Google Built the Transformer. Every Author Left.

How the company that published "Attention Is All You Need" lost all eight of its authors to competitors, and why $269 billion in market cap evaporated when the market noticed

Author

Justin Wright
June 29, 2026 • Est. Reading Time: 6 minutes

Happy Monday!

The most important AI paper of the decade. All eight authors listed Google affiliations. None work there today. (source: arxiv)

In 2017, eight Google researchers published a paper called "Attention Is All You Need." It introduced the transformer architecture, the foundation underneath every major large language model shipping today: GPT, Claude, Gemini, Llama, Grok. Every one of them runs on the architecture those eight people built.

As of last week, zero of the eight still work at Google.

Noam Shazeer, co-lead of Gemini and the most prominent of the eight, announced on June 18 that he was leaving for OpenAI. Two days later, John Jumper, who won the 2024 Nobel Prize in Chemistry for AlphaFold, announced he was leaving Google DeepMind for Anthropic. By June 24, two more senior Gemini researchers, Jonas Adler and Alexander Pritzel, followed Jumper to Anthropic. Four departures in six days. Alphabet's stock fell 6.8%, erasing roughly $269 billion in market cap.

The company that invented the technology powering the AI revolution is watching its architects build the revolution at competitor labs.

Google lost four senior AI researchers in six days: transformer co-author Noam Shazeer to OpenAI, Nobel laureate John Jumper to Anthropic, and two key Gemini researchers to Anthropic. All eight authors of "Attention Is All You Need," the 2017 paper that invented the transformer, have now left Google. Alphabet shed $269 billion in market cap. The departures were triggered not by compensation but by internal compute allocation disputes and organizational friction. The company that invented modern AI cannot keep the people who built it.

TL;DR

The $2.7 Billion Boomerang

Noam Shazeer's departure is the one that tells you the most about the structural problem.

In 2021, Shazeer left Google after the company refused to release Meena, a chatbot he and colleague Daniel de Freitas had built. He then founded Character.AI, a generative AI platform that allows users to chat with interactive, AI-driven personas. In 2024, Google paid $2.7 billion to bring Shazeer and part of his team back through an acqui-hire. It was the most expensive talent acquisition in AI history.

Less than two years later, he left again, this time for OpenAI.

The trigger, according to reports, was not compensation. Google reportedly reassigned computing capacity from one of Shazeer's projects to another DeepMind team in London. The reallocation was intended to consolidate pretraining work, but internal competition for advanced chips has become a systemic friction point at Google. The co-author of the most cited AI paper in history left because the company that employs more GPUs than almost anyone on earth could not give him the ones he needed.

Sam Altman welcomed the hire by saying Shazeer was "one of the people he had most wanted to work with since OpenAI's early days." The person who invented the architecture now works at the company that commercialized it.

The Nobel Laureate Who Walked Away

John Jumper's departure two days later compounded the signal. Jumper led AlphaFold at DeepMind for nearly nine years. AlphaFold predicted the 3D structure of over 200 million proteins and is widely regarded as one of the most significant scientific achievements produced by AI. Jumper shared the 2024 Nobel Prize in Chemistry with DeepMind CEO Demis Hassabis for the work.

He left for Anthropic, although his specific role has not been disclosed. In his announcement, Jumper thanked Hassabis for "taking a real chance letting me lead the AlphaFold team just six months after finishing" his PhD, and called DeepMind "a special place." He then walked out the door.

The market treated the departures as a single event. Alphabet dropped 6.8% on June 22, its steepest single-day decline in a year. The Globe and Mail ran a headline asking whether John Jumper was "worth $250 billion."

The Nobel laureate who left the lab where he won it.

Where the Eight Went

The transformer paper had eight authors. Here is where they are now.

"Attention Is All You Need" Authors (2026)

Author

Role on Paper

Current Position

Noam Shazeer

Co-lead

OpenAI (June 2026)

Ashish Vaswani

Lead author

Nvidia (via Essential AI acqui-hire)

Niki Parmar

Author

Anthropic

Jakob Uszkoreit

Author

Inceptive (biotech startup)

Llion Jones

Author

Sakana AI (Tokyo)

Aidan Gomez

Author

Cohere (co-founder/CEO)

Lukasz Kaiser

Author

OpenAI

Illia Polosukhin

Author

NEAR Protocol (co-founder/CEO)

Zero authors still at Google, three at direct competitors (OpenAI, Anthropic), and two leading their own AI startups. One more at Nvidia, one in biotech, and one building blockchain infrastructure for AI agents.

