David Silver left DeepMind and raised $1.1B to teach AI to learn from scratch

David Silver is one of those names in AI that carries real weight. He was a key figure at DeepMind, co-invented AlphaGo, and spent years pushing reinforcement learning forward. So when he quietly left DeepMind a few months ago to start something new, people paid attention.

Now we know why. His new company, Ineffable Intelligence, just announced a $1.1 billion funding round at a $5.1 billion valuation. That’s not a typo. A lab that barely existed months ago is already worth over five billion dollars.

The pitch: learning without human data

Silver’s core thesis is something he’s been talking about for years: current AI is too dependent on human-generated data. We scrape the internet, we label datasets, we fine-tune on human preferences. But that approach has limits. Human data is messy, biased, and finite. If you want AI that can truly generalize and discover new things, you need it to learn from its own experience.

That’s the promise of reinforcement learning at scale. Instead of feeding the model human examples, you let it explore an environment, try things, fail, and figure out what works. AlphaGo learned to play Go at superhuman level this way, but Silver wants to apply the same principle to real-world problems. Robotics, scientific discovery, maybe even software engineering.

The $1.1B will go toward building the compute infrastructure to train these models. That’s the expensive part. Training an agent that learns entirely from scratch in a complex environment requires massive amounts of simulation and compute. We’re talking thousands of GPUs running for months.

Is this actually new?

Reinforcement learning isn’t new. Silver himself helped pioneer it. What’s new is the scale and the ambition. Most RL research has been confined to games and simulated environments because real-world RL is brutally hard. You can’t just let a robot wander around a factory floor for a million hours learning by trial and error. The hardware wears out, the physics is messy, and safety constraints get in the way.

So the real question is whether Silver has figured out something the rest of the field hasn’t. He’s not saying much publicly yet, which is typical for someone with his track record. But the investors are betting he can pull it off. The round was led by some big names in venture capital, and the valuation reflects the belief that this could be the next paradigm shift.

I’m cautiously optimistic. Silver has the credentials and the vision. But I’ve seen too many ambitious AI labs raise huge sums and then struggle to deliver. The field is littered with companies that promised to reinvent AI and ended up pivoting to something more mundane. I hope Ineffable Intelligence is different.

What this means for the rest of us

If Silver succeeds, it could change how we think about AI training. Instead of hoarding data and worrying about copyright and licensing, we might shift toward models that generate their own training signals. That would be a big deal for privacy, for scientific discovery, and for domains where human data is scarce.

But even if he fails, this investment signals something important: the market believes there’s still room for radical new approaches. Not everything has to be another large language model trained on the public internet. Reinforcement learning at scale is a bet worth taking.

I’ll be watching closely. Silver doesn’t hype things up lightly, and his track record at DeepMind speaks for itself. But $1.1B is a lot of money, and the pressure to show results will be intense from day one.