Google DeepMind's recent leadership shakeup exposes a deeper problem: a two-year power struggle between its London headquarters and Mountain View offices that slowed decision-making and fractured one of AI's most important research labs. When Google merged DeepMind and Google Brain in 2023, the company promised a unified AI operation. Instead, it created competing power centers across two continents, each with its own culture, resources, and competing interests.

What Went Wrong With the DeepMind and Google Brain Merger?

The 2023 merger was supposed to streamline Google's AI research by combining DeepMind, which Google acquired in 2014, with Google Brain, the company's internal AI research team. But the integration masked a fundamental structural problem: Demis Hassabis, DeepMind's founder, ran operations from London, while Jeff Dean, Google's chief scientist and 30-year veteran, remained in Silicon Valley. Both reported directly to CEO Sundar Pichai, creating a dual-leadership structure that looked unified on paper but operated as two separate organizations.

The cultural differences between the two offices became a major source of friction. Former employees described the divide as the "America-fication" of DeepMind, with Mountain View operating under a more aggressive, competitive dynamic where individual researchers claimed credit for projects and pushed harder for resources. London-based researchers, by contrast, worked in a more collaborative environment that valued collective achievement. One longtime London employee explained the contrast: "It's just how people work in the states. It's the loudest person in the room. Everybody seems to claim the project is theirs, they're leading it now. There's just a sharper edge".

How Did Information Silos and Resource Competition Damage the Organization?

Beyond cultural differences, the merger created practical barriers that undermined trust and collaboration. In some cases, London-based DeepMind researchers could access projects and documents from Mountain View colleagues, but the access didn't work in reverse. This asymmetry bred suspicion and resentment. When projects overlapped between offices, teams were sometimes favored based on their location and reporting structure rather than merit. As one former employee noted, "Things are quite politicized based on who you report to. There's some animosity between the two teams".

Computing resources, one of the most precious assets in AI research, became another flashpoint. Teams in each office competed fiercely for access to the GPUs (graphics processing units) and other hardware needed to train large AI models. The tension came to a head in June 2026 when computing resources dedicated to a Mountain View researcher's project were reassigned to a London-based team, contributing to that researcher's eventual departure.

Steps to Understanding How Organizational Structure Impacts AI Research

• Leadership Geography: When senior leaders operate from different continents and report to the same executive, decision-making becomes slower and more political because there's no single authority center to break ties or set priorities.
• Information Asymmetry: Unequal access to project data and documents between offices creates mistrust and prevents teams from building on each other's work, reducing the efficiency gains a merger is supposed to deliver.
• Resource Allocation: Competing teams in different locations will fight over limited computing power and funding, leading to inefficient use of resources and resentment among researchers who feel their work is deprioritized based on office politics rather than scientific merit.

The structural problems were compounded by sluggish decision-making. With leadership split across two continents and multiple layers of bureaucracy, the organization couldn't move quickly. "They're not able to make decisions," said Ben Pouladian, an AI infrastructure analyst who tracks Google's internal dynamics. "There's a lot of bureaucracy slowing things down".

Why Did Top Researchers Start Leaving?

The organizational dysfunction coincided with a wave of high-profile departures that signaled deeper problems. In June 2026, John Jumper, who won the Nobel Prize alongside Hassabis for their work on AlphaFold, a breakthrough AI system for predicting protein structures, joined competitor Anthropic. Two days earlier, Noem Shazeer, a key researcher and one of the original authors of the transformer architecture that powers modern AI, announced he was leaving for OpenAI. These losses were particularly stinging because they represented some of DeepMind's most accomplished scientists.

The departures raised questions about what was happening behind the scenes. Google had squandered an early advantage in AI research; the company's Google Brain team invented the transformer in 2017, the foundational architecture that made modern generative AI possible. Yet all eight authors of the original transformer paper, including Shazeer, eventually left the company. Though Shazeer was brought back in 2025 after Google essentially acquired him for $2.7 billion, his second departure suggested that internal problems persisted.

What Changes Are Coming to Google DeepMind's Leadership?

On August 6, 2026, Google announced a major leadership restructuring designed to consolidate power and streamline decision-making. Hassabis is stepping back from day-to-day leadership of Google DeepMind to become chairman and will focus more on Isomorphic Labs, a DeepMind spinout focused on AI-driven drug discovery and biotechnology that he has led since 2021. Jeff Dean is departing Google entirely to found Discover Loop, a startup focused on using AI to accelerate scientific research, backed by investors including Khosla Ventures and Radical Ventures.

Koray Kavukcuoglu, a Turkish-born engineer who studied under AI pioneer Yann LeCun at New York University, is taking over as head of Google DeepMind. Kavukcuoglu embodies a bridge between the two offices; he joined DeepMind in London two years after its founding before relocating to Mountain View. He's known for his ability to navigate large organizations and reportedly serves as Sergey Brin's go-to person when he needs something accomplished at Google DeepMind. A Google executive told Forbes that "Koray was already running probably the majority of DeepMind for quite some time," suggesting the transition may be more of a formalization than a dramatic shift.

The restructuring also reflects a broader shift in power toward Mountain View. Koray Kavukcuoglu moved from London to Google headquarters in the past year, and Sebastian Borgeaud, who leads an important coding effort, also relocated from the UK to California. This geographic consolidation suggests Google is betting that centralizing leadership in Silicon Valley will reduce the friction that plagued the merged organization.

Google has publicly stated it was not aware of any rivalries between employees in the two cities and that the merger went smoothly. The company also emphasized its continued commitment to London, where it opened a new office earlier in 2026, and described Google DeepMind as a "global team with a global footprint." However, the departures of Dean and Jumper, combined with the leadership changes, suggest the internal tensions were more significant than the company acknowledged.

The shakeup marks a critical moment for Google's AI ambitions. After losing ground to OpenAI and Anthropic in the generative AI race, the company needs to demonstrate that it can retain top talent and execute at the speed required to compete in a rapidly moving field. Whether consolidating leadership under Kavukcuoglu will solve the underlying organizational problems remains to be seen.