Noam Shazeer, Transformer Co-Author, Leaves Google for OpenAI as Lead for Architecture Research

Noam Shazeer, the engineer whose ideas sit at the foundation of nearly every modern large language model, is leaving Google for OpenAI — a defection that lands less than two years after Google paid roughly $2.7 billion to bring him back, and just days after OpenAI quietly set the machinery of an initial public offering in motion.

Shazeer announced his move on Wednesday, June 18, 2026, confirming he will join OpenAI as Lead for Architecture Research. In the role, he will steer the lab's hunt for the next generation of model designs — the kind of foundational work that could determine whether the transformer, the architecture he helped invent, gets succeeded by something faster, cheaper, or smarter. He had been a vice president of engineering at Google and a co-lead of the Gemini models, the very products built to compete with OpenAI's GPT line.

The hire is a marquee win for OpenAI, and chief executive Sam Altman treated it like one. "noam is one of the people I have most wanted to work with since the very beginning of openai," Altman wrote in a post on X. "only took 10 years. i think it will be worth the wait!" The decade-long courtship underscores Shazeer's stature: this is not a routine senior hire but the acquisition of a researcher whose name appears on the papers that define the field.

A Resume That Reads Like the Field's Syllabus

Shazeer's credentials are hard to overstate. He was a co-author of the 2017 paper "Attention Is All You Need," which introduced the transformer architecture now underpinning virtually every major large language model — GPT, Gemini, Claude, and countless open-source descendants. A year earlier, in 2016, he co-authored foundational work on the Sparsely-Gated Mixture-of-Experts layer, a technique that has become central to scaling today's largest and most efficient models. He also developed Multi-Query Attention, an inference optimization that quietly makes modern chatbots cheaper to run.

His career has also become a case study in the bidding war for elite AI talent. Shazeer left Google in 2021 to co-found Character.AI, the consumer chatbot startup. In a 2024 deal valued at roughly $2.7 billion, Google effectively reabsorbed him and part of his team through a licensing-and-hiring arrangement around Character.AI, installing him as a co-lead on Gemini. That return lasted only about 22 months before OpenAI pulled him away again — a churn rate that says as much about the industry's poaching economy as it does about any single researcher.

The timing is conspicuous. OpenAI filed a confidential S-1 with the U.S. Securities and Exchange Commission on June 8, 2026, working with Goldman Sachs and Morgan Stanley on an offering that reporting suggests could value the company at up to $1 trillion. Shazeer was not the only high-profile name to arrive in the same window: Dean Ball, an AI policy scholar who advised the Trump administration's AI agenda, also confirmed he is joining OpenAI, where he will lead a new team called Strategic Futures as Head of Strategic Futures, with a start date of July 6.

Why It Matters

For OpenAI, the message to prospective IPO investors is unmistakable: even amid ferocious competition from Google, Anthropic, and others, the company can still attract the most coveted researchers in the world. Landing the man who co-invented the transformer to lead architecture research is the kind of signal that reassures Wall Street the lab's technical edge is not slipping as it transitions toward public markets.

For Google, the loss cuts deeper than one departure. Shazeer was a co-lead of Gemini, the product line at the center of Alphabet's AI strategy, and his exit revives uncomfortable questions about whether the company can retain the talent it pays so dearly to acquire. The market reaction was telling and volatile. On the day of Shazeer's announcement, Alphabet shares actually ticked up about 1.17%, as investors initially shrugged. The mood soured fast. By Monday, June 22, Alphabet stock fell roughly 7% — its worst day in about a year — after a second landmark researcher, Nobel laureate and AlphaFold2 co-creator John Jumper, left Google DeepMind for Anthropic. The back-to-back exits erased an estimated $250 billion in market value, transforming a single high-profile hire into a narrative about a talent exodus.

That narrative is the real stake here. The transformer era was built by a small cohort of researchers, many of whom passed through Google. Watching them scatter to rivals — and command billion-dollar price tags doing it — suggests the competitive moat in AI is not just compute or data, but the few dozen people who know how to invent the next architecture. Whoever holds them holds the frontier.

What to Watch

The immediate question is what Shazeer actually builds. His mandate — architecture research — points beyond incremental scaling toward whatever might replace or extend the transformer, and any signal that OpenAI is moving away from the design Google still relies on would reverberate across the industry. Watch, too, for Google's response: whether it can stem further departures, and whether Alphabet's stock stabilizes or continues to price in a talent drain. And as OpenAI's IPO process advances, expect every senior hire to be read as a data point on the company's valuation story. For now, the man who declared attention was all you need has decided OpenAI is where he wants to spend his.