Google published the paper, funded the research, and employed all eight, yet Google retained none of them.

The Structural Problem

The pattern is not random. It points to a structural tension between Google's business model and what its best researchers want.

Google is an advertising company. In Q1 2026, advertising accounted for roughly 75% of Alphabet's revenue. The AI research that Google funds is extraordinary, but it exists inside an organization whose primary incentive is to protect and extend an advertising business. When compute allocation decisions come down to "which project gets the GPUs," the answer is shaped by which project serves the business, not which project serves the science.

Anthropic and OpenAI do not have this problem. They are AI companies, so every GPU serves the core mission. Pre-IPO equity at both companies offers upside that Alphabet's already-public stock cannot match. Anthropic's $965 billion valuation and OpenAI's confidential IPO filing create a gravitational pull for researchers who want both the freedom to do their best work and the financial upside of building something new.

The DeepMind-Google Brain merger in 2023 added organizational friction. Multiple reports describe original DeepMinders feeling the merger "formalized a shift away from fundamental research toward applied AI." David Silver, one of DeepMind's earliest employees and a pioneer in reinforcement learning, left to start his own company. The pattern is consistent: the further Google moves from pure research toward product integration, the more its best researchers leave.

Google raised its 2026 capex projection to $180-190 billion, driven by AI infrastructure demand. The company is spending more on AI than ever, but the researchers who built the technology that justifies that spending are walking out.

What This Means for Practitioners

For engineering leaders, the Google exodus is a case study in what retention actually requires. Google offered the highest compensation, the most compute, and the most prestigious brand in AI research; none of it was enough. What the departing researchers wanted was organizational alignment: a company structured around the work they care about, not a company that funds their work as a side effect of an advertising business. If your best people are leaving despite competitive pay, the problem is not the comp package, it is the org chart.

For AI practitioners choosing tools, follow the talent. Anthropic just added a Nobel laureate, two senior Gemini researchers, and (via Niki Parmar) a transformer co-author. OpenAI added the co-lead of Gemini and another transformer co-author. The models these people build over the next two years will reflect that concentration of talent. Google's models will not get worse overnight, but the pipeline of breakthroughs that produced AlphaFold and the transformer is now distributed across its competitors.

For investors, the $269 billion market cap loss on a single day suggests the market is pricing in a structural risk, not a one-time event. Google can hire replacements, but it cannot replace the institutional knowledge that nine years of AlphaFold development or decades of transformer research represent.

The Bottom Line

Google published "Attention Is All You Need" in 2017. Nine years later, all eight authors are gone. The company that invented the transformer, built AlphaFold, and pioneered large-scale language modeling is hemorrhaging the people who achieved those feats.

The departures are not about money, they are about structure. Google is an advertising company that funds AI. Anthropic and OpenAI are AI companies. That distinction, which seemed like semantics five years ago, is now clearly visible in the stock price.

In motion,
Justin Wright

If the company that invented the transformer, won a Nobel Prize for AlphaFold, and spends $190 billion a year on AI infrastructure cannot keep any of the eight people who built its most important technology, what does retention actually depend on, and is any company immune?

Food for Thought

Quick Hits

  • The Nasdaq fell more than 6% from its June 2 peak, erasing over $1.3 trillion from semiconductor stocks. South Korea's Kospi crashed 10%, triggering a circuit breaker. Amazon's trailing free cash flow collapsed 95% due to AI infrastructure costs. (CNN)

  • ChatGPT's market share fell below 50% for the first time since launch (46.4%). Gemini rose to 27.7%, Claude to 10.3% with 245 million monthly users (4x December 2025). Claude leads paid conversion at 13%. (TechCrunch)

  • OpenAI launched GPT-5.5-Cyber through its Daybreak initiative, scoring 85.6% on CyberGym (highest single-model score ever). "Patch the Planet" program launched with 30+ open-source projects including cURL, Go, and Python. (OpenAI)

  • China unveiled a $295 billion, five-year AI infrastructure plan mandating 80% domestic chips, effectively locking out Nvidia. State firms will operate the data centers. Target: a connected national compute network by 2028. (Bloomberg)

If you haven’t listened to my podcast Mostly Humans: An AI and business podcast for everyone yet, new episodes drop every week!

Episodes can be found below - please like, subscribe, and comment!