Noam Shazeer didn't invent modern AI alone, but he helped write the paper it still runs on. His move from Google to OpenAI is a clean win for Sam Altman and an awkward one for the company that paid $2.7 billion to bring him back.
OpenAI has hired one of the people Google most wanted to keep. Noam Shazeer, the Character.AI co-founder and Gemini co-lead who helped write the 2017 transformer paper 'Attention Is All You Need,' said Wednesday that he is leaving Google for OpenAI, according to Business Insider. If you run a frontier AI lab, this is the kind of hire you don't bury in a staffing memo.
The irony is hard to miss. Google's researchers helped create the transformer architecture, then watched OpenAI turn that idea into ChatGPT, an enterprise business, a consumer brand, and now an IPO story. Shazeer was one of eight authors on the paper that gave the industry its basic technical shape. GPT, Claude, Gemini and Llama don't all look the same, but they all owe a serious debt to that work.
Google has already paid heavily once to get Shazeer back. The Wall Street Journal reported in 2024 that Google paid about $2.7 billion in a deal with Character.AI, giving it access to the startup's technology and bringing Shazeer and other employees back into the company. Character.AI stayed independent, which is the neat legal part. The practical part was simpler: Google wanted the people.
That deal followed a familiar Google AI story. Shazeer and Daniel De Freitas had built a chatbot inside Google before leaving in 2021 to start Character.AI. The startup became one of the most visible consumer chatbot companies before Google pulled its founders back through the licensing arrangement. Shazeer then became a technical lead on Gemini, alongside names including Jeff Dean and Oriol Vinyals. That is not a ceremonial role. Google put him near the center of the model program it needed to make competitive with OpenAI.
Now he is gone again.
You can dress this up as the normal movement of senior engineers in a hot market, but that undersells it. Shazeer isn't just another well-paid researcher moving between logos. He is one of the small number of people whose work sits underneath the whole current AI boom. Business Insider reported that he joined Google in 2000, left for roughly three years to build Character.AI, returned through Google's 2024 deal, and is now headed to OpenAI. That arc tells you more about the AI labor market than another compensation rumor would.
Google paid for access, not permanence
The 2024 Character.AI arrangement always looked like an acquisition in everything but the corporate shell. Google got non-exclusive rights to the technology and brought back the founders. Character.AI remained a separate legal entity. If you were watching the market then, the move sat beside similar deals across AI, where big companies avoided buying startups outright and instead hired the teams they cared about most.
That structure may have made regulatory sense. It didn't solve the harder problem. You can't license loyalty.
Shazeer's reported ownership stake in Character.AI made him wealthy long before this move. Several reports at the time estimated that his share of the Google transaction could have landed in the hundreds of millions of dollars. The exact number matters less than the condition it creates. Once someone has already had that kind of payout, another job move is not mainly about rent, school fees, or the next bonus cycle. It is about where the work feels most important.
That is the uncomfortable part for Google. The company didn't lose a person it couldn't afford to pay. It lost a person who had already seen the inside of both a giant lab and a fast-moving AI startup, then chose OpenAI at the exact moment OpenAI is trying to prove that its lead can survive the next generation of models.
OpenAI gets a cleaner IPO signal
OpenAI's reported IPO preparations make the timing sharper. Axios reported in May that the company was preparing a confidential IPO filing, and other reports have put a possible public listing into the 2026 conversation. Investors looking at OpenAI have one blunt question: is this company still technically ahead, or is it just the best-known brand from the first ChatGPT wave?
Hiring Shazeer helps answer that question in the language investors understand. It doesn't prove OpenAI will build the next model better than Google DeepMind, Anthropic, Meta, or xAI. Frankly, no single hire proves that. But it does show that one of the people who helped define the field believes OpenAI is worth joining now, after the first rush of hype and before the public market starts marking the company every quarter.
OpenAI has not announced Shazeer's exact role. That matters, because a title can tell you whether this is a research hire, a model leadership hire, or a broader technical strategy hire. Until that is clear, the honest reading is narrower: the move is confirmed, the resume is unusually important, and the receiving company is the one Google has been chasing since ChatGPT turned a research lead into a consumer habit.
For readers building companies around AI, the point is not that one engineer changes everything overnight. That is not how frontier model work happens. The point is that the best people in this field still choose environments, not just employers. They go where the compute, peers, product pressure, and ambition line up.
Google still has DeepMind, Jeff Dean, massive infrastructure, and plenty of researchers who don't need introduction. Nobody serious should write it off because one famous engineer left. But this particular departure lands badly because it repeats the same old pattern: Google helps invent the future, hesitates over the product moment, pays to recover talent, then watches OpenAI turn the optics its way.
That is why this hire matters. Not because Shazeer is a mythic lone inventor, but because he is a real engineer with a real claim on the technology everyone is fighting over. OpenAI now gets to put that fact beside its IPO story. Google has to explain why a man it spent billions to bring home decided not to stay.
Also read: Wall Street is betting on Zhipu AI as Washington's export crackdown hands Chinese models a gift • Nvidia's ENPIRE lets AI coding agents train robots to install the GPUs that run AI • An Indonesian gold retailer's 115-place Fortune ranking jump tells you everything about where Southeast Asia's money is going
