John Jumper helped win a Nobel Prize at Google DeepMind. In June, he said he was leaving anyway. He is headed to Anthropic, where the science is still unproven1.
He has company. Jeff Dean ran Google's AI research for 27 years. In August, he left to found Discovery Loop with three colleagues2. They are betting that a focused lab beats a famous one at the next discovery.
Four Scientists, Two New Labs
Jumper spent nearly nine years building AlphaFold, the system that predicts how proteins fold. He shared the 2024 Nobel Prize in Chemistry for it with Demis Hassabis1. "After nearly nine years, I have decided to leave Google DeepMind and join Anthropic," Jumper said.
Google DeepMind thanked him on the way out. "We are grateful for John's significant contributions to Google DeepMind's work in advancing science and AI," a spokesperson said1. Jumper's title at Anthropic and his start date stayed unannounced.
Dean's exit hit harder. He joined Google in 1999, its 30th employee, and spent 27 years building its systems2. On Aug. 5, he posted his own announcement and named three co-founders: Sanjay Ghemawat, Oriol Vinyals and Quoc Le3.
"The four of us have worked together for 14 to 30 years, and have helped build some of the world's most used products, infrastructure and AI models, and we're excited to turn our attention to this ambitious endeavor," Dean wrote3.
Jeff Dean@JeffDeanAnnouncing Discovery Loop!
I am very excited to announce that, along with my longtime friends and collaborators @Sanjay_Ghemawat, @OriolVinyalsML and @quocleix, we are founding Discovery Loop (@DiscoLoopAI), a Public Benefit Corporation whose mission is to automate machine…
Discovery Loop runs as a public benefit corporation, a structure that lets owners weigh mission alongside profit. Radical Ventures and Khosla Ventures led its first funding round, with Alphabet joining too2.
Anthropic made its own move three months later. On Sept. 24, it opened a life sciences research group and said Claude agents had found a new enzyme system10. The claim arrived before anyone confirmed Jumper had started his new job1.
Four scientists left Google and DeepMind for science startups in 47 days
Dated moves and milestones in AI-for-science research, 2026
Each pin marks one dated move. The magenta pins are scientists leaving a big lab; the gray pin is Anthropic's own science claim that followed.
Sources: CNBC [1]; TechCrunch [2]; Phys.org [10]. Chart by Ryan Elliott Dennis.
The numbers behind this chart
| Date | Event |
|---|---|
| June 19, 2026 | Jumper leaves DeepMind. For Anthropic, after nearly nine years |
| Aug. 5, 2026 | Dean's group founds Discovery Loop. Dean, Ghemawat, Vinyals and Le leave Google |
| Sept. 24, 2026 | Anthropic opens science lab. Claude agents report a new enzyme system |
The Money Chasing an Unproven Bet
Two more labs chase that same bet, with figures attached. Periodic Labs closed a $300 million seed in September 2025, led by Andreessen Horowitz45. Lila Sciences has raised $550 million since 2023, confirmed on its own site6.
Periodic's own newsroom lists only that seed. Outside reporters put a 2026 round at a $7 billion valuation, citing talks ahead of any signed close7. Discovery Loop's backers floated a $10 billion valuation within weeks of its founding, a figure the company itself has yet to confirm11.
Discovery Loop's reported $10 billion dwarfs Lila's confirmed $550 million
Largest valuation reported for each AI-for-science startup, against the funding each has confirmed itself, as of Nov. 2, 2026
Each bar is one company's largest reported number. Magenta bars are estimated valuations from press talks; the blue bar is money Lila confirmed itself.
Lila's bar is money raised, confirmed by Lila. Periodic's and Discovery Loop's bars are estimated valuations from press talks, a different measure read on its own.
Sources: Lila Sciences [6]; Dealroom [7]; Dealroom [11]. Chart by Ryan Elliott Dennis.
The numbers behind this chart
| Item | Value | Note |
|---|---|---|
| Discovery Loop (estimate) | $10.00B | Valuation reported in talks, Sept. 2026 |
| Periodic Labs (estimate) | $7.00B | Valuation reported in talks, March 2026 |
| Lila Sciences | $0.55B | Total funding confirmed by Lila itself |
That gap carries its own lesson. A seed round buys a building and a payroll. Reported valuations price a different bet, on what a lab might find, sometimes years before a paper proves it711.
Why a Lab Beats a Model at Discovery
Ekin Doğuş Çubuk left Google's materials and chemistry team5. He co-founded Periodic Labs with Liam Fedus, once a top OpenAI research leader. Their pitch rests on one limit: a model only repeats what its training already holds.
"Scientific discovery is almost by definition what you haven't been trained on," Çubuk said. "And that's why we're building these labs, so that models can use these labs to tinker with the universe, because we don't feel like you can make a big discovery without trying things"8.
Jeff Dean makes the same case for Discovery Loop's automated experiments, at the company's launch. "You will get both a higher quantity and a higher quality of experiments, and that will lead to scientific breakthroughs and advances," he said2.
Anjney Midha, the Andreessen Horowitz partner who led Periodic's seed round, sizes the target market in trillions. "The industries that Periodic will impact, advanced manufacturing, materials science, semiconductors, energy, aerospace, represent roughly $15 trillion of global GDP," Midha said. "These are the sectors AI has barely touched because you can't transform them with models trained only on text"4.
Geoffrey von Maltzahn, Lila's chief executive, frames the race the same way. "The teams who can turn that wheel with the highest intelligence, at the largest scale, and at the greatest speed will deliver the most transformative advances," he wrote when Lila closed its Series A6.
