Demis Hassabis and AI's next chapter

I finished Sebastian Mallaby’s The Infinity Machine on 5 August. Before the day was over, the book had acquired an unexpected epilogue.

Demis Hassabis announced that he was handing Google DeepMind’s day-to-day operations to Koray Kavukcuoglu. He becomes Chair of Google DeepMind and Chief Scientist of Alphabet, while remaining involved in AGI strategy and continuing to lead Isomorphic Labs.

His message to staff included the sentence that frames the transition:

“It’s time for AI to prove its unequivocal value to the world.”

That line stayed with me because it cuts through much of the current AI conversation. Today, much of that conversation is dominated by large language models. US companies remain prominent at the proprietary frontier, while Chinese labs have produced powerful open-weight alternatives and narrowed the performance gap. Hassabis’s story is a reminder that another important path began in Europe: DeepMind progressed from training AI agents on existing games to AlphaGo, AlphaFold and now medicine.

An unlikely route to the Nobel Prize

What amazed me most was that someone without formal training in chemistry could share the Nobel Prize in Chemistry.

Hassabis was a chess master at thirteen. At seventeen, he worked with Peter Molyneux on Theme Park. He studied computer science, built games and later completed a PhD in cognitive neuroscience. His route into chemistry passed through computer science, neuroscience and artificial intelligence.

That is not a weakness in the story. It is the point.

Games gave DeepMind controlled environments in which intelligence could be measured and trained. AlphaGo showed that a machine could find solutions that did not come from copying a human playbook. AlphaFold took the same underlying ambition into biology: could a learning system discover useful structure in a domain too complex for people to search directly?

The result was more consequential than another benchmark. AlphaFold2 transformed protein-structure prediction, and around 200 million predicted structures were made available to researchers. In 2024, Hassabis and John Jumper shared half of the Nobel Prize in Chemistry for protein-structure prediction.

AlphaFold does not eliminate laboratory work, and it does not prove that every scientific problem will yield to AI. It does show what “unequivocal value” can mean: not a more fluent chatbot, but a tool that changes how scientists begin an investigation.

Not a victory lap

Mallaby’s book is valuable because it does not turn Hassabis into a flawless visionary.

He underestimated language models, believing that symbols detached from physical experience could not produce genuine understanding. DeepMind invested heavily in reinforcement learning while language models improved rapidly through more data and compute.

The book also documents failed attempts at AI governance. Safety boards, review panels and plans for greater independence repeatedly collided with corporate incentives and competitive pressure. Scientific brilliance did not solve institutional design, and concern about AI risk did not stop the race to build more capable systems.

That tension still matters. Hassabis wants AI to create a new age of scientific discovery, but he also helped build one of the organisations driving the frontier race.

Europe created it, America scaled it

DeepMind was founded in London in 2010 and acquired by Google in 2014. The easy version of this story says that Europe created the innovation and America captured it. The truth is more complicated.

Google supplied capital, computing infrastructure and research runway that helped make AlphaGo and AlphaFold possible. Hassabis and his co-founders made a rational trade: more capacity in exchange for less independence. Without that scale, some of DeepMind’s achievements might not have happened.

But when this trade repeats across strategic industries, it becomes more than an isolated business decision.

I have spent most of my career in the European semiconductor industry working for US-controlled companies, including Synopsys, Emerson Automation and Sigma Designs. At Sigma Designs, I was there when Silicon Labs acquired its Z-Wave business. Nanopower, with whom I currently collaborate, is an exception: a semiconductor company controlled by European capital.

Recent data gives this pattern scale. Hello Tomorrow’s 2026 Semiconductor Innovation Report, which analyses more than 1,500 startups, estimates that Europe produces about 25% of the world’s semiconductor startups but attracts only about 7% of global funding. Only 6% progress beyond Series A, compared with 24% in the US and 37% in China. Europe’s problem is not creating semiconductor companies. It is financing them through scale-up.

This experience changes how I read DeepMind’s story. Europe clearly has the engineers, researchers and ideas. What it often lacks is the capital and institutional capacity to retain control as those ideas scale. The loss is not only financial. It affects economic sovereignty, technological sovereignty and, in strategic sectors, security.

At the Porto Business School Alumni Day last July, Arturo Bris gave a keynote focused precisely on this problem. His argument was not that Europe lacks innovation, but that it fails to mobilise its capital: European savings and institutional money finance US companies, which then acquire European innovation. DeepMind was one of his examples.

The point is not that foreign capital is bad. DeepMind itself shows why access to US capital and infrastructure can be transformative. The problem is dependence. If European companies repeatedly need foreign ownership to reach global scale, Europe loses the ability to choose where strategic technology is developed, governed and applied.

The next chapter

Hassabis is now creating more room for AGI strategy, science and Isomorphic Labs. The next test is medicine, where success requires far more than pattern recognition. It requires experiments, clinical evidence, manufacturing, regulation and time.

That makes his challenge more credible, not less. AlphaFold proved that AI could cross from computer science into scientific discovery. Isomorphic Labs must show whether it can cross from discovery into safe and effective therapies.

I recommend The Infinity Machine because it explains both the person and the institution: the ambition, the mistakes, the breakthroughs and the compromises required to operate at this scale.

It also left me with a European question. DeepMind and the path from games to AlphaFold began here. Europe should remain open to global capital and collaboration, but openness without the capacity to scale our own strategic companies becomes dependency.

Europe does not merely need more innovation. It needs the capital and institutions required to retain meaningful control of the innovation it creates.

References

1. Demis Hassabis and Sundar Pichai, “The next chapter of our AI momentum”, Google, August 2026.
2. Sebastian Mallaby, The Infinity Machine: Demis Hassabis, DeepMind and the Quest for Superintelligence, Penguin Press, 2026.
3. The Nobel Prize, “The Nobel Prize in Chemistry 2024”, October 2024.
4. European Investment Bank, “The scale-up gap”, 2024.
5. Stanford Institute for Human-Centered AI, AI Index Report 2026, 2026.
6. Hello Tomorrow, 2026 Semiconductor Innovation Report, 2026.