AlphaGo’s David Silver bets on Google Cloud for superintelligence

Ineffable Intelligence, the London-based startup launched by the DeepMind researcher who created AlphaGo, has selected Google Cloud as its infrastructure partner for a frontier AI lab. The partnership, unveiled at Google Cloud’s London summit on June 16, provides a company with no product but enormous ambition with the computing power its founder insists the mission requires.

The weight of this announcement rests squarely on the founder’s shoulders. David Silver spearheaded the research behind AlphaGo, the program that defeated the world champion at Go, and AlphaZero, which mastered chess, shogi, and Go from only the rules. After over a decade at DeepMind leading its reinforcement-learning group, Silver left to establish Ineffable. The venture bets that the self-play methodology underlying AlphaZero can be applied to the entire world.

That bet is called a “superlearner.” Ineffable envisions a system that acquires knowledge through its own experience, from fundamental motor skills to scientific discoveries, without relying on human-generated data. Unlike today’s large models trained on static datasets scraped from the internet, a superlearner would generate, evaluate, and learn from experience in real time. The company frames this goal in cosmic terms, describing it as making “first contact” with superintelligence.

Regardless of the framing, the computing requirements are substantial and distinctive. Experience-based learning, according to the company, imposes different demands on hardware than training on fixed data. It requires tightly coupled training and inference, high-performance networking, and enormous scale. To meet these demands, Ineffable will deploy one of the largest clusters of Google’s A5X instances, powered by Nvidia’s upcoming Vera Rubin NVL72 GPUs.

The choice of partner carries strategic significance. Silver said the lab evaluated the market and selected Google Cloud for its integrated AI Hypercomputer architecture rather than a basic “box of chips” GPU rental. This systems-level setup bundles processors with networking and storage. “Training frontier models requires more than just raw compute,” he stated, framing the decision as one about orchestration rather than sheer power.

For Google Cloud, securing Ineffable is a major win in the fiercely competitive market for AI workloads. Thomas Kurian, its chief executive, said Ineffable is using “our full-stack AI Hypercomputer, from Jupiter networking to our optimized storage,” and pitched the deal as evidence that frontier labs are choosing Google to accelerate progress. Landing a lab of Ineffable’s profile is the kind of reference customer that cloud providers aggressively pursue.

The deal also deepens the Google Cloud and Nvidia partnership, with Nvidia’s Vera Rubin platform anchoring the arrangement, at a time when both companies are jointly pushing their hardware into ever-larger training clusters.

There is a geographic dimension the companies emphasized. By basing the lab in London and building on Google’s infrastructure, Ineffable positions itself as a cornerstone of Europe’s AI ambitions and a magnet for engineering talent that might otherwise move to the United States. For a continent concerned about its dependence on American AI, a homegrown frontier lab of this profile is the kind of project policymakers highlight, even if the underlying compute is American-supplied.

The deal follows a startup origin story that is striking even by current standards. Ineffable raised a $1.1 billion seed round, billed as the largest in European history, and later received backing from Sequoia and Nvidia at a reported $5.1 billion valuation. All of this for a company with no product, no revenue, and no public roadmap. The market is betting on the founder’s track record and the viability of the thesis, not on anything shipped.

That is the wager in plain terms. A researcher with a genuine claim to having built systems that taught themselves, now armed with a billion dollars and a massive cluster, is trying to replicate that success at the scale of all human knowledge. The infrastructure is now in place, but whether the superlearner arrives is something no amount of compute can guarantee.