The Discovery That Drew Doubt
Anthropic's enzyme claim gave the bet its first public test. Roughly 950 Claude agents spent 21 hours scanning DNA sequences. They surfaced a system the company calls ART, built from genetic repeats that also appear in CRISPR10.
Chief executive Dario Amodei called the find a possible "new gene editing mechanism that could have applications in gene therapy"10. He predicted "profound progress in the near future on diseases that have plagued humanity for millennia." The result stayed preliminary and unpublished.
Ilya Finkelstein teaches molecular biosciences at the University of Texas at Austin. He questioned how much of the find was the model's alone. "They can pursue a very large number of unproductive paths rapidly," Finkelstein said. "They are really good at grunt work when orchestrated by experts in the field"9.
Mario Rodríguez Mestre is a computational biologist at the University of Copenhagen. He had studied the same enzymes for four years, and went further. Anthropic's account, he said, understated the human steering behind it. "There was considerably more prior scientific knowledge and human direction behind the search than the phrase 'autonomous discovery' suggests," Mestre said9.
Two more researchers added caveats of their own. Aaron Engelhart, of the University of Minnesota, called the results "very interesting preliminary results"9. They "don't demonstrate gene editing or provide a mechanistic picture of exactly what's happening," he said.
Gustavo Sudre, of King's College London, put the gap in plainer terms. "The gap between 'interesting pattern' and 'useful knowledge' is where most of the work lives," Sudre said9.
What to do
Judge these startups on a named result. A round size only proves funding arrived. Ask what Periodic, Lila or Discovery Loop has published. Let an outside scientist review it, the way Finkelstein and Mestre reviewed Anthropic's claim.
Track where the founders worked before, and what they built there. Fedus led research at OpenAI. Çubuk led materials science at Google. Von Maltzahn built Lila inside Flagship Pioneering for two years before it left stealth. Their prior record is the only proven thing on the table today.
Separate a company's own numbers from a reporter's. Lila's $550 million sits on its own site. Periodic's $7 billion and Discovery Loop's $10 billion sit in press accounts the companies have yet to confirm.
By the numbers
- Nine years: how long John Jumper spent building AlphaFold at Google DeepMind before leaving for Anthropic in June 20261.
- $300 million: Periodic Labs' confirmed seed round, closed in September 202545.
- Jeff Dean's tenure at Google ran 27 years before he left to found Discovery Loop on Aug. 5, 20262.
- Lila Sciences has confirmed $550 million in total funding since 20236.
- 950: the Claude agents Anthropic says analyzed DNA sequences for its enzyme claim10.
- $7 billion: a valuation press reports put on a 2026 Periodic Labs round the company has yet to confirm closing7.
- Four named researchers publicly doubted Anthropic's enzyme claim within two weeks9.
- Discovery Loop's backers floated a $10 billion valuation within weeks of its August 2026 founding11.
What to watch
Watch for a named result a journal accepts first. A round a company announces still proves money alone. Periodic's materials claims, Lila's molecules and Discovery Loop's research cycles all face the same test: a paper an outside lab can check.
A confirmed valuation would be the next best signal. Periodic or Discovery Loop staying quiet about its own reported number extends a silence well past a single news cycle.
Anthropic's enzyme system needs a follow-up study most of all. Finkelstein and Mestre asked for exactly that, and a confirmed mechanism would answer their doubt better than any new round would.
Sources
- CNBC, "John Jumper to leave Google DeepMind for Anthropic," June 19, 2026, https://www.cnbc.com/2026/06/19/john-jumper-to-leave-google-deepmind-for-anthropic.html
- TechCrunch, "Jeff Dean and other top AI researchers are leaving Google to launch their own startup," Aug. 5, 2026, https://techcrunch.com/2026/08/05/jeff-dean-and-other-top-ai-researchers-are-leaving-google-to-launch-their-own-startup/
- Jeff Dean, "Announcing Discovery Loop," X, Aug. 5, 2026, https://x.com/JeffDean/status/2085034604172603724
- Anjney Midha, "Investing in Periodic Labs," Andreessen Horowitz, Sept. 30, 2025, https://a16z.com/announcement/investing-in-periodic-labs/
- TechCrunch, "Former OpenAI and DeepMind researchers raise whopping $300M seed to automate science," Sept. 30, 2025, https://techcrunch.com/2025/09/30/former-openai-and-deepmind-researchers-raise-whopping-300m-seed-to-automate-science/
- Geoffrey von Maltzahn, "Announcing Lila's $350M Series A and Incredible Partners on Our Mission," Lila Sciences, Oct. 10, 2025, https://www.lila.ai/news/series-a-235-million
- Dealroom, "Periodic Labs eyes $7B valuation, fivefold jump in six months," March 26, 2026, https://dealroom.co/news/127042-periodic-labs-eyes-7b-valuation-fivefold-jump-in-six-months/
- Latent Space, "Synthesis Superintelligence: from Semiconductors to Superconductors," Jan. 20, 2026, https://www.latent.space/p/periodic
- Katie Hunt, "What to know about Anthropic's 'Claude-led' biological discovery, and why scientists aren't convinced," CNN, Oct. 8, 2026, https://www.yahoo.com/news/science/articles/anthropic-announcement-claude-led-crispr-145413559.html
- Phys.org, "Anthropic touts AI-led biology discovery," Sept. 24, 2026, https://phys.org/news/2026-09-anthropic-touts-ai-biology-discovery.html
- Dealroom, "Ex-Google engineers' Discovery Loop eyes $50B valuation a month after launch," Sept. 10, 2026, https://dealroom.co/news/161695-ex-google-engineers-discovery-loop-eyes-50b-valuation-a-month-after-laun/
