David Silver's Ineffable Intelligence Raises $1.1B to Replace Human Training Data

Ineffable Intelligence, a London-based AI lab founded months ago by former DeepMind researcher David Silver, has closed a $1.1 billion seed round at a $5.1 billion valuation — a so-called "coconut round," the industry's term for outsized early financing clustering around star-researcher-led AI labs.

The bet is architectural: Silver believes human-generated data is a ceiling for AI capability, not a foundation. Ineffable's stated goal is to build a "superlearner" that discovers knowledge and skills entirely through reinforcement learning — the trial-and-error paradigm Silver used at DeepMind to produce AlphaZero, a program that mastered chess and Go without studying human games. Silver spent more than a decade at DeepMind leading the reinforcement learning team; that approach produced systems that defeated the world's top programs in both games through self-play alone. The new lab aims to generalize that principle beyond games to broad reasoning and knowledge acquisition.

The round was led by Sequoia Capital and Lightspeed Venture Partners, with Index Ventures, Google, Nvidia, the British Business Bank, and Sovereign AI — the U.K.'s newly launched sovereign venture fund for AI — participating. The investor list signals both commercial and geopolitical conviction: Nvidia and Google have direct infrastructure interests in a post-LLM training paradigm; the British Business Bank and Sovereign AI reflect the U.K. government's push to anchor frontier AI development in London.

The dominant LLM scaling playbook depends on massive licensed or scraped text corpora, creating data-moat advantages for incumbents and ongoing legal exposure around copyright. A credible reinforcement-learning-first alternative would decouple model capability from data provenance entirely, removing both the compliance risk and the competitive moat that large proprietary datasets currently provide. Companies would need to reexamine procurement and build-versus-buy decisions around data pipelines and fine-tuning infrastructure if Ineffable's approach proves viable at general-reasoning scale.

That "if" is load-bearing. Self-play and world-model techniques have produced superhuman performance in bounded environments — games with fixed rules and clear reward signals. Generalizing to open-ended domains requires specifying what "winning" means across arbitrary tasks, a problem that remains unsolved. Silver's site frames the ambition in sweeping terms: "If successful, this will represent a scientific breakthrough of comparable magnitude to Darwin: where his law explained all Life, our law will explain and build all Intelligence." That is a claim about long-run research trajectory, not near-term product delivery.

Ineffable fits a pattern of researcher-led lab formations attracting outsized early capital. Last month, AMI Labs — co-founded by Turing Award winner Yann LeCun — raised $1.03 billion at a $3.5 billion pre-money valuation. Recursive Superintelligence, co-founded by DeepMind's former principal scientist Tim Rocktäschel and incorporated in the U.K., reportedly raised $500 million with enough demand to extend to $1 billion. The pattern positions London as a competing frontier-AI geography to the Bay Area.

Several former DeepMind staffers are reportedly joining Ineffable's executive team. Silver is not building from scratch; he is reconstituting a team with deep experience in the technical paradigm he is scaling. The $1.1 billion buys substantial compute runway for a pre-product lab. Whether reinforcement learning without human data can generalize beyond games is the research question that will define the return on that capital.