The co author of the 2017 paper that introduced the transformer architecture behind ChatGPT, Claude, and other modern AI systems is leaving Google for OpenAI, two years after a $2.7B deal that returned him to close a reasoning model gap Google has not closed.
On Wednesday, Noam Shazeer announced he is leaving Google for OpenAI. In a post on X reported earlier by the-decoder, an AI industry trade publication, Shazeer wrote: "It was a difficult decision to move on." The departure closes a loop on a $2.7 billion deal Google struck in 2024 to bring Shazeer back, a deal meant to close a reasoning-model gap Google still has not closed.
Shazeer is one of eight co-authors of "Attention Is All You Need," the 2017 Google research paper that introduced the transformer architecture. That design, which processes text by weighing the relationships between words in parallel rather than one at a time, became the foundation for the large language models behind OpenAI's ChatGPT, Anthropic's Claude, Google's own Gemini, and nearly every other consumer-facing AI system. Shazeer first joined Google in 2000 and worked on early search improvements, including the spell checker, before rising to Vice President of Engineering.
In 2021, he left Google to co-found Character.AI, a consumer chatbot startup. Three years later, in a deal reported at $2.7 billion, Google brought Shazeer and Character.AI co-founder Daniel De Freitas back to Mountain View along with parts of Character.AI's research team. The stated purpose was specific: close a gap on reasoning models, the AI systems designed to work through multi-step problems rather than just produce fluent text. At the time, Google was widely seen as trailing OpenAI and Anthropic on that category.
Two years on, that gap remains, and Shazeer is joining OpenAI, the company Google hired him to help catch. The-decoder, the AI trade publication that first reported the move on Wednesday, frames it as the biggest AI talent shift of 2026, alongside researcher Andrej Karpathy's reported move to Anthropic. The framing also amounts to a verdict on the strategy behind the 2024 deal: when a company spends billions to reacquire a researcher and the researcher walks out the door two years later, the question is not whether the talent move failed but whether talent acquisitions can solve structural model gaps at all.
Neither Google nor OpenAI has commented on the move. Shazeer's X post gives no reason beyond the "difficult decision" line, and the-decoder's reporting does not name a specific one. The public record is short on the internal conversations that produced the exit and on what Shazeer will work on at OpenAI. What it does say is that the researcher Google paid to close its reasoning gap is now at the company currently setting the pace on it.
The next test will be whether Shazeer's second departure produces a different outcome than his first. He left Google once for a startup, came back in a multibillion-dollar deal, and is now leaving for a direct competitor. If Google's reasoning models improve in the quarters ahead, the 2024 reverse-acquihire will be remembered as a costly delay. If they do not, it will read as an admission that the gap cannot be closed by buying back the people who were there when it opened.