Alphabet Inc. is undergoing what may be the most profound artificial intelligence leadership restructuring in its nearly 30-year history. The persistent underperformance of its flagship Gemini models, coupled with a mass exodus of top-tier talent, has forced the company to make a long-overdue choice between being a "great research institution" and a "ruthless model product company." This decision is fundamentally reshaping the company's AI strategy.
On August 5, 2026, Alphabet announced that Demis Hassabis would step down as CEO of Google DeepMind to become the Chief Scientist of Alphabet and Chairman of Google DeepMind. Just minutes before the announcement, Jeff Dean, a 27-year veteran of Google, along with Oriol Vinyals, Sanjay Ghemawat, and Quoc Le, jointly announced their departure to found a new AI research company called Discovery Loop. Simultaneously, Koray Kavukcuoglu was appointed to take over the daily operations of Google DeepMind, overseeing model development, frontier research, the Gemini application, and the developer ecosystem, reporting directly to CEO Sundar Pichai. Following the news, Alphabet's market capitalization evaporated by over $180 billion in a single day.
This is the final institutionalization of a power shift that has been brewing for some time, not a sudden personnel earthquake. The anticipated summer 2026 launch of Gemini 3.5 Pro has been repeatedly delayed, while competitors like OpenAI and Anthropic continue to lead in key areas such as coding agents and long-horizon task execution. Less than two months ago, Noam Shazeer, who helped design the Gemini technical architecture, defected to OpenAI. John Jumper, who shared the Nobel Prize with Hassabis, joined Anthropic. The collective departure of Google's "golden generation" of AI talent marks the end of an era.
Delayed Flagship Model and Accelerating Talent Drain
The delay of Gemini 3.5 Pro is the most direct trigger for this management upheaval. Reports indicate the model was planned for a summer launch, but as of late July, it was still in partner testing. Google has only stated it will be widely released "when it's ready." Meanwhile, internal teams have already begun training on Gemini 4.
The talent drain is accelerating. Oriol Vinyals, who left with Jeff Dean, was a key general in Google's fight against Anthropic and OpenAI. Prior to that, Noam Shazeer joined OpenAI, and John Jumper moved to Anthropic. Jeremy Nixon, who previously worked on research at Google Brain, described Jeff Dean's departure as "almost unimaginable" and warned: "This may be Google's first true moment of crisis in a long time. No one can match the speed of Google's talent loss."
In July, Google released several small AI models, including one focused on cybersecurity. While these models performed reasonably well on benchmarks for coding and financial tasks, they failed to surpass competitors. At the same time, Google maintains massive capital expenditure. The company expects to spend between $195 billion and $205 billion on data centers and AI infrastructure this year, more than double the $85 billion spent last year.
The Culmination of Three Years of Organizational Conflict
The roots of this restructuring can be traced back to the April 2023 decision to merge Google Brain with DeepMind. Initially, the merger was widely interpreted as a strategic move to concentrate top AI talent and accelerate research. However, Google confused two fundamentally different questions: whether talent could be centralized, and what that centralized talent should do.
From its inception, Google Brain was deeply integrated with Google's search, advertising, and infrastructure systems, giving it an inherent engineering and delivery focus. DeepMind, on the other hand, has always positioned itself as a project akin to the "Apollo Program"—first solving intelligence itself, then using that intelligence to solve all problems. Hassabis's core interest has never been to boost the next quarter's token sales, but rather AGI, world models, and scientific discovery. AlphaFold's Nobel Prize-winning achievement gave this long-termism undeniable legitimacy.
For the past three years, Google has tried to have it both ways: demanding that DeepMind serve the commercial product needs of a billion-user scale, while being unwilling to give up DeepMind's aura as a premier research sanctuary.
According to industry sources close to Google's AI leadership, during the sprint to launch Gemini 2.5, teams responsible for data pipelines and RLHF (Reinforcement Learning from Human Feedback) were working around the clock at a startup pace. Yet, parts of the DeepMind team that retained a research institute culture continued to operate in a relaxed manner, "exploring how to combine two directions."
Hassabis nominally held overall leadership but his focus remained on the AGI and scientific mission. Jeff Dean retained the title of Chief Scientist, but the Google Brain roadmap had gradually lost organizational dominance after the merger. Koray Kavukcuoglu was quietly being given more and more product integration responsibilities, meaning the power structure had already shifted while the titles still reflected the old order.
Power Shift Completed a Year Ago
In fact, this personnel adjustment is merely the formal institutionalization of a power transition that has been unfolding for the past year.
In June 2025, Google created the specific position of Chief AI Architect for Kavukcuoglu, promoted him to Senior Vice President, and changed his reporting line to report directly to Pichai. In an internal memo, Pichai stated that Kavukcuoglu's new mission was to get Google's most advanced models into products faster, achieving smoother integration, quicker iteration, and higher efficiency.
Google's official description of Kavukcuoglu's role also shifted subtly at that time—no longer just the technical lead for DeepMind, but explicitly leading the development of Gemini generative AI models and their large-scale integration into Google products. Kavukcuoglu, an early DeepMind member who helped build the DQN deep reinforcement learning system and studied under AI pioneer Yann LeCun, has a solid research background. Unlike Hassabis, his discussions about technical roadmaps always ultimately return to specific product delivery and user needs. In Kavukcuoglu's view, the value of native multimodality is not about benchmark scores, but about enabling models to complete tasks that were previously impossible. This pragmatic orientation is exactly what Google needs at this stage. Google did not decide yesterday that Kavukcuoglu was better suited to lead Gemini; it was simply unwilling to publicly contradict Hassabis's leadership arrangement.
Google's 'TBD Strategy' and the Age of Industrialized AI
Over the past three years, Google essentially pursued a "TBD (To Be Determined) strategy." It was unwilling to make a clear choice between being a "great research institution" and a "ruthless model product company," hoping for the best of both worlds, but ultimately failing to excel at either.
This logic could hold in the early stages of AI development, when a single breakthrough in underlying technology could create a generational gap, making a "genius moment" worth waiting for. However, in the current phase, model competition increasingly relies on high-density experimentation, data engineering, inference optimization, cost control, and continuous delivery, rather than waiting for the next "AlphaGo moment." Not falling behind every three months has become a more urgent survival imperative than "changing science in the next decade."
Following this restructuring, Google has established a new division of labor: Kavukcuoglu leads high-intensity training and iteration; Pichai is responsible for embedding models into every Google business; Hassabis turns his focus to AGI, scientific discovery, and AI-driven drug research at Isomorphic Labs; and Jeff Dean, along with his longtime colleagues, starts a new venture, with Google providing investment and cloud support, waiting for future technological breakthroughs to be absorbed into the ecosystem.
This is a realistic arrangement, a clear admission of failure. The Google that once convinced the world that a large corporation could let a group of brilliant scientists determine the direction of AI is becoming history. Whether Google can rebuild its competitiveness under Kavukcuoglu's logic of industrial execution, or whether it will step into a deeper vortex of talent centrifugal force, the market is seeking the answer. The timing of the Gemini 3.5 Pro launch will be the first major test.
Comments