NVIDIA Deploys 35 Supercomputers Across Europe

According to NVIDIA's press release (as carried by HPCWire and other outlets), 35 NVIDIA AI HPC supercomputers are in development across 23 European countries, representing roughly 800 AI exaflops of deployed or announced capacity since last year. The announcement was presented at ISC High Performance 2026 and the rollout is described as spanning national supercomputing centers, AI factories and academic research institutions, with the aim of equipping more than 3 million researchers (sources: NVIDIA press release; HPCWire; Wccftech).

Per NVIDIA's announcement, the new systems are built on NVIDIA platforms including Blackwell and Hopper, networked by NVIDIA Quantum InfiniBand/Quantum-X800 technology and running NVIDIA software such as CUDA-X, NIM microservices and NVIDIA AI Enterprise (sources: NVIDIA press release; HPCWire). Reported examples with capacity figures include Barcelona Supercomputing Center's MareNostrum5 AI upgrade, cited as delivering about 20 exaflops of AI training and 33 exaflops of AI inference performance for that installation, and BavariaAI's Blue Swan, reported to include 1,000 GPUs via GB200 NVL4 systems (sources: Wccftech; HPCWire).

Editorial analysis: Large-scale procurements that consolidate on a single vendor stack, including accelerator hardware, interconnects and ecosystem software, accelerate standardisation of tools and workflows across institutions. Observers have noted similar patterns in prior national and regional AI factory initiatives where a dominant hardware/software supplier accelerates deployment cadence but also concentrates dependency on one vendor.

Editorial analysis: For ML engineers and research teams, the announced capacity increases change available options for large-scale training and inference. Greater continental access to systems optimised for AI workloads can shorten iteration cycles for foundation models and scientific AI workloads, but practical access will depend on allocation policies, software containerisation, GPU-sharing mechanisms and data governance at each centre. Reported stack components such as NVIDIA Quantum InfiniBand and CUDA-X imply an emphasis on high-throughput, GPU-native workflows (source: NVIDIA press release).

Editorial analysis: Key indicators to monitor include published user-access policies from the named centres (for example, Barcelona Supercomputing Center and BavariaAI), published benchmarks or utilisation reports showing how the exaflops translate into usable throughput for model training, and announcements about multi-tenant software layers or open access programs. Observers should also track whether alternative vendor stacks or multi-vendor procurement choices emerge as the deployments move from announcement to commissioning.

HPCWire carried a quote from NVIDIA CEO Jensen Huang: "AI is the new instrument of science, and Europe is building the infrastructure to put it in the hands of millions of researchers." Named projects cited in coverage include Barcelona Supercomputing Center's MareNostrum5 AI upgrade, BavariaAI's Blue Swan, IT4I/IT4LIA, HLRS's HammerHAI and NAISS's Mimer EuroHPC AI Factory (sources: HPCWire; Wccftech; NVIDIA press release).

What is reported publicly is a set of announced deployments and technology stacks; details such as delivery timelines, exact allocation models, total installed GPU counts per site (beyond the examples reported) and contract terms were not published in full in the press materials and media coverage. Several outlets reproduced NVIDIA's forward-looking language; those statements are risk- and assumption-dependent as noted in NVIDIA's press materials (source: NVIDIA press release).

Overall, the announcement documents a major, vendor-led expansion of AI-optimised supercomputing capacity across Europe and sets clear near-term expectations for large-scale, GPU-accelerated scientific compute in the region (sources: NVIDIA press release; HPCWire; Wccftech).