Google DeepMind's London Power Base Erodes as Hassabis Steps Back to Chairman Role

Demis Hassabis, the Londoner who built DeepMind into the crown jewel of British artificial intelligence before selling it to Google in 2014, is stepping back from running the operation day-to-day. He becomes Chairman of Google DeepMind and Chief Scientist of Alphabet, while Chief Technology Officer Koray Kavukcuoglu — who relocated from London to Mountain View years ago — takes the reins of daily operations as SVP, reporting directly to Sundar Pichai. On paper, it reads as a tidy succession plan. In practice, it is the clearest confirmation yet that the centre of gravity inside one of the world's most consequential AI labs has finished its multi-year migration from the UK to California.

DeepMind was, for a decade, Britain's proudest claim to AI relevance — a Bloomsbury-founded research outfit that out-published rivals, cracked protein folding with AlphaFold, and gave the UK government a talking point every time it wanted to boast about "world-leading" domestic AI capability. That claim is now looking increasingly threadbare. Since Google folded DeepMind together with Google Brain in 2023, authority, headcount decisions, and computing resources have steadily gravitated towards Mountain View, and this leadership reshuffle is less a single event than the final entry in a ledger that has been running for three years.

From Founder-CEO to Chairman: What Actually Changed

Hassabis's new titles sound elevated, and in a narrow sense they are — Chief Scientist of Alphabet is a company-wide remit, not a divisional one. But stepping back from operational control of Google DeepMind removes him from the daily decisions that determine where compute goes, which teams get resourced, and who gets promoted. That authority now sits with Kavukcuoglu, a Turkish-British machine learning veteran who spent years as one of DeepMind's most senior London-based researchers before the job pulled him to California.

Kavukcuoglu's Relocation Was the Tell

Kavukcuoglu's own move from London to Mountain View years ago now reads less like a personal choice and more like an early signal of where Google intended the lab's decision-making to live. He was not alone. Sebastian Borgeaud, who leads coding efforts at the company, made the same journey west. When the people running the technical roadmap physically relocate to the same campus as Google's broader AI infrastructure and leadership, the organisational chart tends to follow the desks, not the other way around.

A Second Senior Departure Compounds the Signal

The reshuffle did not happen in isolation. Jeff Dean — Google's 30th employee, architect of much of its foundational infrastructure, and Alphabet's outgoing Chief Scientist — is leaving after 27 years to found his own venture, Discover Loop AI. Losing the company's most tenured technical leader in the same news cycle as Hassabis's step-back only sharpens the sense that Alphabet's AI leadership bench is being reshuffled from the top down, with Silicon Valley absorbing the authority that flows out of it.

London's Fading Centre of Gravity

The 2023 decision to merge Google Brain and DeepMind into a single organisation was framed publicly as a way to concentrate Alphabet's AI firepower against fast-moving rivals. Internally, according to accounts from people who worked across both offices, it produced something closer to a slow-motion power transfer — one in which London, the founding home of DeepMind, gradually lost the leverage it once had as an independent research culture with its own priorities.

The "America-fication" of a British Institution

One former longtime London-based employee described the cultural shift bluntly: the lab has undergone what they called an "America-fication," characterised by a management style where "it's just how people work in the states" rewards "the loudest person in the room" and produces "a sharper edge" than the more collegial, research-first culture DeepMind was known for in its independent years. Industry analyst Ben Pouladian, commenting on the broader dysfunction inside the merged organisation, noted that the combined team has struggled with decision-making speed, observing that staff "are not able to make decisions" quickly because of "a lot of bureaucracy slowing things down" — friction that reporting lines increasingly resolve by defaulting to Mountain View.

That friction reportedly extended to compute allocation and information access. Teams inside the merged organisation have competed directly for GPU and TPU resources, and in one notable instance a project led by Noam Shazeer lost compute access to a London-based team shortly before Shazeer departed for OpenAI in June 2026. Separately, some London researchers could view what Mountain View teams were building, but the visibility did not run in the other direction — an asymmetry that former staff say bred lasting mistrust between the two offices.

London Versus Mountain View: Where the Power Sits

Google did open a new London office earlier in 2026, a move that on the surface reads as continued investment in Britain. But a fresh building is not the same as retained authority, and the pattern across leadership, compute, and reporting lines tells a more one-directional story.

