All 8 'Attention Is All You Need' Authors Have Left Google

All eight Transformer paper authors have left Google. SemiAnalysis sees a strategic pivot to TPU infrastructure monetization.

All eight co-authors of the 2017 'Attention Is All You Need' paper have now left Google, per SemiAnalysis. Jeff Dean and Sanjay Ghemawat departed to build Discovery Loop, with Oriol Vinyals and Quoc Le following.

Key facts

• All 8 'Attention Is All You Need' authors left Google
• Jeff Dean and Sanjay Ghemawat founded Discovery Loop
• Noam Shazeer joined OpenAI; John Jumper joined Anthropic
• Demis Hassabis stepped back from daily DeepMind leadership
• SemiAnalysis sees pivot to TPU infrastructure monetization

All eight co-authors of the 2017 "Attention Is All You Need" paper — the foundational Transformer architecture paper — have now left Google, according to SemiAnalysis. The exodus includes Jeff Dean and Sanjay Ghemawat, who left to build Discovery Loop, with Oriol Vinyals and Quoc Le joining them. Noam Shazeer is at OpenAI, John Jumper at Anthropic, and Demis Hassabis has stepped back from running DeepMind day-to-day.

This is not a slow trickle; it's a structural break. The people who created the single most cited AI architecture in the last decade no longer work at the company that funded the work. Google's DeepMind and Google Brain — now merged into Google DeepMind — have lost the entire authorship team of the paper that launched the modern AI boom.

SemiAnalysis frames this as a strategic inflection point: Google is shifting from frontier-model competition toward monetizing TPU infrastructure and cloud services. The analyst firm argues the departures reinforce that pivot. If the people who build the models are gone, the remaining value extraction is in selling the hardware and compute those models run on.

The pattern is consistent with what we've seen across the industry. OpenAI has Noam Shazeer, Anthropic has John Jumper, and a host of new ventures like Discovery Loop are being built by former Google researchers. Google's response has been to lean into its infrastructure advantage — TPUs, cloud, and data centers — rather than trying to out-innovate the people it trained.

What remains unclear is whether Google can sustain its research edge without the Transformer architects. The company has not publicly commented on the departures, and the exact scope of Discovery Loop's work has not been disclosed.

Key Takeaways

• All eight Transformer paper authors have left Google.
• SemiAnalysis sees a strategic pivot to TPU infrastructure monetization.

What to watch

Watch for Discovery Loop's first public research output or funding round, which could signal where Dean and Ghemawat are steering the next big AI bet. Also monitor Google's next TPU generation announcement and whether it doubles down on cloud AI services over frontier model releases.

---

Originally published on gentic.news