A Petabyte of Answers to a Question Biology Could Not Afford to Ask
On September 8, 2026, Google DeepMind published the AlphaGenome Atlas, a free public dataset that scores the likely molecular effect of every one of the roughly 9 billion possible single-letter changes to the human genome. Each variant gets thousands of predictions across gene regulation in hundreds of human and mouse cell types, plus a single summary number DeepMind calls the AlphaGenome Variant Impact score, built for both the 2 percent of the genome that codes for proteins and the 98 percent that does not. The whole resource is 1 petabyte, more than 30 times the AlphaFold Protein Structure Database, and it is available through a free web portal, a public API, and as a skill inside Google's Antigravity coding tool.
| Resource | Size | What it predicts |
|---|---|---|
| AlphaFold Database | 200+ million protein structures, up from about 190,000 experimentally solved ones | 3D protein shape |
| AlphaGenome Atlas | 1 petabyte, over 30 times the AlphaFold Database | Molecular effect of 9 billion possible single-letter DNA changes |
Before this release, predicting what a single mutation does to gene regulation required running a specialised model yourself, one variant at a time, with your own compute. DeepMind has now pre-computed the answer for essentially every possible variant and put it behind a free lookup. For a European genomics lab, a rare-disease diagnostics team, or a pharma research group screening candidate targets, that turns a weeks-long compute job into a search query.
Free Today Is Not the Same as Free Always
Google's own announcement states non-commercial academic use is free now and commercial access will come through Google Cloud's Model Garden, with no price disclosed. That sequencing matters more than the science. AlphaFold, DeepMind's earlier release, followed the identical path: a free public database first, then paid enterprise deployment routes opened later as the tool became embedded in industry workflows. Nothing about that is dishonest, but it means the AlphaGenome Atlas should be read as a loss leader, not a permanent public utility, and any organisation building it into a production pipeline should plan for a paid tier from day one rather than being surprised by it.
The Procurement Question Nobody Is Asking Yet
The practical move for a European biotech, university lab, or health-tech vendor is to build against the open API and the documented AlphaGenome Variant Impact score now, while it is free and while the interface is still being shaped, rather than waiting for the commercial version to stabilise. Code written against a well-documented free API tends to port cleanly to the paid version later; code written to work around a moving, undocumented workaround does not. The variant impact score itself, not the raw predictions, is the part most likely to become a paid feature, since it is the compressed, directly usable output DeepMind would have the clearest incentive to gate.
Servola Journal
We do this for everyone trying to keep up with what technology is doing to our lives. The people who build it, and the people it happens to. The Servola Journal exists so that what we learn belongs to all of them.
Nobody pays us for this. No ads, no paywall, free to everyone. We just believe that understanding what's happening to all of us shouldn't depend on who can afford to pay for it.
If it gave you something today, tell us to keep going. Follow us, leave a like, or write a positive comment. We read every one, and they are what keeps us going.
Read next: Half of Google's Weather Claims Are Independently Checked | Gemini 3.8 Flash: Three Weeks, Pricing Risk
