems need to learn from their own experience.
In a paper co-authored with Richard Sutton — the godfather of reinforcement learning — Silver argued that the next breakthrough would come from “the era of experience,” where agents learn predominantly through interaction with their environment rather than from static human-generated datasets. Instead of training once on internet text and shipping a frozen model, these systems would use internal “world models” — simulations that let them predict the consequences of their actions — and continuously improve over months or years of operation.
This is the approach that powered AlphaGo and AlphaZero. Where those systems mastered board games, Silver’s stated goal for Ineffable Intelligence is to build “an endlessly learning superintelligence that self-discovers the foundations of all knowledge.”
The Money
Sequoia Capital managing partner Alfred Lin and partner Sonya Huang reportedly flew to London to meet Silver personally. Nvidia, Google (DeepMind’s parent Alphabet), and Microsoft are all in discussions to invest, though negotiations are still ongoing and terms could shift.
The $4 billion valuation for a pre-product company is extreme by any standard. For context, Mistral AI — which had working models and revenue — was valued at roughly $2 billion at its Series A in late 2023. Silver is commanding a 2x premium on reputation and research pedigree alone.
But the investor logic tracks. If there is even a small probability that Silver’s reinforcement learning approach produces capabilities that LLMs cannot, the upside dwarfs the cost of the bet. And Silver has done it before: AlphaGo and AlphaZero were both considered impossible by the mainstream AI community until they worked.
The Strategy
Ineffable Intelligence represents a growing counternarrative in AI research. While the industry has poured hundreds of billions into scaling transformer-based language models — the architecture behind ChatGPT, Claude, and Gemini — a faction of elite researchers is arguing that this approach has fundamental limits.
Silver is not alone. Yann LeCun, formerly Meta’s chief AI scientist, raised 500 million euros for AMI Labs at a 3 billion euro valuation, also pursuing alternatives to pure language model scaling. The shared premise: predicting the next token in a text sequence, no matter how much compute you throw at it, may not be sufficient for general intelligence.
The difference between Silver and most LLM skeptics is his track record. He has built systems that genuinely surpassed human capability in constrained domains. The question is whether reinforcement learning techniques that conquered Go can be generalized to open-ended real-world problems.
Who Wins, Who Loses
Winners: London’s AI ecosystem gets a flagship lab. Sequoia diversifies its AI portfolio beyond LLM bets. If Silver’s approach works, early investors in a $4 billion company own a piece of something potentially worth orders of magnitude more.
Losers: No one yet — this is a long-duration research bet, not a near-term market disruptor. But if reinforcement learning labs start producing results that LLMs cannot match, the hundreds of billions currently flowing into language model infrastructure could face a brutal revaluation. The companies most exposed would be those whose entire value proposition depends on LLM scaling continuing to deliver returns.
The most interesting dynamic is that Google, Nvidia, and Microsoft — all deeply invested in LLM scaling — are considering backing a venture built on the premise that LLMs are insufficient. That is less hypocrisy than hedging. In a race with uncertain outcomes, the smartest players bet on multiple horses.