The man who taught a machine to beat the world at Go thinks AI’s next breakthrough won’t come from scraping more of the internet. It’ll come from machines that learn the way scientists do — by experimenting, failing, and discovering knowledge on their own.
David Silver, the DeepMind researcher who led the AlphaGo and AlphaZero projects, just raised $1.1 billion in seed funding for his new startup Ineffable Intelligence — the largest seed round ever recorded, valuing the London-based company at $5.1 billion before it has a single product on the market.
What Makes This Different
Every major AI company — OpenAI, Anthropic, Google, Meta — is racing to build bigger models trained on more data. More parameters. More tokens. More GPUs. The entire LLM paradigm runs on one assumption: the path to smarter AI is feeding it more human-written text.
Silver doesn’t buy it.
Ineffable Intelligence is building what Silver calls “superlearners” — AI systems that use advanced reinforcement learning to discover knowledge autonomously through experience, rather than pattern-matching against existing human data. Think AlphaZero teaching itself chess from scratch, but scaled to scientific discovery, reasoning, and problem-solving.
The approach is fundamentally different from the transformer-scaling orthodoxy. Where ChatGPT learns by predicting the next word in internet text, a superlearner would learn by interacting with environments, testing hypotheses, and building understanding from the ground up. It’s the difference between memorising a textbook and actually running experiments.
The Investor Lineup Is Absurd
Sequoia Capital and Lightspeed Venture Partners co-led the round. Nvidia and Alphabet (Google) are backing it. The UK Sovereign AI Fund — the British government’s strategic AI investment vehicle — chipped in too, making this as much a geopolitical bet as a commercial one.
That last detail matters. The UK has been watching its crown jewel, DeepMind, operate under American ownership for a decade. Having Silver — a DeepMind co-founder who spent over a decade there — build his next act in London with British government backing is a deliberate strategic signal.
Silver has also pledged to donate all equity proceeds to charity, which is either genuinely admirable or the most effective PR move in startup history. Possibly both.
Why It Matters
This isn’t just another well-funded AI startup. It’s a direct challenge to the dominant paradigm in AI development.
The LLM scaling approach has delivered impressive results, but the cracks are showing. Models are running out of high-quality training data. The cost of marginal improvements is exploding. And the fundamental limitation remains: you can’t discover what isn’t already in your training data.
Silver’s bet is that the next paradigm shift comes from self-supervised discovery — systems that can learn from interaction with environments rather than from static datasets. If he’s right, it wouldn’t just produce better AI; it could produce genuinely new knowledge, things no human has written down because no human knows them yet.
Of course, if he’s wrong, $1.1 billion is a very expensive way to learn that reinforcement learning doesn’t scale the way the LLM crowd hopes it doesn’t.
The NZ Angle
For New Zealand, which is still grappling with how to integrate AI into education and industry, the Ineffable Intelligence story illustrates the widening gap. The UK is explicitly backing alternative AI paradigms with sovereign capital. Meanwhile, NZ’s AI policy conversations are still largely about whether to use ChatGPT in classrooms.
The countries that invest in understanding these paradigm shifts early — not just adopting existing tools — are the ones that’ll benefit when the next wave arrives.
The Bottom Line
$1.1 billion for a company with no product. A founder who already revolutionised AI once. A technical approach that challenges the entire industry’s orthodoxy. And the British government treating it like a national priority.
Ineffable Intelligence might be the most important AI startup you’ve never heard of — or the most overhyped seed round in history. Either way, the bet has been placed.
🔍 The real story isn’t the money. It’s the thesis: that scraping the internet harder isn’t the path to superintelligence. Learning from scratch might be.
