If you've used Google's Gemini chatbot, or gotten one of those AI-generated summaries at the top of a Google Search result, some significant part of what you experienced was shaped by two people. Last week, both of them said they were leaving Google. For competitors.

Noam Shazeer, VP of Engineering and co-lead of Google's Gemini models, announced June 18 he's joining OpenAI as Lead for Architecture Research. John Jumper, the 2024 Nobel Prize in Chemistry laureate and co-creator of AlphaFold, announced June 19 he's leaving Google DeepMind for Anthropic. Alphabet's stock fell roughly 7% on June 22 — its worst single-day drop in over a year — wiping approximately $250 billion in market cap.

That's the news. Here's what's underneath it.

Why these two people specifically

Shazeer is one of eight co-authors on "Attention Is All You Need" — the 2017 paper that introduced the transformer architecture (the mathematical framework that every modern AI model, including GPT, Gemini, and Claude, is built on). Every chatbot you've ever used runs on transformers. Shazeer didn't invent AI, but he co-invented the structure that made modern AI possible. He's spent the years since working on what comes next: efficient attention, long-context scaling, mixture-of-experts architectures (the technique that lets large models be accurate without being prohibitively expensive to run).

The $2.7B context matters here. In 2024, Google struck a deal with Character.AI — a startup Shazeer had co-founded after a previous stint at Google — worth approximately $2.7 billion. The deal gave Google technology access and brought Shazeer back to co-lead the Gemini project. He stayed 22 months. Then he announced OpenAI. Sam Altman posted: "noam is one of the people I have most wanted to work with since the very beginning of openai. only took 10 years."

That sentence is more honest than Altman probably intended.

Jumper is a different kind of important. AlphaFold — which he co-created at Google DeepMind — predicted the three-dimensional structure of over 200 million proteins, a problem that had resisted progress for fifty years. Biologists can now find in days what used to take PhD-student-years in a lab. The Nobel committee gave Jumper chemistry's top prize for it in 2024. That's not an honorary badge. It's the actual prize. And now he's joining Anthropic, whose focus on scientific AI has been growing visibly through early 2026.

Source spread

• CNBC — Alphabet has its worst day in over a year on AI concerns after high-profile exits — skeptic. Leads with investor concern; D.A. Davidson analyst frames this as Google "losing the war for talent at the frontier of AI."
• TechCrunch — Nobel laureate John Jumper is leaving DeepMind for rival Anthropic — builder. Focuses on Jumper's scientific legacy and what his move signals for AI-for-science ambitions.
• Axios — Top AI researcher leaves Google for OpenAI — skeptic. Clean timeline of the Character.AI deal; best reporting on what the $2.7B actually covered.
• Bloomberg — Alphabet Shares Slide 7% After Second AI Star Departs for Rival Firm — hype. Market angle; treats the stock move as the primary story.

What's real:

• These are genuinely non-fungible positions. The transformer paper had eight authors. Shazeer was one of them. AlphaFold had a small core team and Jumper was central to it. This isn't a case of losing two strong senior engineers who can be backfilled from the hiring market. The specific architectural intuition Shazeer brings, and the specific scientific vision Jumper brings, don't substitute cleanly.
• The Character.AI deal makes the departure worse, not better. Google ran the expected-value calculation on Shazeer, concluded he was worth $2.7B in effective terms, and signed the agreement. Then he left twenty-two months later. Either the calculation was wrong about what he'd contribute, or something about Google's environment — culture, autonomy, mission alignment — couldn't be bought. Both explanations are uncomfortable.
• Two departures to direct competitors in 48 hours is a pattern, not a coincidence. D.A. Davidson analyst Gil Luria said it plainly after the announcements: the exits were "raising the concern that Google is losing the war for talent at the frontier of AI." One departure is noise. Two, to OpenAI and Anthropic respectively, within two days, is a signal.

What deserves a side-eye:

• Stock markets over-react to narrative. Alphabet's 7% drop may reflect capex fears as much as talent fears. The company's Q1 2026 free cash flow fell 47% year over year — that's a separate story with its own gravity. Attributing $250B in market cap loss purely to two departures is too clean.
• Google still has Demis Hassabis, a deep DeepMind bench, and compute infrastructure no one else has. Saying Google is "losing the AI war" based on one week of exits is a leap. Two departures don't dissolve one of the world's strongest research institutions.
• Jumper's actual role at Anthropic is unannounced. He could work on biology and scientific AI (the obvious bet). He could do something completely different. The move matters, but the actual impact is unknown until he builds something we can evaluate.

What to do about it

This one is more watch-list than action-list.

• If you use Gemini regularly, watch what changes over the next 12-18 months. Model architectures are set far in advance. The Gemini you use today reflects Shazeer's work. The one you use in 2027 reflects whoever leads after him. Monitor whether quality on complex tasks tracks the same direction.
• OpenAI products are about to have serious architectural focus behind them. Shazeer's specialty — making transformers more efficient and capable at the architectural level — could show up in GPT products in late 2026 or early 2027. Worth knowing if you're deciding which AI tools to commit to for the long term.
• For anyone in science or medicine: Jumper at Anthropic is a longer-game story. AlphaFold changed drug discovery; what he builds next with Claude's capabilities is an open question. Track it, especially if scientific AI is part of your work.
• For builders: this doesn't change what you ship this quarter. Existing models don't change because of job announcements. The signals matter for 2027 roadmap planning, not for the API call you make this week.

Further reading

• CNBC — Alphabet has its worst day in over a year on AI concerns after high-profile exits
• TechCrunch — Nobel laureate John Jumper is leaving DeepMind for rival Anthropic
• Axios — Top AI researcher leaves Google for OpenAI
• CNBC — Google Gemini co-lead Noam Shazeer leaves for OpenAI
• Bloomberg — Alphabet shares slide 7% after second AI star departs for rival
• "Attention Is All You Need" — the 2017 transformer paper (arXiv)
• AlphaFold Protein Structure Database