Behind the huge Google (GOOGL.US) personnel shock: businessmen take charge of DeepMind, AI competition crushes scientific research idealism

The Zhitong Finance App learned that Google (GOOGL.US) co-founder Sergei Brin has continued to put pressure on the core AI team in recent months to fully focus on the company's flagship model Gemini in order to catch up with competitors. In this context, parent company Alphabet announced a large-scale restructuring of Google's DeepMind leadership structure on August 5. According to some analysts, this adjustment marks the further weakening of DeepMind's sovereignty since it was acquired in 2014, and commercial urgency is completely overwhelming the research-oriented culture of the London AI laboratory for more than 10 years.

People familiar with the matter revealed that after Anthropic first released a preview of its powerful Claude Mythos model, Brin addressed hundreds of employees at an all-staff conference in April, urging the DeepMind team to speed up the pace.

Google's Gemini model briefly surpassed its rivals in November of last year, but subsequent updates to Anthropic and OpenAI once again put Google in a catch-up mode. According to two people familiar with the matter, internal tests showed that the performance of the new flagship version of Gemini still lags behind the competition in key areas such as programming, which caused Google to delay the launch by two months.

DeepMind's “change of command”: Hassabis gives way, commercials take power

On August 5, local time, the tech giant, headquartered in Mountain View, California, announced a comprehensive leadership change for Google DeepMind. DeepMind CEO Demis Hassabis stepped down and was replaced by his deputy Korai Kavukuoglu. Khasavis was transferred to the position of chairman, and management responsibilities were reduced accordingly. At the same time, the two original technical leaders of the Gemini project also left their jobs and co-founded a startup.

Under the leadership of Hassabis, Google DeepMind has carried out a wider range of AI research projects than its competitors, while also constructing a Gemini model that is critical to commercial interests. However, there are reports that in the past few years, Hassabis has resisted some efforts that could bring new revenue to Alphabet or help it gain a better position in the AI competition.

Although some industry insiders question whether Google is withdrawing from the model competition, seven people familiar with Gemini's development work said that Alphabet's commercialization strategy and last week's structural restructuring are all closely related to the company's intention to pursue a leading position in the model.

This adjustment marks a further weakening of DeepMind's autonomy. Since acquiring this London lab in 2014, Google has been weakening DeepMind's autonomy and continuing to push its AI technology to empower more software products and generate more revenue. Two people familiar with the matter revealed that Alphabet's current restructuring also included transferring some teams from DeepMind and integrating them into Google's corporate division.

On the day the restructuring was announced, Alphabet's stock price fell 4%. At the departmental general meeting last Thursday, Hassabis and Kavukuoglu tried to assuage employees' concerns about the future of the AI lab — most employees only learned about the change when the news became public.

People familiar with the matter revealed that the two leaders said the lab's daily work will remain the same. People familiar with the matter said that Kawukuolu will continue to be responsible for Gemin, who took over the supervisory responsibility of model development more than a year ago. Hassabis, on the other hand, delivered a speech on how AI can drive scientific progress and change society. These long-term topics will form the new focus of his work as chairman.

However, the General Conference also revealed new concerns that DeepMind's autonomy is being encroached upon by Google. According to people familiar with the matter, Khasavis and Kavukuolu told employees that several non-technical teams are being removed from DeepMind and incorporated into Google's corporate reporting structure. For an organization that once encouraged employees to work on ancillary projects, the message of regaining the dominance of the model was just as clear.

Bollin puts pressure behind the scenes: betting on “recursive self-evolution”

Google's personnel changes come at a time when Brin is trying to boost employee morale. He also took similar action after OpenAI shocked Google by announcing ChatGPT in 2022.

Despite not taking an executive role, 52-year-old Brin influenced model training strategies. People familiar with the matter said that over the past few months, he has continued to stress to core employees that Google needs to catch up with the cutting edge.

People familiar with the matter also said that Brin used his implicit power as Google's co-founder to lean resources towards specific areas, such as “recursive self-improvement” — the concept that technology can improve itself without human intervention.

Two people familiar with the matter said that soon after Gemini was first launched in 2023, Hassabis reduced its participation in the project and handed over related responsibilities to a deputy.

Although Google caught up at the cutting edge with the release of Gemini 3 in November last year, it soon fell behind again. Four people familiar with the matter attributed Google's backwardness to deep-seated structural challenges. People familiar with the matter said that, for example, computing power limitations and differences among Gemini's many project leaders have led to insufficient resources in key areas such as programming. Google's bureaucracy has also caused its models to be released more slowly than more flexible competitors.

Google declined to comment on these claims. However, CEO Pichai told Wall Street analysts last month that the company will continue to expand the scale of infrastructure to ease model training and TPU computing power bottlenecks required for consumer-grade and enterprise products.

Power is concentrated in Kawukuolu, and the center of power shifts from London to California

In 2025, Google CEO Sundar Pichai appointed Kawukuoglu, who was the chief technology officer of DeepMind at the time, as Google's chief AI architect. The task was to integrate Gemini into the company's entire product line. People familiar with the matter said that over the next few months, the influence of some DeepMind leaders gradually waned, and Kawukuolu became the central figure in shaping the direction of Gemini. One of the people familiar with the matter said that Kavukuolu was supported by Brin.

People familiar with the matter revealed that as Hassabis reduced its participation in Gemini and Cavucuoglu moved to Google's headquarters, the influence of some London headquarters leaders on the direction of Gemini's technology has clearly weakened.

Kavukuoglu is also gradually building a reputation for focusing more on commercialization in other divisions of Google. According to a person directly familiar with the matter, even before his relocation, Kavukuolu, rather than Khasabis, was already the main interface between DeepMind and Google Cloud (the company's main AI revenue engine).

A Google spokesperson said last week that Kawukuolu will now have the final say in DeepMind's major decisions.

In addition to Kawukuolu, some of the most notable leaders include Gemini's two original technical co-heads Jeff Dean and Oriol Vinyals, who have left their posts in this restructuring. Three people familiar with the matter said that the teams that reported to some of the leaders had rare autonomy to conduct research not directly related to Gemini's training.

After they left their jobs, Google did not immediately announce the arrangement plans for these teams or projects, which made some employees uneasy and worried that the lab's final barrier against commercial interests would further collapse.