David Silver walked out of DeepMind and raised $1.1B to build an AI that doesn’t need us

David Silver, the DeepMind researcher who helped shape modern reinforcement learning, has raised $1.1 billion for a new lab called Ineffable Intelligence. The company is barely a few months old and already carries a $5.1 billion valuation. That’s a lot of money for a lab with no product, no revenue, and probably not much more than a whiteboard and some GPUs.

But Silver isn’t some random founder cashing in on hype. He was a key figure behind AlphaGo, AlphaZero, and the reinforcement learning algorithms that underpinned much of DeepMind’s early success. If anyone has earned the right to ask for a billion dollars on a vision, it’s him.

The pitch is straightforward: build AI that learns without relying on human-generated data. No scraping the internet for text, no labeling images by hand, no fine-tuning on human preferences. Just an agent that figures things out by interacting with an environment, the way a child learns to walk or a wolf learns to hunt.

This is the reinforcement learning dream that Silver has been chasing for years. The problem is that RL has never quite scaled the way supervised learning did. AlphaZero crushed games, but it needed millions of self-play games to get there. DeepMind’s work on RL for robotics was impressive but brittle. The real world is noisy, expensive to explore, and doesn’t reset when you fail.

Still, the timing might finally be right. Compute is cheaper than it was five years ago. Simulators are getting better. And the limitations of the current paradigm — massive datasets, human annotation, scaling laws that seem to hit diminishing returns — are becoming obvious even to the people who benefited from them.

Ineffable Intelligence is reportedly focusing on what they call “intrinsic motivation” systems, where the AI sets its own goals and rewards rather than waiting for humans to define them. That’s been a research area for years, but nobody has made it work at scale. Silver might have some ideas that haven’t been published yet.

The valuation is the part that makes me uncomfortable. $5.1 billion for a company that hasn’t shipped anything is a bet on reputation, not results. Silver is brilliant, but so were a lot of founders who burned through huge rounds without delivering. The difference here is that the investors — and the reports suggest sovereign wealth funds and big tech VCs are involved — are betting on a paradigm shift, not incremental improvement.

If Silver pulls this off, it changes the entire AI industry. No more data moats, no more scraping the entire internet, no more worrying about copyright lawsuits over training data. An AI that learns from scratch, in the wild, is the kind of thing that makes current foundation models look like parlor tricks.

If he doesn’t, $1.1 billion buys a lot of compute and a very nice severance package for a lot of talented researchers. Either way, it’s going to be fun to watch.