David Silver, the DeepMind researcher who built AlphaZero, just pulled in $1.1 billion for a new AI lab that wants to skip human data entirely. According to TechCrunch AI, Silver’s London-based startup, Ineffable Intelligence, raised the round at a $5.1 billion valuation only a few months after launching. Sequoia Capital and Lightspeed Venture Partners led the deal, with Google, Nvidia, Index Ventures, the British Business Bank, and the UK’s new Sovereign AI fund all piling in.
This is a big deal, and not just because of the size of the check.
What Ineffable is actually building
Silver wants to build what he calls a “superlearner.” The pitch: an AI that figures things out the same way AlphaZero figured out chess and Go, by playing against itself and learning from experience instead of training on human-generated text or examples. No scraped internet, no human feedback loop, no curated datasets. Just trial, error, and reinforcement learning at scale.
That’s Silver’s home turf. He led DeepMind’s reinforcement learning team for over a decade before leaving, and his name is on the systems that crushed the world’s strongest chess and Go engines without ever being shown a human move. AlphaZero is the headline example.
The company’s site frames the ambition in terms that would make a marketing team blush: “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.” Silver calls Ineffable his “life’s work” and told Wired any money he personally makes will go to high-impact charities.
Why this matters
Large language models are the dominant paradigm right now, and they’re running up against the limits of available human-generated data. Several frontier labs have started warning that fresh text is getting harder to source. Silver’s bet is that the next leap doesn’t come from more data. It comes from agents that generate their own experience.
If that approach works at general-purpose scale, it sidesteps the data wall entirely. If it doesn’t, $1.1 billion is a very expensive science experiment.
The ‘coconut round’ pattern
Ineffable joins a small but growing club of AI labs founded by celebrity researchers and funded with seed-stage checks the size of late-stage rounds. TechCrunch AI reports the industry has started calling these “coconut rounds,” a tongue-in-cheek escalation of the usual “seed” label.
Recent entries in the same category:
- AMI Labs: Co-founded by Yann LeCun (Turing Award winner, former Meta AI chief). Raised $1.03 billion at a $3.5 billion pre-money valuation last month.
- Recursive Superintelligence: Co-founded by former DeepMind principal scientist Tim Rocktäschel. Reportedly raised $500 million, with demand pushing it toward $1 billion.
- Ineffable Intelligence: $1.1 billion at a $5.1 billion valuation. Pentacorn status before shipping anything.
What stands out here: each bet points away from today’s transformer-only playbook. LeCun has been openly skeptical of the LLM-only path. Silver is coming at it from reinforcement learning. Investors are clearly hedging.
London is quietly turning into an AI capital
The geography is worth flagging too. Ineffable, Recursive Superintelligence, and a chunk of the alumni networks behind them all trace back to DeepMind’s London base, which Google bought in 2014 and has kept as a major research center. Several former DeepMind staffers are reportedly joining Ineffable’s executive team.
Jeff Bezos’ new AI lab, Project Prometheus, is reportedly hunting for office space near Google’s London AI hub. The UK government is pouring sovereign capital in through Sovereign AI. The talent network and the funding are coalescing in the same square mile.
What to watch next
A few things worth tracking:
- The team announcement. Who actually leaves DeepMind for Ineffable will tell us how serious the technical bet is.
- First public demos. Reinforcement learning labs usually show a narrow proof point before going broad. Watch for that.
- The data-wall narrative. If LLM scaling keeps stalling, the “no human data” pitch gets a lot more compelling, and these valuations start looking less crazy.
More details are available at the original TechCrunch AI report.