Noam Shazeer announced on 19 June that he is leaving Google DeepMind to join OpenAI as Lead for Architecture Research — a move that represents one of the most consequential talent shifts in the AI industry's history.
Shazeer is a co-author of the 2017 paper 'Attention Is All You Need' that introduced the Transformer architecture, the foundation underpinning every major language model in production today including GPT, Claude, Gemini, and Llama. He also co-authored the Sparsely-Gated Mixture of Experts paper, another architecture now used in essentially all frontier models. At Google, he served as VP of engineering and co-led development of the Gemini model family.
The financial context makes the departure particularly striking. Google paid $2.7 billion to acquire Shazeer's startup Character.AI in August 2024, with the deal structured primarily as a talent acquisition to bring Shazeer back to the company. His departure after just 22 months means Google effectively spent $2.7 billion for less than two years of his leadership on Gemini.
Sam Altman described Shazeer as 'one of the people I have most wanted to work with since the very beginning of OpenAI.' Alphabet's stock rose 1.17 per cent on the news, suggesting investors view Google's computational infrastructure advantages as more durable than any individual researcher — though the long-term impact on Gemini's architectural direction remains uncertain.
For context engineers, Shazeer's move to OpenAI suggests that the company is investing heavily in next-generation model architectures rather than simply scaling existing approaches. His expertise in mixture-of-experts and attention mechanisms could shape the architecture of GPT-6 and beyond, potentially shifting the competitive dynamics between frontier model providers.
