{"slug": "nobody-shut-down-alphafold-check-your-license-instead", "title": "Nobody Shut Down AlphaFold. Check Your License Instead.", "summary": "Google DeepMind dissolved the team that built AlphaFold, but the AlphaFold Protein Structure Database remains operational and is hosted by EMBL-EBI, not Google, with no deprecation or sunset date. The real risk lies in the Google-hosted AlphaFold Server, which has no SLA and could be switched off, while the AlphaFold 3 model weights are restricted by separate non-commercial terms. The open ecosystem has already moved beyond Google's involvement, with forks and alternatives emerging.", "body_md": "[AI](https://sourcefeed.dev/c/ai)Article\n\n# Nobody Shut Down AlphaFold. Check Your License Instead.\n\nGoogle dissolved the team, not the database, and the real migration risk was never Google's org chart.\n\n[Priya Nair](https://sourcefeed.dev/u/priya_nair)\n\nThe headlines this week say Google shut down AlphaFold. It didn't. The [AlphaFold Protein Structure Database](https://alphafold.ebi.ac.uk/) is still serving more than 200 million predicted structures over the same REST endpoints it did last month. [AlphaFold Server](https://alphafoldserver.com/) still takes jobs. The AlphaFold 3 repo is still on GitHub. Nothing has been deprecated, no sunset date has been published, and if you have a pipeline hitting `alphafold.ebi.ac.uk/api/prediction/`\n\nin production, it will keep working tomorrow.\n\nWhat actually happened, per the Financial Times, is an org chart change: [Google DeepMind](https://deepmind.google/science/alphafold/) has dissolved the team that built AlphaFold. Most authors of the original papers were reassigned over the past year — to Gemini-adjacent work, enzyme design, fusion, genomics, or over to Isomorphic Labs. Roughly a quarter of the full-time DeepMind authors left the company outright. Nobel co-laureate John Jumper announced his departure for Anthropic on June 19, and two core AlphaFold researchers, Jonas Adler and Alexander Pritzel, followed him. DeepMind's research VP Pushmeet Kohli framed it as strategy that \"has evolved\" away from singular grand challenges.\n\nThat's a real story. It's just not the story the aggregators wrote, and the difference matters if you're the one on call.\n\n## The dependency you should actually be auditing\n\nSort your AlphaFold exposure into three buckets, because they carry completely different risk.\n\n**The database is the safest thing you depend on**, and it's the one people panic about. AFDB is operated by [EMBL-EBI](https://www.ebi.ac.uk/), a publicly funded European institute, not by Google. The predictions are CC-BY 4.0. The partnership was renewed in October 2025, and the most recent release synced coverage to UniProt 2025_03 and added downloadable MSAs. Your risk here has never been Google's headcount — it's EMBL-EBI's grant cycle and, more practically, update cadence. If nobody at DeepMind is generating new bulk predictions, AFDB freezes as a snapshot rather than dying. For most consumers that's tolerable. If it isn't, the mitigation is boring and you should have done it already: pull the bulk archives from the Google Cloud public dataset or EBI's FTP, pin a version, and stop treating a free public API as a runtime dependency. Rate-limited third-party HTTP calls inside a hot path were a bad design before this week.\n\n**AlphaFold Server is where the single-vendor risk actually lives.** It's a Google-hosted web app, non-commercial use only, capped at a few dozen jobs a day — 30, at last count. No SLA, no contract, no migration commitment. It's the one artifact in this story Google could switch off unilaterally on a Friday, and the team that would have argued internally against switching it off no longer exists. If AlphaFold Server is load-bearing for anything beyond exploratory one-offs, that's your action item.\n\n**The model weights were never yours to begin with.** This is the part the shutdown narrative buries. The [AlphaFold 3 code](https://github.com/google-deepmind/alphafold3) is Apache 2.0, which sounds great until you read `WEIGHTS_TERMS_OF_USE.md`\n\n: the parameters ship under separate terms, you may only use them if you received them directly from Google, and commercial use is off the table. There's an application form and a review. Anyone building a product on structure prediction hit that wall in 2024 and routed around it years before this reorg.\n\n## The community already forked the future\n\nHere's the argument the news coverage misses entirely: AlphaFold's institutional decline barely dents the field, because the open ecosystem stopped waiting for Google two years ago — and it did so specifically because of the licensing, not the science.\n\n[OpenFold](https://openfold.io/) shipped OpenFold3 with weights, training code, evaluation scripts, and — as of March 2026 — the training data itself, all under Apache 2.0, at near-parity with AF3 across most modalities. Boltz-2, from MIT and Recursion, went out under a plain MIT license in June 2025 and did something AF3 doesn't: joint structure and binding-affinity prediction, roughly 18 seconds on one consumer GPU, which its authors benchmark as approaching FEP-quality affinity estimates around a thousand times faster. ByteDance's Protenix-v1 is Apache 2.0 for academic and commercial use. Chai-1 loosened its terms after launch under the same competitive pressure.