Noam Shazeer: The Man Who Co-Wrote the Paper That Started

In 2017, eight researchers at Google published a paper called ‘Attention Is All You Need.’ The paper introduced the Transformer architecture. Every major AI model you have ever used, GPT, Gemini, Claude, Llama, all of them run on some version of what that paper described. It is arguably the most consequential piece of academic writing of the decade.

One of those eight co-authors is Noam Shazeer. On June 18, 2026, he announced he was leaving Google to join OpenAI.

In a talent war where each hire gets celebrated like a transfer window signing, this one lands differently. This is not a talented engineer moving between companies. This is a founding father of the technology choosing a side.

The path that led here

Shazeer left Google the first time in 2021 to co-found Character.AI, the conversational AI startup that became one of the fastest-growing consumer AI products of that era. Then, in August 2024, Google paid $2.7 billion in a structured partnership to bring him and part of his Character.AI team back into the fold. He returned as vice president of engineering and a co-leader of the Gemini AI models, Google’s flagship product line, CNBC reported.

Less than two years after that $2.7 billion bet, he is leaving again. This time for OpenAI, Google’s most direct competitor in the frontier model race.

What Sam Altman actually said

OpenAI CEO Sam Altman did not try to play it cool. He wrote publicly: ‘Noam is one of the people I have most wanted to work with since the very beginning of OpenAI. Only took 10 years,’ Axios reported.

The ‘only took 10 years’ line is telling. OpenAI was founded in 2015. The Transformer paper came out in 2017. Altman has apparently wanted Shazeer since before the paper made him famous. Now, with Shazeer joining as Lead for Architecture Research, OpenAI has one of the architects of the foundational technology working directly on what comes next.

Why architecture research matters right now

The title matters. Architecture Research is not fine-tuning. It is not safety. It is not product. It is the work of designing the fundamental structure of how AI models process information. The Transformer was an architecture breakthrough. Shazeer is going to OpenAI to help figure out what the next breakthrough looks like.

Every major AI lab is searching for the same thing right now: a way to move beyond the current Transformer paradigm, which is hitting scaling limits that cannot be overcome just by adding more compute. The lab that finds the next architecture advantage will have a significant lead in the next generation of models, Fast Company noted. That is the race Shazeer is now running at OpenAI.

What this costs Google

Google’s loss here is harder to quantify than a departing product manager. Shazeer was not just leading Gemini technically. He was a symbolic anchor. Having a co-author of the foundational AI paper running your flagship model program is a statement about the depth of your research talent. Losing him to a direct competitor is the opposite statement.

Google paid $2.7 billion to get him back. Whatever value that was supposed to protect, the bet did not hold. The company will have to rebuild around his absence on a team that was already under pressure to close the gap with GPT-5, Memeburn reported.

The Colorado AI law landing the same week

Shazeer’s move landed in a week that also saw Colorado’s Consumer Protections for Artificial Intelligence Act scheduled to take effect on June 30, the first comprehensive state AI law in the US. The law requires developers and deployers of high-risk AI systems to protect Colorado residents from algorithmic discrimination. It is narrow in scope but significant in precedent: states are no longer waiting for federal action.

The combination of a major talent shift and the first state AI regulation arriving in the same week is a useful snapshot of where the industry stands. At the frontier, it is still a talent war fought at extraordinary cost. At the regulatory level, the rules are finally starting to arrive.

Why This Matters

The AI talent war has been running for years, but Shazeer’s move has a different quality than a typical senior hire. He is not just bringing skills or credentials. He is a connection to the architectural breakthrough that made the current era possible. If the next era requires a new architectural breakthrough, the lab that has him working on it has a meaningful advantage.

For Google, the wound is both practical and symbolic. Practically, it loses a Gemini leader mid-race. Symbolically, it spent $2.7 billion on a commitment that lasted under two years. In a war where confidence is part of the product, that is an expensive signal to send.

The NewsSparq Takeaway

Three things to hold onto.

One, architecture research is the prize right now. Every frontier lab knows Transformer scaling has limits. Shazeer is at OpenAI specifically to work on what replaces it. That is the most consequential technical bet in the industry.

Two, the $2.7 billion retention failed. Google paid an extraordinary sum and still lost him under two years later. That tells you something about what motivates people at this level of the field, and it is not just money.

Three, the regulatory clock is ticking regardless. Colorado’s AI law takes effect June 30. Federal legislation is moving slowly. The gap between what the labs are building and what the rules cover is shrinking, even if the pace is uneven. The next generation of AI models will be built inside a more regulated environment than the current one was.

The man who helped invent the architecture that powers AI today is now working on what comes next, at OpenAI, not Google. Whether that matters as much as the headlines suggest, only the next generation of models will tell. But in a field where the intellectual lineage of your team shapes what you can build, it matters more than most personnel moves.

Sources: CNBC, Axios, Fast Company, Memeburn.

By The NewsSparq Editorial Desk

The Man Who Co-Wrote the Paper That Started the AI Revolution Just Changed Sides