Spain-Based Startups Advance Regional Innovation With Barcelona Supercomputing Center
Oct 19, 2026, 10:00 PM · NVIDIA Blog

Spain's national supercomputer is opening its doors to startups, with NVIDIA's startup program as a side entrance. Sovereign compute is starting to look like a real product, not just a slogan.
Why it matters
The Barcelona Supercomputing Center says its AI Factory Accelerator Programme is now fully operating after a pilot. It gives startups access to public compute, technical support and training. As of this announcement, some of the BSC AI Factory's services can also be reached through NVIDIA's Inception startup portal.
BSC says its AI Factory, one of Europe's first and the first in Spain, has delivered more than 800 services to public and private clients since launching in April 2025. An upcoming extension of the MareNostrum 5 supercomputer is meant to feed capacity into this and other European AI factories.
The pilot's three named champions show what this kind of access buys: a water-network platform used by more than 250 utilities in 16 countries, a hospital spin-off looking for hidden heart-rhythm problems in normal-looking ECGs, and a voice-driven agent system for pharmaceutical sales and doctor information.
From the desk
We think this is the right use of public supercomputing. Most young companies cannot rent serious GPU time at market prices, and the ones that most need it are often working on things like water systems and stroke prevention rather than chat apps. Qatium training its own models on MareNostrum 5 to help utilities make daily operating decisions is exactly the kind of work that public infrastructure should make cheaper. OneCareAI, spun out of BSC itself, trying to flag atrial fibrillation risk before a stroke, is the kind of payoff that justifies the bill.
That said, it is worth naming whose announcement this is. The source is NVIDIA's own blog, and nearly every startup is described through the NVIDIA tools it uses: NIM microservices, Nemotron models, NeMo libraries, cuGraph. Europe's sovereign AI push is meant to reduce dependence on a few foreign suppliers. Routing access to a public facility through one American chipmaker's startup portal, and building the cohort on that company's software stack, sits awkwardly next to that goal. Sovereign over the building and the electricity is not the same as sovereign over the stack.
The medical and pharmaceutical entries deserve extra care. A model that spots hidden heart-rhythm risk could save lives, and it could also flood clinicians with false alarms if it is not validated on diverse patients. Avatia's agents feeding doctors information about medicines on behalf of drug companies raises an obvious question: curated by whom, and with what checks against the line between information and marketing? The announcement does not say, and it does not report any clinical results.
Where this leads if it scales: a European network of public AI factories that gives startups a real alternative to hyperscaler pricing. That is good. The risk is that the public money builds capacity while the lock-in quietly flows to the vendor whose software everyone learned on. I'm watching for whether BSC's access terms stay open to teams on other stacks.
Context
Europe has been funding a series of AI factories built around national supercomputing centers so researchers and companies can train and run models on publicly owned infrastructure. MareNostrum 5 is BSC's flagship system and is powered by NVIDIA accelerated computing. Other startups in the cohort include Agrikola AI, Elna Health, Lokutor, Montevive AI, Oppy, Pack2Earth and Up2Smart, spanning agriculture, medical imaging, speech, security, sales intelligence and packaging.
Who feels it
- European startups
- Subsidized access to serious compute plus training and mentoring, now reachable through an additional NVIDIA channel.
- Utilities and hospitals
- Early tools for network operations and cardiac risk screening, though validation evidence is not part of this announcement.
- Policymakers
- A test of whether sovereign AI infrastructure reduces dependency or simply relocates it to a single supplier's software.
What to watch
- The timing and size of the MareNostrum 5 extension
- How many startups access BSC services through the NVIDIA Inception portal versus directly
- Published clinical validation for OneCareAI's ECG work
- Whether the accelerator admits teams building on non-NVIDIA stacks
Companies: NVIDIA