Google AI Power Shifts from London to Silicon Valley: Who’s Now Steering Gemini

1. The Quiet Power Broker Behind Gemini

DeepMind CTO Koray Kavukcuoglu, a veteran of the lab for 13 years, has taken full charge of daily operations and now reports directly to CEO Sundar Pichai, inheriting absolute control over the Gemini model and Google’s AI destiny.

Kavukcuoglu helped build DeepMind’s deep‑learning team and led the development of WaveNet and the Deep Q‑Network (DQN). A Turkish‑born engineer who studied under Turing‑Award winner Yann LeCun at NYU, LeCun praised him as a top‑tier engineer with rare research‑management talent.

After relocating to Google’s Mountain View headquarters last year, the focus of Gemini’s management quietly shifted, and Kavukcuoglu now reports directly to Pichai, shortening the reporting line and effectively elevating his authority within Google’s tightly‑controlled bureaucracy.

2. Why Hassabis Stepped Back: Two Complementary Narratives

Perspective 1 – The Scientist’s Fulfilment (official and moderate view)

According to Semafor, Hassabis had already been withdrawing from day‑to‑day administration, handing full responsibility for consumer‑facing AI products and Gemini to Kavukcuoglu. As a pure scientist, Hassabis never loved corporate management; his passion lies in pursuing artificial general intelligence (AGI) and using his spin‑out Isomorphic Labs to tackle human diseases. Stepping down freed him to focus on scientific exploration.

Perspective 2 – Power Struggle and Decline (Bloomberg and critical view)

Bloomberg reporter Parny Olsen and other observers argue that after Gemini’s early‑2024 PR crisis, Hassabis sought and briefly gained greater control, keeping R&D authority in London. However, internal sources say the clash between a “London independent kingdom” and California’s commercial drive became irreconcilable, and Kavukcuoglu, backed by the California headquarters, ultimately won the strategic battle.

Regardless of the true motive, the outcome is clear: the London‑based R&D strategy of the Hassabis era has been fully taken over and re‑engineered in Silicon Valley.

3. The Hard Challenges Kavukcuoglu Inherits

Google’s current AI agenda focuses on “extreme execution” rather than pure invention. Although Google invented the Transformer architecture that underpins the generative‑AI era, competitors such as OpenAI and Anthropic have captured much of the commercial upside. The company now faces three major hurdles:

Delivery delays and performance gaps : The flagship Gemini super‑computing cluster has not been completed as scheduled, and Gemini’s performance in the high‑resource “code generation” domain still lags behind rivals.

Talent attrition : Key figures from the Transformer paper, including Noam Shazeer, and Nobel‑winning scientist John Jumper have been poached by OpenAI, Anthropic and other rivals.

Heavy financial pressure : Google expects capital expenditures of up to $205 billion this year, pushing free cash flow into negative territory for the first time since its 2004 IPO; the news caused Alphabet’s stock to drop about 4%.

Conclusion – The “California Speed” Imperative

Google’s fundamentals remain massive. In the earnings call, Pichai emphasized that Google is the only company with a fully self‑controlled stack—from TPU chips and infrastructure to base models and applications. Gemini’s monthly active users have already surpassed 950 million.

Google never lacks top talent or compute power; its historic weakness has been “speed.” With the AI power center now fully back in Silicon Valley, Kavukcuoglu’s mission is to apply engineering rigor and California‑level velocity to revive the sluggish AI behemoth.