Google's AI talent exodus accelerates as cloud revenue soars

Google is experiencing a significant brain drain in artificial intelligence even as its cloud business posts record growth, exposing fundamental tensions about where the $4 trillion company should focus its AI investments.

Jeff Dean, Google's chief scientist for 27 years, announced his departure this week to launch Discovery Loop, a Google-backed public benefit corporation focused on automating machine learning and scientific discovery. He's joined by prominent Google researchers Sanjay Ghemawat, Oriol Vinyals, and Quoc Le. Simultaneously, Demis Hassabis is stepping down as CEO of Google DeepMind to become chairman, with technology chief Koray Kavukcuoglu assuming daily management.

The leadership changes come as Google Cloud reported 82% revenue growth, with CEO Sundar Pichai noting that 90% of Fortune 100 companies now use Gemini Enterprise. This commercial success stands in stark contrast to the research organization's struggle to retain top talent.

Why it matters

The departures signal a strategic crossroads for Google: investing billions in frontier AI research offers uncertain returns, while selling AI infrastructure and services to enterprises generates immediate, substantial revenue. How Google resolves this tension will determine whether it leads the next wave of AI breakthroughs or becomes primarily an infrastructure provider to competitors.

Compute allocation fuels internal friction

Access to tensor processing units (TPUs) has become a major source of frustration for Google researchers, according to people familiar with internal dynamics. When Google Cloud sells TPU capacity to external customers—including direct competitor Anthropic—it reduces availability for internal research teams pursuing ambitious projects.

Google models demand across research, product serving, and cloud customers years in advance, with capacity shifting based on changing priorities. Pichai has stated on recent earnings calls that DeepMind's compute needs remain the "first priority," though researchers reportedly see the situation differently when large infrastructure commitments go to competing labs.

Dan Niles of Niles Investment Management, a Google shareholder, acknowledged the inherent tension: "Google has all of these other businesses, and they've got to figure out who they're going to give some of these resources to. Somebody's always going to be unhappy in that situation."

The transformer brain drain continues

All eight authors of Google's landmark 2017 paper "Attention Is All You Need"—which established the transformer architecture underlying generative AI—have now left the company. Noam Shazeer departed for OpenAI in June, less than two years after Google paid nearly $3 billion to bring him back. Nobel laureate John Jumper left DeepMind for Anthropic shortly before.

Gil Luria, an analyst at D.A. Davidson, identified a clear pattern: "They're not interested in commercializing AI. They're interested in being part of history, and so they look at Anthropic, OpenAI or another startup as being the place where they can pursue history."

Google's bureaucracy compounds the problem, with multiple approval layers required to move research into products—making nimbler competitors more attractive to scientists focused on breakthroughs rather than balance sheets.

The "good enough" model question

Tomasz Tunguz of Theory Ventures argues that cutting-edge models aren't necessary for most enterprise applications: "Many of the models that are reasonable are good enough." Niles echoed this view, noting that "the models are good enough for 90% of what needs to get done."

This reality may favor Google's infrastructure business over frontier research investments. But for researchers pursuing transformer-scale breakthroughs, being "good enough" isn't sufficient—creating an irreconcilable gap between commercial pragmatism and scientific ambition.

These details were first reported by CNBC.