Barret Zoph, co-founder of AI startup Thinking Machines Lab and recently vice president of research at OpenAI, is joining Google DeepMind as vice president of research, Reuters reported on August 27, citing Zoph's own announcement on LinkedIn that morning. The appointment sends one of the industry's most mobile senior researchers back to the lab where his career started, and it arrives at a moment when DeepMind is navigating its most significant leadership overhaul in years (Demis Hassabis steps down as CEO).
Zoph will begin his role on August 26, 2026, according to Reuters. His return to DeepMind closes an unusual career loop: he spent six years as a research scientist at Google Brain from 2016 to 2022, before that team merged with DeepMind to form Google DeepMind in 2023. He then left for OpenAI, took a brief stint co-founding Thinking Machines Lab earlier this year, made a short return to OpenAI, and is now heading back to Google.
What Zoph brings to DeepMind
Zoph's research background spans some of the most impactful work in modern machine learning infrastructure. At Google Brain, he contributed to foundational research on neural architecture search, efficient model training, and distributed systems for large-scale machine learning. Those areas became increasingly important as the field shifted toward trillion-parameter models that require massive compute coordination.
At OpenAI, he held the same vice president of research title, overseeing research teams working on the models that power ChatGPT and the GPT series. His presence at both companies gives him a rare dual perspective on how two different organizations approach frontier model development, safety, and the tension between open research and commercial product timelines.
The Reuters report does not specify Zoph's exact research remit at DeepMind, his reporting line, or which teams he will manage. Koray Kavukcuoglu, who took over as DeepMind's chief executive officer this month after Demis Hassabis stepped back from day-to-day operations to become chairman, will presumably define Zoph's role in the broader organizational structure.
A company in upheaval
Zoph's arrival fits into a wider pattern of turbulence at Google DeepMind. Earlier this month, the lab lost several of its most senior figures. Jeff Dean, the company's long-time chief scientist and a 27-year Google veteran, departed alongside senior fellow Sanjay Ghemawat and researchers Oriol Vinyals and Quoc Le, who left to launch a startup called Discovery Loop. The exits occurred on the same day that Demis Hassabis announced he would step down as DeepMind's CEO.
This summer's departures were part of an ongoing trend. In June, Gemini co-lead Noam Shazeer left for OpenAI, and John Jumper, who shared the 2024 Nobel Prize in Chemistry for AlphaFold, left for Anthropic. AlphaFold's original 29 named authors have seen 13 depart since the paper was published. David Silver, the reinforcement-learning pioneer behind AlphaGo and AlphaZero, left earlier this year to found Ineffable Intelligence, now valued at $5.1 billion after a $1.1 billion seed round.
The talent drain has been quantified by Zeki Data, a UK-based data intelligence firm that released a detailed analysis in Fortune on August 27. The study tracked 20,900 research and advanced-engineering hires across ten companies and found that DeepMind's share of research and engineering talent in Europe, the Middle East, and Africa fell from 49 percent in 2022 to 2023 down to just 18.6 percent in 2025 to 2026. That is the sharpest market-share drop Zeki recorded for any major AI lab.
The arrivals-to-departures ratio at DeepMind fell from roughly 12 to 1 in the second quarter of 2023 to about 2 to 1 by the third quarter of 2026 (detailed analysis from Fortune). For comparison, Anthropic's ratio stood at 22 to 1, OpenAI's at 5.7 to 1, and Meta's at 3 to 1.
What this means for the AI talent war
Zoph's move is significant not because it is unexpected. The revolving door between OpenAI, DeepMind, Anthropic, and emerging startups has been the defining feature of the AI talent market for years. It matters because of what it signals about where DeepMind stands relative to its competitors.
Google's advantage remains its resources. DeepMind still has access to Google's compute infrastructure, Alphabet's cash reserves, and the institutional knowledge that comes from being embedded in the world's largest technology company. But those advantages mean less in a market where the best researchers can choose almost anywhere, and the pay packages at pre-IPO companies like Anthropic and OpenAI can include equity that turns early employees into billionaires.
The Zeki analysis identified a specific weakness: DeepMind lost more LLM and multimodal-systems specialists than it hired over the past 12 months. That is the core capability that matters most for the current frontier model race. Meanwhile, DeepMind is gaining ground in robotics and embodied AI, reflecting where the company is trying to build new strengths.
Why Google hired Zoph now
Hiring a senior researcher with Zoph's profile suggests Google is trying to stabilize DeepMind's research direction after months of turbulence. Someone who understands both OpenAI's product-driven approach and Google's research traditions can bridge the gap between the lab's academic heritage and the commercial pressures that have grown as Gemini has become Alphabet's flagship AI product.
It is also worth noting that Zoph's brief time at Thinking Machines Lab did not produce a public product or model before he left. His move back to a well-resourced lab may reflect the difficulty of sustaining frontier research from a startup without the compute and data access that Google and OpenAI control, a pattern we saw with Google's own chip investments.
What to watch next
Zoph's specific research focus at DeepMind has not been announced. If he takes on responsibility for large language model training, the hiring could signal that Google is preparing to make a significant investment in its next-generation Gemini models. If his role leans toward infrastructure or efficient training, it may reflect the growing importance of cutting compute costs as model scale continues to increase.
The broader question is whether Zoph's arrival, on its own, can reverse DeepMind's talent attrition. The Zeki data shows the problem extends beyond any single hire. Competitors like Anthropic have maintained a 22-to-1 arrival ratio, and Google's own internal restructuring, including tighter publication rules introduced in April 2025, may be pushing researchers away from the open-ended work that originally attracted them to the lab.
Zoph's career path through Google Brain, OpenAI, and Thinking Machines Lab gives him a perspective that few people in the industry possess. How he applies that perspective at DeepMind could shape the lab's trajectory for years to come.