Before the next pandemic knocks on the door, scientists may already have a molecular blueprint waiting. Chip giant Nvidia, alongside Google DeepMind and the European Molecular Biology Laboratory's European Bioinformatics Institute (EMBL-EBI), announced on Thursday, 24 September 2026 that AI-predicted protein complex structures for more than 2,800 viruses are now openly available to researchers worldwide — giving virologists and drug-discovery teams a structural head start against potential outbreaks.

What is actually being released — and why it matters

Protein complex structures reveal how a virus's proteins interact — with each other and, critically, with host cells. Knowing that geometry is the first step toward designing drugs, vaccines, or diagnostics. Historically, resolving these structures required expensive, time-consuming laboratory techniques like cryo-electron microscopy. AI prediction collapses that timeline from years to hours.

The new dataset covers predicted complex structures — proteins working together, not just individual chains — across viruses, a significantly harder modelling problem than single-protein prediction. Making them openly available means a researcher in Pune, Nairobi, or São Paulo can access the same starting point as a team at a well-funded Western university.

Standing on AlphaFold's shoulders

This release is a direct descendant of the AlphaFold Protein Structure Database, jointly launched by Google DeepMind and EMBL-EBI in July 2021. That database eventually covered predictions for more than 200 million proteins and is widely credited with accelerating structural biology research globally. The new viral-complex dataset pushes the frontier further: from cataloguing individual proteins to modelling how they assemble and interact inside a pathogen.

Nvidia's role here is as AI-infrastructure backbone — the GPU computing power that makes running protein-folding models at scale feasible. The collaboration underlines a broader pattern: hardware companies, AI labs, and bioinformatics institutes converging on open biological data as a shared public good, particularly after COVID-era urgency demonstrated how slow preparedness can cost lives.

Open access as pandemic preparedness strategy

The phrase 'openly available' carries real weight. Proprietary databases lock structural data behind institutional agreements; open ones let any scientist integrate the structures into existing pathogen databases, test drug candidates computationally, or publish peer-reviewed findings without a licensing barrier. Public health researchers tracking emerging zoonotic viruses — the category most likely to seed the next outbreak — now have a dramatically expanded structural library to search.

What to watch next: whether the dataset gets integrated into major pathogen surveillance platforms, how quickly peer-reviewed studies cite the structures, and whether follow-on releases extend coverage to host-pathogen interaction complexes — the molecular handshakes where viruses actually invade cells.

The blueprint exists. The next race is to use it before an outbreak forces the world to build one under pressure.