Google's new AI chief inherits delayed Gemini 3.5 Pro, talent exodus

Koray Kavukcuoglu takes over Google DeepMind as the flagship Gemini 3.5 Pro slips and four senior researchers depart in the same week.

Google's new DeepMind chief Koray Kavukcuoglu inherits a division racing to close a coding gap with OpenAI and Anthropic after delaying Gemini 3.5 Pro and losing four senior researchers.

"Google has fallen behind Anthropic and OpenAI by more than one position in coding capability," Malik Ahmed Khan, senior equity analyst at Morningstar, said.

Kavukcuoglu, who joined DeepMind in 2012 and served as chief technology officer, now reports directly to Alphabet CEO Sundar Pichai. The promotion follows the departure of Jeff Dean, Sanjay Ghemawat, Oriol Vinyals and Quoc Le, who left to cofound Discovery Loop, an AI research startup backed by Alphabet. Alphabet shares fell about 4 percent, erasing roughly $185 billion in market value, after the moves were announced.

The transition lands as Alphabet raises 2026 capital expenditure to $195 billion to $205 billion and Google Cloud revenue accelerates to 82 percent growth. Kavukcuoglu's mandate is clear: deliver Gemini 3.5 Pro and prove Google can still compete at the frontier.

Gemini 3.5 Pro delay exposes coding weakness

The flagship model's release has been pushed back because it falls short in programming and other critical areas, according to Bloomberg. Google launched Gemini 3 in November 2025 to strong reviews, but the competitive picture shifted through 2026 as OpenAI and Anthropic accelerated their release cadence.

Khan said Google's focus on search monetization and multimodal capabilities diverted resources from building a killer coding application, the category where Anthropic's Claude and OpenAI's models have established clear leads. Google has formed an internal team called "Code Strike" to address the gap, though the company disputes the characterization that it has drifted from its AGI roadmap. Google maintains that the path to AGI spans agents, robotics and world models, not just language-model benchmarks.

The delay carries market consequences. Alphabet shares fell in July partly on the Gemini 3.5 Pro concerns, and the company has not given a release date, saying only that the model is undergoing partner testing. Meanwhile, Google has shipped three lighter-weight Gemini models to keep developers engaged while the flagship waits.

From research to execution

Constellation Research chief analyst Ray Wang said Kavukcuoglu's elevation signals a shift from research-driven development to product execution, which could mean faster release cycles and a more focused product roadmap. Demis Hassabis, who led DeepMind since Google acquired the lab in 2014, becomes chair of the unit and Alphabet's chief scientist while continuing to run Isomorphic Labs.

The handover was not abrupt. Kavukcuoglu had been absorbing model development and major Gemini product responsibilities from Hassabis over the past year, according to people familiar with the matter. His 13-year tenure at DeepMind includes founding the deep learning team and working on WaveNet and DQN breakthroughs.

Futurum Group AI lead Nick Patience said the first priority is delivering Gemini 3.5 Pro on a predictable schedule and rebuilding the programming and pre-training teams. Sebastian Mallaby, author of "The Infinity Machine," cautioned that the departures should not be read as a systemic crisis, noting the Gemini team numbers in the thousands.

The stakes are measurable. Alphabet's second-quarter revenue rose 24 percent year over year to $119.8 billion, its 12th consecutive quarter of double-digit growth, with operating margin at 34 percent. Google Cloud revenue reached $24.8 billion, with growth accelerating to 82 percent. But the company's free cash flow turned negative last quarter for the first time on record as AI infrastructure spending surged. At roughly $362 per share, Alphabet trades at about 27 times forward earnings, a premium that assumes Gemini stays competitive. The Gemini app has crossed 950 million monthly users, giving Google a distribution advantage even as its frontier model lags.

This article is for informational purposes only and does not constitute investment advice.