Noam Shazeer Leaves Google for OpenAI: Transformer Co-Author Defects Ahead of IPO

Noam Shazeer joined Google in 2000, spent 21 years there doing research and building systems, then left in 2021 to cofound Character.AI. Three years later, Google spent $2.7 billion to buy him and his team back, tasking him with co-leading Gemini’s pre-training work. On June 18, he posted a short message on X: he is joining OpenAI.

“I’m excited to share that I’ll be joining OpenAI and look forward to working with the exceptional team there.” Google’s $2.7 billion investment in keeping him just ended.

Why the Transformer Paper Changes the Context

Most people know ChatGPT, Claude, and Gemini all run on Transformer architecture. Few can say where transformers came from. The answer is a 2017 paper published by a Google research team: “Attention Is All You Need.”

Shazeer was one of its eight co-authors. The paper’s central contribution replaced the dominant recurrent neural network approach with a self-attention mechanism, allowing models to process relationships across all positions in a sequence simultaneously rather than reading token by token. That shift made training at scale practically viable. The commercial wave of large language models that followed traces directly back to it.

Calling Shazeer the sole “father of the transformer” overstates it. Eight people are on that paper. Within that group, his role in core architecture design carries weight that AI research circles widely acknowledge.

$2.7 Billion Bought Two Years

When Shazeer left Google in 2021, Character.AI started as an entertainment chatbot company. The underlying technology required significant compute and research investment. In 2024, Google structured a $2.7 billion licensing deal that pulled Shazeer and his core team back. The stated purpose was licensing Character.AI’s technology. The practical effect was reassigning him to the Gemini project.

Gemini’s trajectory improved after that. By October 2025, Gemini had 650 million monthly active users, with its growth rate temporarily surpassing ChatGPT’s during that period. Shazeer’s name came up with increasing frequency in Google’s public communications.

From Google’s perspective, the return on that $2.7 billion investment lasted roughly 24 months.

What OpenAI Gets Before Its IPO

Shazeer’s work at OpenAI will cover two areas: exploring new model architectures, and driving the continued evolution of the existing Transformer framework. The first is a research question about what comes after the transformer. The second keeps OpenAI’s current model lineup competitive during that transition.

OpenAI is targeting a public listing around September 2026, with private-market valuations ranging from $850 billion to $1 trillion. Bringing in a Transformer co-author at this moment sends a clear signal for IPO roadshows: the company is attracting researchers capable of defining next-generation architectures, not just scaling headcount.

Reuters reported that Shazeer’s move marks a notable acceleration in elite AI researcher movement in 2026. OpenAI lost research VP Bob McGrew in 2024 and Ilya Sutskever before that, both departures read as organizational setbacks at the time. Shazeer’s arrival partially addresses that perception.

Google’s Remaining Position

Google’s public response: “We’re grateful for Noam’s meaningful contributions.” Short and final. For a company that typically publishes detailed technical blog posts for significant developments, this level of brevity reads as acceptance rather than rebuttal.

Gemini’s research bench is deep enough to survive one departure. But pre-training work involves architectural decision-making that is difficult to transfer in the short term. There will be a continuity risk between now and when a successor settles into the role.

Anthropic filed its IPO confidentially months ago, with timing expected close to OpenAI’s. Google faces an AI talent market where two well-capitalized competitors are spending aggressively around their respective public listings. Shazeer’s departure adds a concrete dimension to that pressure.

If this was useful, subscribe to the newsletter for weekly AI PM insights and GenAI case studies.

---

Related Reading:

• SpaceX Acquires Cursor for $60 Billion Four Days After IPO
• Anthropic Files Confidential IPO at $965 Billion Valuation
• Claude Fable 5 and Mythos 5 Released: Anthropic’s Dual Flagship Models