Ineffable Intelligence is the London lab founded in late 2025 by David Silver, DeepMind's longtime reinforcement-learning lead and the architect of AlphaGo and AlphaZero. It came out of stealth in April with $1.1 billion, Europe's largest-ever seed round, at a $5.1 billion valuation, to build AI that learns without human data. Silver calls human data a fossil fuel.
I have been reading about him in the new Demis Hassabis book, and the thing that strikes me is how consistent David Silver has been. The AlphaZero result, system learning purely from self-play, does seem to be (or at least used to be) his whole philosophy on AI. With AlphaFold I get the impression he resisted deprioritizing RL a lot, even though it didn't work that well in that domain. So is he now really betting against the entire LLM industry? Only LeCun's AMI seems to be similarly heterodox, of the "neolabs", the billion-dollar startups founded by researchers who left the big AI labs.
Frustratingly for forecasting, Ineffable does not have any public output, no papers, no benchmarks, no demos, no timeline. The information vacuum is so complete that the company has no Wikipedia article, and the pages ranking for "when will Ineffable show results" are SEO farms inventing dates. On my set of neolab forecasts its frontier-model median came out at June 2035, a decade out, which to me implies it has no chance of being relevant. So I decided to dig into it a little more.
| Question | Forecast |
|---|---|
| First public result, paper, or demo | Feb 2028 (Feb 2027 to Jun 2030) |
| The domain of its first headline result | Games or simulated environments, 33% (nothing by 2028, 23%) |
| Still UK-headquartered at the end of 2027 | 93% |
First, as I did for the other neolab forecasts, here's the basic factual timeline I anchored to:
- November 2025. Silver leaves DeepMind after two decades and incorporates the company in the UK. The Companies House registered office is in Cheshire.
- April 27, 2026. Out of stealth with the $1.1 billion seed at $5.1 billion, co-led by Sequoia and Lightspeed, with Nvidia and Google as strategic investors and DST, Index, EQT, and Bond participating. The UK government's Sovereign AI Fund, via the British Business Bank, puts in $20 million.
- April 2026. Silver pledges 100% of his personal equity gains to charity through Founders Pledge, the largest commitment in that organization's history, and gives the "fossil fuel" framing for human data.
- Spring and summer 2026. Hiring from the DeepMind reinforcement-learning bench, with early staff carrying Atari, ProcGen, and XLand lineage. No roadmap or timeline is given.
- June 2026. The training cluster comes online.
- Since then, nothing. No paper, no demo, no benchmark, no Wikipedia article.
The median for Ineffable's first public technical artifact is February 2028, roughly two years after formation, with a p90 stretching to mid-2030. The SEO farms claiming a "demo by Q2 2027" are making it up; the forecast puts only about a quarter of the probability before July 2027. The reasoning is straightforward. The giant cluster only came online in June, novel reinforcement-learning environments take one to two years to mature, and Silver has a taste for the single AlphaGo-style reveal over incremental updates. Wired describes him as "cagey" about what the training simulations even are. The early shoulder is real though. A recruitment-driven technical note, or a joint infrastructure paper with Nvidia, could drop at any time.
What does the first result look like? Games or simulated environments lead at 33%, mathematics at 18%, and there is a 23% chance of nothing at all by the end of 2028. Software agents get just 10%, and the reason is a Silver quote I had missed: "generative language, video, code, and more, all are in good hands." He is deliberately not competing where LLMs are already strong. The team roster backs the games call, since the early hires carry Atari, ProcGen, and XLand lineage, and the investor letters talk about "theorems we have not yet proved." Robotics, despite the long-term mission, gets 5%. Hardware iterates too slowly for a first result.
The strangest line in Ineffable's cap table is the UK Sovereign AI fund, a national government taking a direct position in a pre-product research lab, alongside Sequoia, Nvidia, and Google. The forecast says 93% that Ineffable is still UK-headquartered at the end of 2027. The fresh $1.1B removes the classic Delaware-flip pressure, the Sequoia and Lightspeed partners already sit as directors on the UK entity, and only about 3% of European startups ever formally relocate. One detail I enjoyed from the research is that the registered office in Companies House is in Cheshire, not London. The legal analysis found the sovereign fund has no contractual lock-in keeping the company British, so the remaining 7% is real, an acqui-hire or a US-led mega-round flip. But through 2027, the national-champion setup holds.
Silver spent two decades at DeepMind proving that patient, unfashionable bets can produce the field's most famous results. The forecasts here say the patience will be tested in public. My prediction from the bench stands, that whatever Ineffable builds first will surface somewhere leaderboards don't look, and the first-result date above is when we find out where. Watch for reinforcement learning results, and watch the post-reorg DeepMind, because that is the talent pool Ineffable recruits from.
Forecast these yourself in the FutureSearch app , the moment Ineffable publishes anything, or the moment the next DeepMind researcher leaves.