AivexaNewsSearch
AI news for builders and product teamsChecked every hour

Spain-Based Startups Advance Regional Innovation With Barcelona Supercomputing Center

Collected Oct 20, 2026

The Barcelona Supercomputing Center's AI Factory Accelerator Programme has moved from a pilot into full operation, and NVIDIA used the milestone to announce that Inception members can now reach some BSC AI Factory services through the Inception portal. BSC AI Factory is described as among Europe's first AI factories and the first in Spain; since its April 2025 launch it has delivered more than 800 services to private and public-sector clients. NVIDIA dates the new Inception portal access to today.

The programme's pitch to startups is sovereign compute, technical support and training across industries. NVIDIA's contribution runs alongside it: Deep Learning Institute open courses for training, business mentoring, venture capital introductions and other resources. Inception, for context, is NVIDIA's long-running programme for early-stage companies, typically offering credits, technical guidance and go-to-market help; pairing it with a national supercomputing accelerator means a startup can draw on both commercial and public computing channels.

The compute side is MareNostrum 5, BSC's public infrastructure for scientific and technological innovation, which runs on NVIDIA accelerated computing. An upcoming extension of MareNostrum 5 is slated to deliver supercomputing access through the BSC AI Factory and other AI Factories across Europe. The source does not attach a date or capacity figure to that extension. For developers, this is the practical detail to watch: access terms, queue priority and allocation rules on a shared public machine usually determine whether an accelerator programme is useful for training runs or mainly for experimentation.

Three Inception members were named AI Champions at the pilot's demo day. Qatium operates a water-network platform already used by more than 250 utilities in 16 countries, and is now in early production with a Neural Water Intelligence technology built on proprietary models trained on MareNostrum 5. Its stack includes NVIDIA accelerated computing, the cuGraph library for graph analytics, the PhysicsNeMo framework for physics-informed machine learning and Nemotron open models. Water networks are naturally graph-shaped — pipes, junctions, pumps and demand nodes — so graph tooling plus simulation-aware models is a sensible fit for operational decisions such as leak detection, pressure management and outage planning. A likely trade-off is that models trained on one utility's topology and sensor layout may need retraining before transferring to another.

OneCareAI, a BSC research spin-off, used NVIDIA NIM microservices to build technology intended to flag signs of hidden atrial fibrillation, an irregular heart rhythm, in ECGs that otherwise look normal. The stated aim is helping clinicians spot patients at risk of stroke before one occurs. NIM microservices are NVIDIA's packaged, containerized inference endpoints for models; using them shortens the path from research prototype to a deployable service. In practice, screening tools of this kind stand or fall on sensitivity and false-positive rates, and the source gives no performance figures.

Avatia built a voice-first, multi-agent system for pharmaceutical companies: agentic coworkers that support sales representatives and give doctors curated, transparent information about medicines, with the stated goal of getting the right treatment to the right patient sooner. It uses NIM microservices, the Nemotron 3 open model family and the open source NeMo Curator library, which handles data curation and filtering for training and evaluation. The word transparent is doing real work here — in regulated life-sciences settings, a system that answers clinicians needs traceable sourcing for its claims, and architecture choices around retrieval and logging matter as much as model quality. The source does not describe how that transparency is implemented.

Seven further startups round out the cohort: Agrikola AI, Elna Health, Lokutor, Montevive AI, Oppy, Pack2Earth and Up2Smart. Across them the source lists precision agriculture, medical imaging and knowledge access, speech services, enterprise security, sales intelligence and sustainable packaging. The stated applications include detecting breast cancer subtypes before biopsy, putting medical knowledge at doctors' fingertips faster, replacing petroleum-based plastic with biobased compostable materials and protecting crops with robots instead of chemicals.

Why it matters: startups in Spain and across Europe get a second route to sovereign supercomputing through a commercial portal they may already use, which could shorten procurement for early experiments and production pilots. For developers, the interesting signal is the stack pattern — accelerated computing, graph and physics libraries, NIM microservices and open model families — repeated across water, cardiology and pharma use cases. Whether the arrangement translates into sustained compute access once the MareNostrum 5 extension arrives is, by inference, the open question to track.

Read at NVIDIA Blog

Based on reporting from the original publisher. Visit the source for full context and later updates.

Publisher excerpt

The Barcelona Supercomputing Center (BSC)’s AI Factory Accelerator Programme, now in full operation following a pilot initiative, gives startups access to sovereign compute, technical support and training to support innovation across industries. Many of these companies are also members of the NVIDIA Inception program for cutting-edge startups — and as of today, access to some […]