Noam Shazeer, vice president of engineering at Google and co-lead of its Gemini AI models, announced on X that he is leaving Google to join OpenAI.
In his post, Shazeer wrote: “I’m excited to share that I’ll be joining OpenAI and look forward to working with the exceptional team there. It was a difficult decision to move on. I’m incredibly proud of the amazing team at Google and everything we’ve built together. It has been an honor and a pleasure to work with all of you.”
The move comes less than two years after Google brought Shazeer back in August 2024 as part of a reported $2.7 billion deal that also included licensing technology and bringing over a team of researchers from Character.AI, the startup he co-founded after previously leaving Google in 2021.
Background and Role
As reported by Reuters, Shazeer joined Google in 2000. He is a co-author of the seminal 2017 “Attention Is All You Need” paper that introduced the transformer architecture and helped catalyze the current AI boom. At Google, he was appointed co-lead of Gemini development in 2024 and has been credited as a key figure in helping Gemini close the performance gap with OpenAI’s ChatGPT.
The departure underscores the ongoing high-stakes talent competition among frontier AI labs as companies race to advance model capabilities.
Actions to Take
Talent and org watchers at frontier labs and AI-native companies: Track follow-on moves from the Gemini team. Shazeer’s exit as co-lead of a major model effort often precedes additional researcher or engineering departures in these environments.
Enterprise AI strategy teams: Re-evaluate reliance on any single lab’s model roadmap. A high-profile architecture lead move like this can accelerate internal debates about multi-model strategies or building more proprietary scaffolding around base models.
Investors and acquirers evaluating AI talent moats: Factor the $2.7B Character.AI return cost into retention math. Short tenure at this level (under two years post-return) highlights how expensive and fragile top-tier model architecture talent remains.
Compute allocation and roadmap planners: Watch for any public signals on whether Shazeer’s new role at OpenAI involves architecture or pre-training leadership. Such moves frequently correlate with shifts in training cluster priorities or new model family direction at the destination lab.
Competitive intelligence teams: Monitor OpenAI’s next model releases and technical reports for any architectural or scaling innovations that align with Shazeer’s known areas of expertise (transformers, mixture-of-experts, efficient architectures).
This is one of the more significant single-researcher moves in the current cycle given Shazeer’s foundational contributions and recent role running one of Google’s flagship model efforts.