Noam Shazeer, co-creator of the transformer architecture, leaves Google for OpenAI

Noam Shazeer, one of the world's most respected AI engineers, has announced he is leaving Google to join OpenAI. The decision comes as a surprise: Shazeer returned to Google just two years ago as part of a $2.7 billion deal, and since August 2024 had co-led the development of Google's Gemini models as Vice President of Engineering.

A Two-Year Homecoming

Shazeer's history with Google is unusual even by the standards of an industry where star transfers have become routine. He worked at the company from 2000 to 2021, then left to co-found Character.AI with Daniel De Freitas, a startup that lets users chat with virtual versions of pop-culture characters and build their own AI assistants.

In August 2024, Google struck a $2.7 billion deal with Character.AI. Formally, it was a license for the startup's technology rather than an acquisition, allowing Google to sidestep the lengthy merger-review process regulators require. In practice, the deal had one main purpose: bringing Shazeer and his team back to Google. The US Department of Justice is now examining whether this deal structure was designed to circumvent antitrust law.

A Decade of Courting by Altman

Sam Altman had long sought to work with Shazeer. After the departure from Google was announced, the OpenAI chief wrote on X that Shazeer was among the people he had most wanted to work with since OpenAI's earliest days.

Noam is one of the people I most wanted to work with from the very beginning of OpenAI. It only took 10 years. I think it was worth the wait - Sam Altman, CEO of OpenAI

I'm excited to share that I'm joining OpenAI, and I can't wait to work with this exceptional team - Noam Shazeer

In its official comment, Google limited itself to thanking Shazeer for his contribution to the company, without giving reasons for his departure or plans for filling his role on the Gemini team.

The AI Talent War

Shazeer's departure is the latest episode in a months-long battle for top AI researchers among Google, OpenAI, Anthropic and Meta. Companies are offering multimillion-dollar pay and equity packages to lure away the authors of key research papers and the architects of the most important models.

Shazeer co-authored the 2017 paper Attention Is All You Need, which introduced the transformer architecture that today underpins nearly every large language model, from GPT-5.6 to Gemini 3 and Claude Opus 5. Losing one of the creators of that architecture is a reputational blow for Google, especially at a time when Gemini is competing for the top spot against models from OpenAI and Anthropic.

It is the second high-profile departure to a rival in a short period for Google - Nobel laureate John Jumper previously left Google DeepMind for Anthropic. The company is finding it increasingly difficult to convince its own researchers that continuing to work on Gemini offers greater opportunities than rival labs.

What It Means for the Market

For OpenAI, landing Shazeer strengthens its research team at a time when the company is preparing for further product expansion and technological rivalry with Google and Anthropic. The role he will take on at OpenAI has not yet been disclosed.

The case also shows how hard it is to retain key people even after paying billions of dollars for them - contracts and financial packages do not guarantee loyalty when a competitor offers a different team, different computing resources, or simply a different vision for the technology's development. For Polish companies investing in AI teams, it is a signal that retaining talent in this industry is growing more expensive and less predictable, regardless of the sums involved.