DeepMind’s reorganization of its Nobel-winning AlphaFold team is not a retreat from science — it is the end of one organizational model and the arrival of another.
AlphaFold — From Grand Challenge to Platform
November 2020
AlphaFold 2 wins CASP14; solves protein-structure prediction with near-experimental accuracy — a 50-year biology grand challenge.
July 2021
DeepMind and EMBL-EBI launch the AlphaFold Protein Structure Database — initial release covers ~350,000 structures including the full human proteome.
July 2022
Database expands to ~200 million predicted structures — covering nearly every known protein — becoming a standard tool for biologists worldwide.
October 2024
Demis Hassabis and John Jumper awarded a share of the Nobel Prize in Chemistry for AlphaFold’s contributions to protein-structure prediction.
July 29, 2026
Financial Times reports DeepMind has disbanded the dedicated AlphaFold team. Most members reassigned to Gemini, enzyme design, fusion, genomics, and Isomorphic Labs. ~25% of original paper authors have left Google entirely — some to Anthropic.
The Financial Times reported on July 29 that Google DeepMind has disbanded the dedicated team behind AlphaFold, the protein-structure prediction system that earned Demis Hassabis and John Jumper a share of the 2024 Nobel Prize in Chemistry. Most of the team’s key members — including the majority of the original paper authors — have been reassigned within Google and Alphabet. The destinations are telling: Gemini, Google’s large language model program; scientific verticals including enzyme design, nuclear fusion, and genomics; and Isomorphic Labs, Alphabet’s drug-discovery spinoff. Roughly a quarter of the original authors have left Google entirely, with some moving to Anthropic.
Precision matters here. The AlphaFold model itself has not been shut down. The public protein structure database — expanded in 2022 to approximately 200 million predicted structures — remains a standard research tool for biologists globally. Isomorphic Labs is not winding down; it has unveiled a drug-design engine it describes as a step beyond AlphaFold and is moving toward human clinical trials. Google frames the reorganization as reassignment and evolution, not abandonment. The word “dismantled,” used by the FT, reflects the organizational reality — the dedicated team is gone — but the science continues through multiple vehicles.
DeepMind Research Vice President Pushmeet Kohli has framed the change explicitly: it represents a fundamental shift away from DeepMind’s nine-year strategy of attacking singular grand scientific challenges, toward building Gemini-powered automated research systems capable of addressing many scientific problems simultaneously. That framing is the real story.
The key insight: DeepMind is not retreating from science. It is retiring one model of doing science — the hero project, one team, one grand artifact, one Nobel — and replacing it with a platform model: one general system (Gemini) aimed at many scientific problems at once, plus a commercial layer (Isomorphic) built to capture and distribute the value. Even the most celebrated purpose-built model in scientific AI history is folding into the general one.
Nearly a quarter of the original AlphaFold-paper authors have left Google DeepMind (some for Anthropic); most others were reassigned – to Gemini, Isomorphic Labs, and science areas like enzyme design, fusion and genomics. Figures approximate. Source: Financial Times.
The Structural Read
For nine years, DeepMind’s organizing logic was the grand challenge: pick one problem at the frontier of human knowledge, concentrate the best people on it, and produce a singular artifact of world-historical significance. AlphaGo. AlphaFold. The model worked — it produced a Nobel Prize. But it is being retired, and the reason is structural, not sentimental.
The underlying shift is one of value migration. In the early era of deep learning applied to science, the value resided in the specialized artifact itself — the trained model, the architecture, the paper, the database. AlphaFold was the value. But as general-purpose foundation models have grown powerful enough to be pointed at scientific domains, the artifact becomes a case study rather than a moat. The durable value migrates in two directions: upstream, to the general platform that can replicate and extend the capability across many problems; and downstream, to the application and distribution layer that turns capability into commercial outcomes.
That is precisely the two-vector move DeepMind is executing: Gemini (the general platform) absorbs the scientific reasoning capability, and Isomorphic Labs (the application layer) captures the drug-discovery value. The hero project was the right organizational model for a world where specialized models were state of the art. The platform model is the right organizational model for a world where general models are.
