AlphaGo's Creator Raises $1.1B Seed Round to Pursue Superintelligence Through Reinforcement Learning

In the crowded field of AI labs racing toward superintelligence, David Silver has perhaps the most credible track record of anyone who has ever attempted to build one. The researcher who led Google DeepMind’s reinforcement learning team for a decade — producing AlphaGo, AlphaZero, AlphaStar, and AlphaProof, systems that successively conquered go, chess, StarCraft II, and mathematical olympiad problems — has now launched his own lab and raised what may be the largest seed round ever recorded in Europe.

Ineffable Intelligence, the London-based laboratory Silver founded in late 2025, announced on April 27 that it has closed a $1.1 billion seed round at a post-money valuation of $5.1 billion. The round was co-led by Sequoia Capital and Lightspeed Venture Partners, with participation from Nvidia, DST Global, Index Ventures, Google, and the United Kingdom’s Sovereign AI Fund. For a company that is only a few months old, the numbers are almost without precedent.

The Thesis: Reinforcement Learning as the Path to ASI

The intellectual foundation of Ineffable Intelligence is a direct extension of Silver’s life’s work — and a pointed departure from the paradigm that dominates today’s frontier AI development.

Current state-of-the-art large language models, including GPT-5 and Claude 3, are primarily trained on supervised learning: vast datasets of human-generated text, code, and imagery, labeled and curated to teach models what good outputs look like. This approach has produced extraordinary results, but it comes with a fundamental ceiling: the model can only ever be as knowledgeable and capable as the human data it learned from.

Silver’s position, refined across a decade at DeepMind, is that this ceiling is the wrong architecture for the final stretch toward human-level and beyond-human-level intelligence. His AlphaGo breakthrough in 2016 was built on a different principle: reinforcement learning, where an agent learns by interacting with an environment, receiving rewards for successful actions, and iterating toward better and better strategies without ever requiring human demonstrations of the correct move.

AlphaZero, the 2017 successor, went further — starting with only the rules of chess, go, and shogi and playing against itself until it surpassed every human and human-informed chess engine ever built, in a matter of days. No human chess data was ever used. The system discovered entirely novel strategies that human players had never considered.

Ineffable Intelligence intends to extend this approach beyond games and into the full domain of human knowledge. Silver has described the goal as building “a superlearner that discovers all knowledge from its own experience, from elementary motor skills through to profound intellectual breakthroughs.” The company’s stated mission is to “make first contact with superintelligence.”

Why Now, Why $1.1 Billion

The scale of the seed round reflects both the credibility of the founding team and the specific capital requirements of cutting-edge reinforcement learning research. RL training is computationally expensive in ways that differ from supervised learning: because the agent must generate its own training data through interaction, the compute requirements for sophisticated RL regimes at frontier scale are enormous. The $1.1 billion seed is, in significant part, a commitment to the infrastructure needed to run serious RL experiments at a scale competitive with the world’s best-resourced AI labs.

Silver’s pedigree makes the investment decision easy to justify. With more than 300,000 academic citations to his name, he is one of the most cited researchers in the history of machine learning. His systems at DeepMind didn’t just win games — they consistently solved problems that experts had believed were decades away from machine solution. AlphaGo’s 2016 victory over world champion Lee Sedol was predicted by most AI researchers to be a decade away. AlphaProof’s performance on International Mathematical Olympiad problems in 2024 was similarly ahead of schedule.

The investor roster amplifies the signal. Sequoia and Lightspeed are the two most consistent backers of category-defining AI labs globally; their co-leadership of the seed is a strong endorsement of both the technical thesis and the go-to-market potential. Nvidia’s participation carries its own significance: the chip giant does not typically join seed rounds without a view toward the training infrastructure requirements the company will create. Google’s involvement, meanwhile, creates an interesting dynamic given Silver’s decade at DeepMind — the company’s former employer has a direct stake in the success of the approach it pioneered.

Reinforcement Learning’s Resurgence

Ineffable Intelligence is not alone in betting on reinforcement learning as a key to frontier AI. OpenAI’s o3 and o4 models use reinforcement learning extensively for reasoning tasks, applying RL to teach models to generate and evaluate intermediate reasoning steps before arriving at an answer. Anthropic and Google DeepMind have both invested heavily in RL-based approaches for specific capability domains.

What sets Ineffable Intelligence apart — at least in terms of stated ambition — is the scope: not RL applied to specific domains, but RL as a general architecture for discovering knowledge across all domains without relying on human data as a substrate. This is a much harder problem. Human data serves as a convenient shortcut that encodes millennia of human reasoning and knowledge; removing that shortcut means the system must build its own internal representations of the world from scratch, through trial and error, at a scale that makes even current frontier training runs look modest.

The technical community is divided on whether this approach is more likely to reach AGI before, after, or alongside the current scaling-plus-RLHF paradigm. Silver’s career suggests that when he makes a claim about what RL can achieve, the timeline is often shorter than the skeptics expect.

London’s AI Moment

The $1.1 billion figure is officially the largest seed round in European history, according to the company’s own announcement. It reflects a broader pattern: London has emerged as a credible alternative to San Francisco for frontier AI research, home to DeepMind, Stability AI’s founding team, and now Ineffable Intelligence.

The UK’s Sovereign AI Fund participation is particularly notable — a government-backed investment in a domestic AI lab pursuing superintelligence is a policy statement as much as a financial one. In the context of global AI competition, the British government is placing a direct bet that homegrown expertise in reinforcement learning may produce systems of strategic national importance.

Silver has not specified a product roadmap or a commercialization timeline. The lab is, for now, in pure research mode — which is precisely what a $1.1 billion seed round is designed to protect. The only deliverable stated publicly is “first contact with superintelligence.” Whether Ineffable Intelligence achieves that or not, the bet that reinforcement learning — the engine behind the most surprising AI breakthroughs of the last decade — still has its most consequential results ahead of it deserves to be taken seriously.