Indicator	London (DeepMind Origin)	Mountain View
Top operational leadership	Hassabis now Chairman, non-operational	Kavukcuoglu runs daily operations, reports to Pichai
Senior technical leadership relocations	Original base for Kavukcuoglu and Borgeaud	Both relocated here to lead their functions
Physical footprint in 2026	New Google office opened earlier this year	Existing hub for Google Brain legacy and Alphabet AI leadership
Reported cultural trajectory	Described by former staff as losing its distinct research culture	Described as setting the assertive, fast-moving operating norm

The Competitive Pressure Behind the Reshuffle

None of this is happening in a vacuum — it is happening because Google is under sustained pressure to prove it can out-execute OpenAI and Anthropic, and Alphabet leadership evidently believes a more centralised, Mountain View-anchored command structure is the way to do it. Google invented the transformer architecture that underpins the entire modern AI boom, publishing the foundational "Attention Is All You Need" paper in 2017. Every one of that paper's eight authors has since left the company. OpenAI, not Google, turned that research into the product that defined the era.

The Departures Piling Up

The talent bleed has continued into 2026, and it cuts in both directions — researchers leaving Google DeepMind for rivals, and Google pulling its own remaining top talent physically closer to headquarters to try to stem the losses:

• John Jumper, Nobel laureate and AlphaFold co-creator, departed for Anthropic in June 2026
• Noam Shazeer, whose team lost compute priority to a London-based group beforehand, left for OpenAI the same month
• Jeff Dean, Alphabet's Chief Scientist, is exiting after 27 years to launch Discover Loop AI
• Demis Hassabis steps out of daily operational control, even as he retains a company-wide scientific title

For a company trying to project stability and singular strategic direction against rivals that move fast and ship aggressively, a reshuffle of this scale sends a mixed signal — resolve dressed up as reorganisation, but reorganisation triggered by attrition it could not otherwise stop.

What This Means for Britain's AI Ambitions

For the UK, the practical stakes go well beyond one company's org chart. DeepMind has functioned for over a decade as the anchor tenant of Britain's AI credibility — the example the government reaches for whenever it wants to argue that London, not just Silicon Valley, can produce and retain world-class AI research. A steady drain of operational authority to California complicates that pitch considerably.

The UK Talent Pipeline Under Strain

The knock-on effects for British AI talent are already visible in the pattern of relocations rather than redundancies. Senior UK-trained and UK-based researchers are not being made redundant — they are being pulled towards Mountain View to do the jobs that matter most, which has a similar hollowing-out effect on the domestic ecosystem over time. Graduates and mid-career researchers who might once have seen a clear path to senior technical leadership without leaving Britain now face a structure where the most consequential roles increasingly sit an ocean away.

London's Hub Status Faces a Credibility Test

This matters directly for the UK government's own AI strategy, including its ongoing push to train and reskill the domestic workforce for an AI-driven economy. Programmes built around the promise of a thriving domestic AI industry rely on organisations like DeepMind functioning as genuine centres of gravity, not satellite offices executing decisions made elsewhere. A handful of consequences are worth watching closely:

1. Retention risk: Ambitious UK-based researchers may increasingly view London roles as a stepping stone to California rather than a destination in itself
2. Investment signalling: Venture capital and government funding decisions often track where decision-making power actually sits, not just where offices are located
3. Policy leverage: A DeepMind more tightly integrated into Alphabet's Mountain View command structure gives UK regulators and policymakers less direct influence over a lab operating on British soil
4. Talent pipeline knock-on: University partnerships and graduate recruitment pathways built around DeepMind's London prestige lose some of their pulling power if the most senior roles are understood to require relocation

None of this means DeepMind is abandoning London outright — the new office opened this year is real, and Britain retains a deep bench of AI research talent across academia and industry. But the direction of travel, confirmed rather than reversed by this leadership change, is unmistakable. The lab that put British AI on the map now answers, in practice, to a Mountain View chain of command.

The broader lesson for the UK's automation and AI workforce strategy is one about the difference between hosting AI infrastructure and controlling it. Britain can still claim office buildings, research papers, and headline investment figures. What is harder to claim, on this evidence, is the actual decision-making authority over how the next generation of AI systems gets built — and that authority is the thing that determines where the best jobs, the biggest budgets, and the long-term industrial advantage ultimately land.