Map of AI — Value Migration
From Specialized Artifact to General Platform
In the Map of AI framework, value does not stay fixed at the layer where a breakthrough first occurs. It migrates toward whichever layer achieves the widest distribution at the lowest marginal cost. AlphaFold was a Layer 5 specialized model (scientific application). As Gemini’s scientific reasoning matures, the capability shifts to Layer 3 (foundation model), and the commercial capture moves to Layer 7 (application/distribution — Isomorphic). The organization has to follow the value. DeepMind just did.
The timing of this reorganization, set against the week’s broader signals, is its own argument. Just days after employees at OpenAI and Anthropic petitioned the U.S. government to help deliberately pace automated AI research, the most decorated science lab in AI reorganized itself around exactly that capability: general models doing automated discovery at scale. Anthropic’s own Mythos modelproduced novel cryptanalysis results just this week, a concrete demonstration of what automated scientific reasoning looks like in practice. And Mark Zuckerberg argued this week that invention is the purpose of superintelligence — a thesis DeepMind’s reorganization implicitly endorses by building toward it organizationally.
The talent flow adds a precise data point. The ~25% of original AlphaFold authors who left Google — some specifically to Anthropic — represents the frontier’s gravitational field in miniature: the researchers most associated with a specialized scientific breakthrough migrating toward the lab most focused on scaling general capability. That is not a coincidence. It is a revealed preference about where the next grand challenge will be fought.
Three Implications
ISOMORPHIC LABS IS THE REAL SUCCESSOR
AlphaFold’s scientific legacy does not die — it accelerates commercially. Isomorphic Labs is the vehicle through which DeepMind’s protein-structure and drug-design capability reaches human clinical trials and generates durable economic value. Disbanding the research team is partly how you fund and staff the commercialization push. Watch Isomorphic’s clinical pipeline, not the org chart.
THE HERO-PROJECT MODEL IS OBSOLETE AT SCALE
Every major AI lab is now running some version of this same transition: specialized teams that produced landmark models are being absorbed into general model programs. The era in which a single dedicated team could produce a capability gap large enough to justify permanent organizational independence is closing. Gemini, GPT, Claude — these are the new organizational homes for scientific AI research, not standalone verticals.
ANTHROPIC IS ABSORBING THE TALENT SIGNAL
The researchers leaving specifically for Anthropic — not for academia, not for startups — indicates that the talent perceives Anthropic’s safety-focused scaling approach as the most interesting scientific frontier right now. For Anthropic, landing former AlphaFold authors is both a capability gain and a positioning statement: it is the destination for researchers who care about scientific rigor at the frontier. That compounds Anthropic’s already strong week.
Business Engineer Framework
The Map of AI Redrawn
The AlphaFold reorganization is a live demonstration of value migration across the AI stack — from specialized scientific artifact (Layer 5) to general foundation model (Layer 3) and commercial application layer (Layer 7, Isomorphic). The Map of AI framework maps exactly where in the nine-layer stack each player sits, which layers are compressing, and where durable value is accumulating. DeepMind just redrew its own position on that map. Use the framework to see who else is about to do the same.
The AlphaFold database still exists, Isomorphic Labs is still running experiments toward clinical trials, and the Nobel Prize is not being returned — but the organizational model that produced them is finished. DeepMind has concluded, correctly, that a world where general models can be aimed at any scientific domain is a world that no longer rewards maintaining a separate, dedicated team for each one. The hero project gave AI science its most celebrated artifact. The platform model is now tasked with giving it scale. Whether Gemini-powered automated research can replicate the creative intensity of a focused team working on a single hard problem for years is the open question — and it is the same question the entire industry is now running as a live experiment.
Gennaro is the creator of FourWeekMBA, which reached about four million business people, comprising C-level executives, investors, analysts, product managers, and aspiring digital entrepreneurs in 2022 alone | He is also Director of Sales for a high-tech scaleup in the AI Industry | In 2012, Gennaro earned an International MBA with emphasis on Corporate Finance and Business Strategy.
Scroll to Top
Discover more from FourWeekMBA
Subscribe now to keep reading and get access to the full archive.
