Noam Shazeer Quits Google, Joins OpenAI
The co-inventor of the Transformer just walked out the door Google paid $2.7 billion to keep shut
Less than two years after a record-breaking deal brought him back, one of AI’s most important researchers is switching sides again
The Noam Shazeer OpenAI move landed this week as one of the more talked-about personnel shifts in AI, and the timing makes it sting. Have you ever wondered what it actually costs to keep your best person from walking out the door — and whether that price has a shelf life?
On June 18, Noam Shazeer announced on X that he’s leaving Google to join OpenAI. He was Google’s vice president of engineering and a co-lead of the Gemini model family. He’s also one of eight co-authors of the 2017 paper “Attention Is All You Need,” which introduced the Transformer architecture — the technical foundation underneath essentially every major large language model in use today, including ChatGPT, Gemini, and Claude.
Here’s the part that makes this story land harder than a typical AI hire: Google reportedly paid around $2.7 billion less than two years ago specifically to get Shazeer back. Now he’s gone again, to the company Google competes with most directly.
noam is one of the people I have most wanted to work with
since the very beginning of openai. only took 10 years.
to bring Shazeer back in 2024
actually lasted
the Transformer paper
up to $1T valuation
This isn’t his first time leaving Google
BackgroundShazeer joined Google back in 2000 and spent two decades there before his first departure. In 2021, after Google declined to publicly release a chatbot he and colleague Daniel De Freitas had built internally, the two left to found Character.AI. That startup became one of the most popular AI companion apps on the market.
Google brought Shazeer and De Freitas back in August 2024 as part of a licensing partnership with Character.AI, reportedly paying around $2.7 billion for the arrangement and the return of the researchers. It was framed at the time as a major coup — Google reclaiming talent it had effectively lost to its own caution.
Since returning, Shazeer rose to VP of Engineering and became a technical co-lead of Gemini, Google’s flagship AI model family, which has made real competitive progress against OpenAI over the past year.
This is now the second time Shazeer has left Google for greener pastures. The pattern suggests his loyalty runs to the work and the team, not necessarily to any one employer’s logo.
The $2.7B retention didn’t even last two years
The lessonThis is the detail driving most of the conversation around the move. Massive acqui-hire deals are usually framed as a way to lock in talent for the long haul, but Shazeer’s exit shows that even a multi-billion-dollar arrangement has a shelf life if the person eventually wants to be somewhere else.
Acqui-hire agreements typically include retention periods and incentive structures designed to keep key people in place for several years. Shazeer’s departure, less than two years after the deal, suggests those structures either expired faster than expected or weren’t enough to outweigh the pull of working at OpenAI.
It’s a meaningful data point for any company currently considering a similar high-priced talent deal: money buys presence for a defined window, not guaranteed permanence.
Expect this case to come up in future negotiations and analysis whenever a tech company is weighing whether a giant retention package is actually worth the price tag.
The timing lines up with OpenAI’s IPO push
Strategic contextOpenAI filed a confidential S-1 with the SEC earlier this month, targeting a listing as early as September at an estimated valuation of up to $1 trillion. Bringing in a researcher of Shazeer’s caliber right now functions as both a genuine capability gain and a signal to the market about who wants to work there.
Shazeer is widely respected for his work on model pretraining and efficient architecture — including Mixture of Experts and Multi-Query Attention, two techniques that materially affect how much it costs to run a model at scale. For a company about to go public on the strength of its frontier model roadmap, that expertise is directly relevant to margins, not just prestige.
It also continues a broader pattern: OpenAI has been hiring aggressively from rivals over the past year while expanding its enterprise distribution, treating talent acquisition as part of the same competitive push as product launches and fundraising.
Recruiting one of the Transformer’s co-inventors right before an IPO is about as direct a statement as a company can make about where it believes the frontier of AI research actually lives.
Google’s Gemini roadmap loses its co-lead at a sensitive moment
What Google losesGemini has made real, measurable progress against OpenAI’s models over roughly the past year, and much of that progress runs through architecture decisions Shazeer was directly involved in shaping. Losing one of Gemini’s co-leads doesn’t halt the program, but it does mean Google moves forward without its most architecturally experienced leader at a moment when the competitive gap with OpenAI may be the narrowest it’s been in some time.
Google has continued shipping aggressively in 2026, including new Gemini-branded products and an agent-focused platform announced at its developer conference earlier this year. The departure doesn’t change those product timelines directly, but it does raise questions about institutional knowledge — the kind of hands-on understanding of what works at scale that doesn’t transfer cleanly through documentation.
For now, Google hasn’t named a public replacement for Shazeer’s specific role, and how the Gemini research organization adjusts internally is something worth watching over the coming months.
The loss is less about a single person’s title and more about the architectural judgment that took years to build, walking out the door at a moment Google can least afford it.
Two decades at Google
Joins Google in 2000, later co-authors the 2017 Transformer paper and works on major projects including the chatbot Meena.
Leaves to found Character.AI
After Google declines to publicly release Meena, Shazeer and Daniel De Freitas depart to start their own AI companion company.
Google brings him back for $2.7B
A licensing partnership with Character.AI returns Shazeer and his team to Google DeepMind. He becomes VP of Engineering and Gemini co-lead.
Announces move to OpenAI
Shazeer posts his departure on X. Sam Altman publicly welcomes him hours later, calling it a 10-year-long recruiting effort finally paying off.
For the past few years, the loudest competition in AI has played out in product launches, benchmark scores, and funding rounds. Shazeer’s move is a reminder that there’s a quieter, equally consequential competition happening underneath all of that: the fight to retain the small number of people who genuinely understand how to design and scale the systems everyone else is racing to build on top of.
Massive acqui-hire and retention deals, like the one that reportedly cost Google $2.7 billion, have become a normal tool in that fight, but Shazeer’s exit is evidence that they function more like a multi-year lease than permanent ownership. Companies can pay to extend someone’s tenure, but they can’t fully insulate themselves from a rival simply offering something more compelling — whether that’s a mission, a team, or proximity to an IPO.
The deeper pattern worth watching is what this does to the broader talent market. When a move this prominent happens, it tends to make every other senior researcher’s retention package look temporary rather than settled, which puts pressure on every major AI lab to keep sweetening the deal just to stay in place.