\n\nRead those releases together and the pattern is unmistakable. Every one of them leads with license permissiveness, because permissiveness was the gap. AF3 set the accuracy bar; the open models cleared enough of it that \"can I ship this?\" became the deciding question, and on that question AF3 loses to four alternatives. DeepMind published the seminal work, then priced itself out of the downstream ecosystem it created.\n\n## Who wins, who loses, what to do Monday\n\nThe winners are the consortia. OpenFold, Boltz PBC, and Protenix now inherit the mantle by default — not because they out-researched DeepMind, but because they were the ones you could legally build on when it mattered. Expect the center of gravity in co-folding to keep drifting toward whoever ships weights you can put in a container.\n\nIsomorphic Labs wins differently, and it's worth being clear-eyed about it: the applied protein-structure lineage moved inside a commercial drug-discovery subsidiary. That work will produce results. Most of them won't come with a GitHub repo.\n\nThe losers are academic labs that treated AlphaFold Server as free infrastructure and AF3 weights as the obvious default. That default is now maintained by nobody in particular. Not gone — unowned, which is worse in a subtler way, because there's no announcement to react to.\n\nConcretely: if you consume AFDB, snapshot it and pin the version. If you use AlphaFold Server, stand up Boltz-2 or OpenFold3 locally this quarter; both run on a single GPU and neither will ask you for a license attestation. If you're on AF3 weights under a research grant, understand you were already blocked from commercialization and this changes nothing except the odds that the terms get renegotiated in your favor.\n\nOne honest caveat: the FT report is the sole origin for the reorg details, and everything downstream is aggregation of it. DeepMind has confirmed a strategy shift, not a product wind-down. If AFDB or AlphaFold Server gets a formal deprecation notice, that's news. Right now the news is that a Nobel-winning team was quietly absorbed into a chatbot roadmap, and the field it created had already stopped needing it.\n\n## Sources & further reading\n\n-\n[Google shuts down its Nobel-prize winning AlphaFold project as it focuses on Gemini](https://www.engadget.com/2225849/google-shuts-down-alphafold/)— engadget.com -\n[Deepmind dismantles its AlphaFold team as key authors leave for Anthropic](https://the-decoder.com/deepmind-dismantles-its-alphafold-team-as-key-authors-leave-for-anthropic/)— the-decoder.com -\n[Nobel laureate John Jumper is leaving DeepMind for rival Anthropic](https://techcrunch.com/2026/06/20/nobel-laureate-john-jumper-is-leaving-deepmind-for-rival-anthropic/)— techcrunch.com -\n[EMBL-EBI and Google DeepMind renew partnership and release update to AlphaFold Database](https://www.embl.org/news/science-technology/google-deepmind-partnership-renewal/)— embl.org -\n[google-deepmind/alphafold3](https://github.com/google-deepmind/alphafold3)— github.com -\n[OpenFold Consortium Announces Major OpenFold3 Update and Public Release of Training Data](https://www.businesswire.com/news/home/20260313170622/en/OpenFold-Consortium-Announces-Major-OpenFold3-Update-and-Public-Release-of-Training-Data-for-Reproducible-Biomolecular-AI)— businesswire.com -\n[MIT and Recursion Release Boltz-2: Next Generation AI Model to Predict Binding Affinity](https://ir.recursion.com/news-releases/news-release-details/mit-and-recursion-release-boltz-2-next-generation-ai-model)— ir.recursion.com\n\n[Priya Nair](https://sourcefeed.dev/u/priya_nair)· AI & Developer Experience Writer\n\nPriya covers AI frameworks, developer productivity tooling, and the startup ecosystem across South and Southeast Asia, bringing a researcher's rigour and a practitioner's empathy to every story. She is deeply sceptical of benchmarks and asks hard questions so her readers don't have to.\n\n## Discussion 0\n\nNo comments yet\n\nBe the first to weigh in.", "url": "https://wpnews.pro/news/nobody-shut-down-alphafold-check-your-license-instead", "canonical_source": "https://sourcefeed.dev/a/nobody-shut-down-alphafold-check-your-license-instead", "published_at": "2026-07-29 21:09:17+00:00", "updated_at": "2026-07-29 21:27:33.514999+00:00", "lang": "en", "topics": ["artificial-intelligence", "ai-research", "ai-products", "ai-infrastructure"], "entities": ["Google DeepMind", "AlphaFold", "EMBL-EBI", "AlphaFold Server", "Anthropic", "John Jumper", "Isomorphic Labs", "Pushmeet Kohli"], "alternates": {"html": "https://wpnews.pro/news/nobody-shut-down-alphafold-check-your-license-instead", "markdown": "https://wpnews.pro/news/nobody-shut-down-alphafold-check-your-license-instead.md", "text": "https://wpnews.pro/news/nobody-shut-down-alphafold-check-your-license-instead.txt", "jsonld": "https://wpnews.pro/news/nobody-shut-down-alphafold-check-your-license-instead.jsonld"}}