David Silver walks away from DeepMind with $1.1B for a new kind of AI

David Silver, the DeepMind researcher who helped pioneer reinforcement learning and was a key figure behind AlphaGo, has raised $1.1 billion for a new AI lab called Ineffable Intelligence. The company, founded just a few months ago, is already valued at $5.1 billion.

That’s a lot of money for a lab with no product, no revenue, and not much more than a founder’s reputation. But Silver’s reputation is the kind that makes VCs open their checkbooks. He co-created AlphaGo, the system that beat the world champion at Go in 2016, and was a lead on AlphaZero, which taught itself to play Go, chess, and shogi from scratch with only the rules.

Ineffable Intelligence’s pitch is ambitious: build AI that learns without relying on human-generated data. That’s a direct shot at the current paradigm where models like GPT-4 and Gemini are trained on massive scrapes of the internet — text, images, code — all produced by humans. Silver wants to move past that.

This is higher than I expected for a company that’s essentially a research project at this stage. The valuation implies investors believe Silver can crack something the rest of the field has been struggling with: how to make AI learn from its own experience, not from human examples.

The approach builds on reinforcement learning, Silver’s specialty. In RL, an agent learns by interacting with an environment and receiving rewards for desirable outcomes. AlphaZero did this for games. The question is whether you can scale that to real-world domains where the reward function isn’t as clean as “you won the game.”

Silver’s been talking about this for years. In a 2024 interview, he argued that the current reliance on human data creates a ceiling — models can only be as good as the data they’re trained on. He called for a shift toward what he termed “agentive AI,” systems that generate their own training signals through interaction.

Ineffable Intelligence isn’t the only lab chasing this. DeepMind itself has projects in self-supervised and unsupervised learning. OpenAI has been working on reinforcement learning from human feedback, which is a hybrid approach. But Silver’s bet is more radical: no human data at all, not even for fine-tuning.

The funding round was led by a mix of sovereign wealth funds and Silicon Valley VCs, according to sources close to the deal. That’s unusual for a UK-based lab. Most British AI startups get funded by European VCs or corporate arms. Silver’s pulling in global money, which signals that investors see this as a potential paradigm shift.

There are obvious challenges. Reinforcement learning is sample-inefficient — it takes millions of episodes to learn simple tasks. Scaling that to complex, open-ended environments without human priors is computationally expensive. Silver’s team will need serious hardware, and $1.1B buys a lot of GPUs, but not infinite patience.

The name “Ineffable Intelligence” is a bit much. It sounds like something from a Neal Stephenson novel. But Silver’s track record earns him the right to be grandiose. He’s been right before when everyone else was skeptical.

What I’m watching for is whether he can attract top talent. DeepMind and OpenAI are still the gravitational centers for AI researchers. Silver’s reputation will pull some people, but the $5.1B valuation creates pressure. If they don’t show results within a couple of years, the narrative flips from “visionary” to “overhyped.”

For now, this is the most interesting AI funding story of the year. Not because of the money — though $1.1B is a lot — but because it represents a genuine bet on a different path. Not scaling up the same transformer architecture, not adding more human data, but trying to build something that learns the way animals do: by doing, not by reading.