Noam Shazeer, Co-Author of the Transformer Paper, Leaves Google for OpenAI

When the history of the modern AI era is eventually written, Noam Shazeer’s name will appear on almost every significant chapter. On June 18, 2026, he added another one: Shazeer announced he is leaving Google DeepMind — where he co-led the Gemini program — to join OpenAI as Lead for Architecture Research. Industry observers are calling it the most consequential individual talent move in AI since Andrej Karpathy joined Anthropic.

The Man Behind the Transformer

In 2017, Shazeer was one of eight co-authors on “Attention Is All You Need,” the paper published by researchers at Google Brain that introduced the Transformer architecture. That architecture is the technical substrate beneath every major language model in existence today: GPT-4, Claude, Gemini, Llama, Mistral, and thousands of others. Describing him as influential understates the case; in a real sense, Shazeer helped invent the paradigm that every AI company is now racing to exploit.

After nearly two decades at Google, he left in 2021 to co-found Character.AI, a consumer chatbot company that attracted a devoted following for its highly customizable, persona-driven AI characters. The startup raised hundreds of millions and demonstrated that there was genuine mass-market appetite for conversational AI — well before ChatGPT made the case globally.

Google’s $2.7 Billion Bet That Didn’t Hold

In August 2024, Google executed what many analysts considered a masterstroke: rather than outright acquiring Character.AI, it struck a licensing deal for the company’s technology and, crucially, brought both Shazeer and fellow co-founder Daniel De Freitas back to Google. The reported cost was approximately $2.7 billion — an extraordinary sum to secure the return of a single researcher and his colleague. Shazeer was given the title of Vice President of Engineering and installed as co-lead of the Gemini effort, Google’s flagship AI model family.

The bet held for less than 22 months.

“I’m incredibly excited to announce that I’m joining OpenAI,” Shazeer wrote in his announcement. He cited the scale of OpenAI’s deployment and the opportunity to work on fundamental architecture questions at the frontier of what current models can do.

Sam Altman’s Long-Held Ambition

OpenAI’s CEO Sam Altman made no effort to conceal his excitement. “Noam is one of the people I have most wanted to work with since the very beginning of OpenAI,” Altman wrote on X. “Only took 10 years.” The comment hints that Altman had attempted to recruit Shazeer before — a narrative that would make sense given how thoroughly Shazeer’s work underpins what OpenAI does.

At OpenAI, Shazeer will take on the title of Lead for Architecture Research, a role that places him at the center of the company’s long-term model design decisions. OpenAI has been aggressively recruiting senior researchers as it prepares for a September 2026 IPO that could value the company at over $850 billion. Shazeer is not merely a high-profile hire; his expertise in scaling laws, mixture-of-experts architectures, and efficient attention mechanisms is directly applicable to the next generation of frontier models.

What Google Loses

The departure is a significant blow to Google, and not just symbolically. Shazeer was deeply involved in the architectural decisions underpinning the Gemini 3 series, which has delivered Google its strongest AI model performance to date. Gemini 3.5 Flash is now Google’s highest-traffic model, running as the default in Google Search’s AI Mode and the Gemini app. Gemini 3 Pro recently achieved the highest score yet recorded on the General 365 reasoning benchmark at 62.8% — though even that fell well short of a passing grade, underscoring how far frontier models still have to go.

Google now faces the uncomfortable dynamic of competing against a rival that employs the person who helped architect the models it is competing with. While institutional knowledge, compute infrastructure, and research teams matter far more than any single individual, the psychological and signaling impact of losing Shazeer to OpenAI specifically — rather than to a startup or an academic institution — is difficult to overstate.

A Wider Talent War

The Shazeer move is the latest in an accelerating pattern of elite AI researchers moving between the handful of organizations at the frontier. The talent pool for researchers who deeply understand model architecture at the level needed to push the frontier forward is genuinely small — perhaps a few hundred people globally — and every major lab is competing intensely for it.

Google has responded to past departures by dramatically increasing researcher compensation and expanding its research agenda. The company is unlikely to stand still here. But it is also facing competition not just from OpenAI but from Anthropic, xAI, Meta, and a growing number of well-funded startups that can offer both competitive salaries and the excitement of greenfield research environments.

For OpenAI, the hire sends a clear message ahead of its IPO: the company is not resting on its current architectural approaches. With Shazeer leading architecture research, OpenAI is signaling that it intends to continue pushing on the fundamental design of AI systems — not just on scale, data, and fine-tuning — at a moment when the field’s next major architectural leap is far from obvious.

The Longer View

There is an irony embedded in this story. The Transformer architecture that Shazeer helped create at Google in 2017 is precisely what made OpenAI’s rise possible. GPT-1, GPT-2, GPT-3, and every subsequent OpenAI model depend on the architectural insights in “Attention Is All You Need.” Now, nearly a decade later, Shazeer is joining the company that built its empire on his paper’s foundations.

Whether that irony will eventually resolve into something transformative — a post-Transformer architecture, perhaps, or a major efficiency breakthrough — is a question Shazeer’s work at OpenAI may eventually answer.