35 supercomputers in a single year: Europe steps on the gas
At the ISC High Performance 2026 conference in Hamburg, NVIDIA announced it is delivering infrastructure for a record 35 AI supercomputers across Europe. The machines powered by Blackwell and Hopper architectures represent the largest single-year expansion of supercomputing capacity in the continent's history.
"AI is the new instrument of science and Europe is building the infrastructure to put it in the hands of millions of researchers," said Jensen Huang, founder and CEO of NVIDIA. And the numbers back him up — a total of 800 AI exaflops deployed or announced over the last year means Europe is no longer reliant on data centers in North America or Asia for its AI infrastructure.
For perspective: one exaflop equals a billion billion (10¹⁸) operations per second. 800 exaflops thus represents performance capable of simulating climate models at resolutions undreamed of five years ago, or training language models the size of GPT-5 in days instead of months.
Where the new supercomputers will rise
Key projects include several giants:
MareNostrum 5 in Barcelona — the flagship of the European supercomputing ecosystem EuroHPC. After upgrading with the NVIDIA GB300 NVL72 and GB200 NVL4 architecture, it will offer approximately 20 exaflops for AI training and 33 exaflops for inference. The consortium of Spain, Portugal, and Turkey is thus giving European scientists a tool for climate modeling, biomedical research, or sustainable agriculture development.
IT4LIA at Italy's CINECA — with over 8,000 GPUs, it will deliver a staggering 82 exaflops for training and 164 exaflops for inference. This is a strategic project for European technological autonomy in the areas of agritech, cybersecurity, meteorology, and manufacturing.
Blue Swan in Bavaria — 1,000 GPUs at the University of Erlangen with 11 exaflops for training. Bavaria's Science Minister Markus Blume called the project the foundation of an "independent multimodal AI model for healthcare and robotics that fully meets European standards."
HammerHAI in Stuttgart — the first German AI factory with 850 GPUs and 8 exaflops of performance. It will focus on engineering simulations, large language model inference, and industrial research.
Mimer in Linköping, Sweden — with 400 GPUs and 4 exaflops, it will support Scandinavian research in life sciences, materials research, and autonomous systems.
All systems are connected by NVIDIA Quantum InfiniBand — a high-speed networking technology that allows thousands of GPUs to communicate as a single entity. Without it, even the most powerful chips would remain isolated islands.
AI saves the climate: Siemens Energy and hydrogen turbines
This is not just about scientific simulations in an abstract sense. Siemens Energy is using NVIDIA technology — specifically the Omniverse platform and CUDA-X libraries — to develop gas turbines capable of burning up to 100% hydrogen. Complex physics simulations involving extreme temperatures, fluid dynamics, and combustion processes have been accelerated by up to 77% thanks to accelerated computing.
In practice, this means faster development of low-carbon power plants that can replace coal-fired sources without losing supply stability. Decarbonizing the energy sector is precisely one of the main priorities that European supercomputers are targeting.
Quantum computers as teammates: CUDA-Q bridges worlds
A separate chapter is hybrid quantum-classical computing. The NVIDIA CUDA-Q platform — an open environment for programming both quantum and classical processors — is being integrated into several European centers:
- CINECA and Pasqal are installing a neutral-atom quantum processor (QPU) directly in the supercomputing center and connecting it via CUDA-Q with the Slurm workload manager for batch jobs.
- Fraunhofer FOKUS is linking CUDA-Q with the Eclipse Qrisp programming language, enabling researchers to write complex quantum algorithms and run them in a hybrid environment.
- Barcelona Supercomputing Center has deployed an analog quantum computer from startup Qilimanjaro Quantum Tech, which integrated CUDA-Q into its QiliSDK developer kit.
- Researchers from the Jülich Supercomputing Centre, in collaboration with NVIDIA, set a world record — they completely simulated a universal 50-qubit quantum computer on the JUPITER supercomputer. The JUQCS-50 simulator now enables testing of the largest quantum problems without the need for physical quantum hardware.
This approach is critical — quantum computers are not yet stable enough for standalone deployment, but as "accelerators" for specific computations within classical supercomputers, they can bring breakthroughs in materials research, drug discovery, or supply chain optimization.
A Czech footprint: Ostrava as a European AI hub
The Czech Republic is not lagging behind either. This spring, the Czech AI Factory in Ostrava launched — one of the nodes of the European EuroHPC network. Although its capacity does not match the Barcelona giant, it represents an important step for Czech research. Czech scientists and businesses are thus gaining access to infrastructure that would be financially out of reach for individual institutions. At the same time, it strengthens the Czech Republic's position in European AI projects — the country is transforming from a passive consumer of technology into an active player.
For Czech companies, this means the concrete ability to train their own language models with Czech language support, perform advanced simulations for industry, or test AI models for healthcare directly on European soil — without the need to send sensitive data across the ocean.
Sovereignty instead of dependency
The massive investments into European AI infrastructure are not just a technological story — they are a geopolitical signal. At a time when the United States is restricting access to the most powerful AI models and China is massively subsidizing its own chip industry, Europe has realized that without its own computing capacity, it will remain a second-tier player in the digital economy.
The exaflops in Barcelona, Stuttgart, or Bologna represent an insurance policy — a guarantee that European universities, startups, and industrial enterprises will have access to the infrastructure needed to develop their own AI solutions. And as the example of Siemens Energy or Bavaria's Blue Swan model shows, this is not some distant future — this infrastructure is already delivering concrete results today.
FAQ
What exactly does "exaflop" mean and why is it important?
An exaflop (exa = 10¹⁸, flop = floating-point operation) is a unit of computing performance — one exaflop means a billion billion calculations per second. For comparison: a typical gaming PC reaches single-digit teraflops (10¹²). 800 exaflops thus represents the performance of hundreds of thousands of ordinary computers working simultaneously. In practice, this enables simulating climate models, training large language models, or analyzing protein structures in a timeframe that would take years on ordinary hardware.
When will the new supercomputers be fully operational?
According to NVIDIA's official announcement, the systems are currently under construction. Some — such as MareNostrum 5 or Blue Swan — are already undergoing upgrades to the Blackwell architecture. Full commissioning of most centers is expected during 2026 and early 2027. Individual timelines vary by specific center, but the key point is that 800 AI exaflops have already been deployed or officially announced.
Can Czech research also run on European supercomputers?
Yes. The European EuroHPC network, which includes the Czech AI Factory in Ostrava, is open to researchers from all EU member states. Czech scientists can apply for computing time on any European supercomputer — from Barcelona through Stuttgart to Bologna. Allocation is handled through standard grant mechanisms and is often free for academic research.
