David Silver, the former Google DeepMind principal research scientist who led the team that created AlphaGo, is raising $1 billion for Ineffable Intelligence — a new London-based AI startup valued at $4 billion. Sequoia Capital is leading the round. Silver's credentials are unmatched: AlphaGo's defeat of world champion Lee Sedol in 2016 was the moment the world took AI seriously. AlphaZero generalized that approach to chess, shogi, and Go. AlphaFold revolutionized protein structure prediction. If Silver is building something new, the AI world is paying attention.

The specific focus of Ineffable Intelligence hasn't been fully disclosed, but Silver's research trajectory points toward general-purpose AI systems that learn from experience — moving beyond the current paradigm of static pre-trained models. His work at DeepMind centered on reinforcement learning: AI systems that improve through interaction with their environment. The transition from academic research to a $4 billion startup signals that Silver believes the technology is ready for production applications.

The timing is significant: the AI industry's biggest unsolved problem is exactly what Silver spent his career researching — systems that learn and remember from experience rather than starting from scratch each time.

Why AlphaGo's Creator Matters for AI Memory

AlphaGo didn't just play Go. It learned Go. Through millions of self-play games, it developed strategies that human players had never discovered in thousands of years of competition. AlphaZero generalized this approach — given only the rules of a game, it could teach itself to play at superhuman level. The key insight was that experience-based learning, when combined with sufficient compute, produces capabilities that pure static knowledge cannot.

This insight is directly relevant to the memory problem in current AI systems. Large language models are static once trained. They encode knowledge during pre-training and fine-tuning but don't learn from individual interactions at inference time. Each conversation is independent. Each deployment generates experience that never feeds back into the model's capabilities. It's as if AlphaGo played millions of games but forgot every game before the next one.

Silver's entire career has been about closing this gap — building AI systems where experience improves performance. Ineffable Intelligence likely represents his vision for bringing that approach to the broader AI landscape, beyond games and into general-purpose applications.

The Reinforcement Learning Connection to Memory

Reinforcement learning and persistent memory share a fundamental principle: past experience should inform future decisions. In RL, an agent maintains a value function that encodes what it has learned about the environment. Every interaction updates this function, making future decisions more informed. The agent doesn't just react to the current state — it draws on accumulated experience to choose optimal actions.

Current LLM-based AI systems lack this property. Each interaction is evaluated using the model's static parameters plus whatever context fits in the window. There's no value function that accumulates experience. There's no mechanism for one interaction to improve the quality of the next. The model is as capable — and as ignorant of recent experience — in its millionth interaction as in its first.

The gap between RL systems that learn from experience and LLMs that don't is one of the defining tensions in AI. Silver's new lab presumably aims to bridge this gap — creating systems that combine the language understanding of LLMs with the experiential learning of RL. This is exactly the problem space where persistent memory is the critical infrastructure.

What $1B in Funding Signals About AI's Direction

Sequoia doesn't write billion-dollar checks on incremental improvements. The funding level signals that Silver's vision represents a paradigm shift — not a better chatbot, but a fundamentally different approach to AI systems. At $4 billion pre-revenue valuation, the market is pricing in the possibility that Ineffable Intelligence will solve one of AI's core limitations.

The London location is also significant. Opening alongside OpenAI's London expansion and Google DeepMind's continued presence, the UK capital is becoming the global center for AI research. Silver's deep connections to the London AI community — forged through decades at DeepMind and University College London — give Ineffable Intelligence immediate access to a talent pool that money alone can't buy.

For the broader AI ecosystem, the funding validates a thesis that memory and learning-focused approaches represent the next frontier. The biggest rounds in early 2026 — Anthropic's $30B, ElevenLabs' $500M, Runway's $315M, and now Silver's $1B — collectively signal that the AI industry is moving beyond scaling pre-training toward systems that are more adaptive, more persistent, and more capable of learning from deployment.

Experience Memory as AI Infrastructure

Whether Ineffable Intelligence builds its own memory infrastructure or leverages existing solutions, any system that learns from experience requires persistent memory. RL agents need to store experiences (state, action, reward, next state) in replay buffers. LLM-based agents need to store interaction experiences (context, response, outcome, feedback) in memory systems. The specific format differs, but the requirement is the same: past experience must be accessible to inform future decisions.

This is where MemU's architecture aligns with the trajectory that Silver's work represents. MemU provides persistent experience memory for AI systems — storing not just conversation history but structured knowledge about what happened, why it mattered, and how it should influence future behavior. For RL-informed AI systems, this is the replay buffer scaled to production deployments.

How MemU Provides Experience Memory Infrastructure

MemU is built for exactly the future that Ineffable Intelligence represents: AI systems that learn from experience. Every interaction generates structured memories that capture not just what happened but what was learned. These memories are indexed, retrievable, and available to inform future decisions — whether the agent is an RL system, an LLM, or a hybrid architecture that combines both.

For developers building on any of the emerging experience-learning frameworks, MemU provides the memory infrastructure layer. Write experiences from any agent, retrieve them before any decision, and let accumulated knowledge compound over time. This is the persistent learning loop that current AI systems lack and that the next generation will require.

David Silver is building the next paradigm in AI. MemU provides the memory infrastructure that paradigm will need.

Get Started

Build experience-learning AI systems with persistent memory. Explore MemU at memu.pro and on GitHub.