Noam Shazeer Joins OpenAI After Two Years Back at Google

Noam Shazeer, one of the most influential figures in modern AI research, announced Wednesday that he is leaving Google to join OpenAI. The move ends a second stint at Google that lasted barely two years - and comes after Google paid $2.7 billion in 2024 to bring him back from Character.AI.

Who is Noam Shazeer?

If you use any modern AI system - Claude, GPT-4, Gemini, or any open-source LLM - you are using Noam Shazeer's work.

Shazeer co-authored the landmark 2017 paper "Attention Is All You Need," which introduced the Transformer architecture. That paper fundamentally changed AI research and directly enabled every large language model that exists today. The ideas in that paper - self-attention, multi-head attention, and the attention scaling mechanism - were largely his contributions.

He joined Google in 2000, working on early projects including the search engine's spell checker. Over two decades, he became one of Google's most important AI researchers.

In 2021, Shazeer left Google to co-found Character.AI, a chatbot startup that let users create and interact with AI personas. The company gained millions of users and raised significant funding.

Then, in 2024, Google paid approximately $2.7 billion to bring Shazeer and co-founder Daniel De Freitas back, along with key research team members. Shazeer rejoined Google DeepMind as Vice President of Engineering and co-lead of the Gemini models, specifically tasked with improving Google's reasoning capabilities that were lagging behind OpenAI and Anthropic.

Two years later, he is leaving again - this time for OpenAI.

What HN is Saying

The Hacker News discussion has been active with takes on what this means for both companies.

On the implications for Google:

The dominant sentiment is that this is bad news for Gemini. Multiple commenters noted that Google's brief comeback with Gemini 2.5 Pro last year appeared to be driven by Shazeer's contributions. One commenter pointed out that "Google paid a couple billion dollars to bring Noam back. Really impressive by OAI if this reporting is accurate!"

Others expressed concern about Google's ability to retain top talent. The discussion touched on the classic "big public corp vs private startup" culture divide - once you have to worry about shareholders, regulations, and lawsuits, it becomes difficult to avoid turning into "big corp" culture.

On the "models have no moat" debate:

Several commenters pushed back on the idea that AI models have no moat. One noted that "only like 3-4 companies in the entire world have cutting edge models, that means there is some kind of moat." Others pointed to the deep engineering expertise required: "If it was just a matter of compute on hand and iterating, Meta would be neck and neck with Anthropic, OAI, and Google."

The counter-argument: Google has structural advantages that go beyond individual researchers - custom TPU silicon, more data than anyone else, and phones in 73% of global smartphone users' hands to push AI integration.

On Shazeer's significance:

The thread emphasized that while the "Attention Is All You Need" paper listed authors alphabetically, the critical architectural ideas were largely Shazeer's. One commenter with apparent insider knowledge noted: "The author list was randomized, but the critical idea was truly his."

A detailed reply outlined the paper's history: Jakob Uszkoreit had the initial insight about replacing sequential RNNs with parallel processing layers that could leverage GPU parallelism. When Uszkoreit couldn't get the implementation to outperform RNNs, he brought in Shazeer, who eventually arrived at the performant architecture that became the Transformer.

Why This Matters

This is being called one of the biggest AI talent shifts of 2026, comparable to Andrej Karpathy's earlier move to Anthropic.

OpenAI CEO Sam Altman called Shazeer "one of the people I have most wanted to work with since the very beginning of OpenAI," adding that the partnership was "only 10 years" in the making.

Shazeer himself described it as "a difficult decision to move on."

The move raises questions on multiple fronts:

For Google: What does it say about the internal culture at Google DeepMind that researchers keep leaving despite massive financial incentives to stay? The company has the resources, the data, the hardware - but apparently not the environment that attracts top talent long-term.

For OpenAI: This is a major acquisition of foundational AI expertise. Shazeer brings not just architectural knowledge but also deep experience with scale - both the algorithmic optimizations and the production infrastructure needed to train and serve frontier models.

For the industry: The AI talent market continues to consolidate around a few companies. The same small group of researchers keeps moving between Google, OpenAI, Anthropic, and a handful of startups. The knowledge and techniques they carry with them - including trade secrets - flow between competitors in ways that make the "moat" question genuinely complicated.

What Comes Next

Google DeepMind retains significant bench depth even without Shazeer. They have the hardware infrastructure (TPUs), the data advantage (Search, YouTube, Gmail), and the distribution through Android and Chrome.

But losing the architect of the Transformer twice - first to Character.AI, now to OpenAI - suggests something structural about how Google runs its AI research that top researchers find limiting.

For OpenAI, the question is what Shazeer will work on. His expertise in efficient architectures (he also developed the mixture-of-experts approach) could be applied to making models faster and cheaper to run. Or he could be working on whatever comes after the current Transformer paradigm.

Either way, the AI industry just got a little more consolidated at the top.

Sources

• Noam Shazeer Twitter announcement
• The Decoder: Google's Gemini co-lead Noam Shazeer joins OpenAI
• Benzinga: Sam Altman says it's 10 years in the making
• Hacker News discussion
• Attention Is All You Need (original paper)