Noam Shazeer posted his departure from Google just after midnight Pacific time on June 18, and by morning the news had reordered every conversation about where the best AI talent is heading. Shazeer, Google’s vice president of engineering and a co-lead of the Gemini model program, is joining OpenAI. He is also one of the eight co-authors of the 2017 paper “Attention Is All You Need,” the work that gave the industry the Transformer architecture and, in practical terms, made every frontier model possible.
The timing puts a dollar figure on Google’s loss. In August 2024, CNBC and Axios reported Google paid roughly $2.7 billion to bring Shazeer back as part of its deal with Character.AI, the consumer chatbot company he co-founded after leaving Google in 2021. He has now left again, under two years into that arrangement, and he did not need a two-year runway to make the decision.
Google has not commented publicly. OpenAI CEO Sam Altman welcomed Shazeer on X, describing him as one of the people he had most wanted to work with since OpenAI’s earliest days. That framing is notable: Altman is not treating this as a late-stage hire filling a gap. He is signaling that Shazeer fits into whatever OpenAI is building at the architecture level.
The departure is not purely a research story. Shazeer’s fingerprints are on foundational Google infrastructure predating the Transformer: he materially improved Google’s search spelling correction and wrote the core algorithm behind AdSense. He is an engineer who builds things that compound. Losing him to the primary competitor is qualitatively different from losing a team lead or a product manager.
What makes this particularly pointed is the IPO context. OpenAI has been preparing for a public offering, and talent moves at this level function as proxy signals for investor confidence. When the co-author of the paper that defines the field joins your pre-IPO company from your best-funded competitor, that is the kind of headline that lands in an S-1 roadshow deck whether it appears there or not.
Google has spent three years and billions of dollars trying to hold onto the researchers it lost to the wave of lab spinouts from 2020 to 2022. The Character.AI arrangement was the most visible example of that effort. Shazeer returning and then leaving for OpenAI specifically makes the strategy look expensive rather than effective.
For OpenAI’s technical leadership, the addition matters in a concrete way. Shazeer has been hands-on with model architecture at the scale that produces frontier results. His return to active lab work, now inside the organization building GPT-5 successors and whatever comes after, shifts the composition of that team.
The talent balance between Google and OpenAI has been the subject of speculation since Google’s 2024 model releases improved its competitive standing. This move suggests the balance shifted back before that narrative fully settled.
Any team currently planning a 2026 model vendor commitment should treat Shazeer’s arrival at OpenAI as relevant input: the architecture-level talent concentration at that lab just increased, and the technical leadership gap with Google widened on the day one of Google’s most consequential engineers walked out.
Reported by CNBC and Axios on June 18, 2026.
