The starting point
Europe should be proud of EuroHPC because it turned a strategic weakness into physical public capability. High-performance computing once depended heavily on national facilities with very different scales and on access to machines elsewhere. Since 2018, the European High Performance Computing Joint Undertaking has pooled EU and national resources to procure a distributed family of supercomputers, support software and skills, and allocate computing time across borders. A researcher does not need their own country to own a top-tier machine before proposing work worthy of one.
The achievement is larger than a league table. EuroHPC had procured 12 supercomputers by September 2026, distributed among European hosting centres and designed for different workloads. The family includes systems such as LUMI in Finland, Leonardo in Italy, MareNostrum 5 in Spain and JUPITER in Germany, together with newer mid-range and specialised machines. Their value comes from a governed network: shared procurement, hosting agreements, access calls, national competence centres and user support rather than isolated cabinets competing for a headline.
JUPITER makes the scale visible. In November 2025 it became the first European system officially to cross the exascale threshold at 64-bit precision: one quintillion floating-point operations per second on the benchmark used for the TOP500 list. The system at Forschungszentrum Jülich is operated by the Jülich Supercomputing Centre and built around Eviden's BullSequana XH3000 architecture. EuroHPC lists a sustained performance of one exaflop and a measured peak above 1.2 exaflops.
An exaflop is not a promise that every program automatically runs a quintillion useful calculations each second. Real performance depends on the algorithm, precision, memory, communication, software and how well a workload can use thousands of processors together. JUPITER divides work between a GPU-accelerated Booster for highly parallel computing and a general-purpose Cluster module with high memory bandwidth. The architecture matters because climate, materials, medicine, engineering and AI do not all use the same computing pattern.
The Booster contains roughly 24,000 NVIDIA Grace Hopper superchips. EuroHPC says it can reach very high lower-precision performance for AI workloads, while the system provides around 20 petabytes of ultra-fast flash storage. The Cluster module introduces SiPearl's European Rhea1 processor within the same modular platform. That mix is an honest picture of European capacity: European system design, integration, public procurement and emerging processor technology working with indispensable global components.
JUPITER's intended uses reach from whole-Earth climate simulations and more detailed extreme-weather modelling to materials, energy systems, proteins, cells, brains, quantum-computer validation and large AI models. By mid-2026 Jülich reported more than 120 national and international projects applying for computing time. Early examples included a record-scale simulation of a 50-qubit quantum computer, roughly one-kilometre Earth-system climate modelling and a foundation model for analysing brain microarchitecture.
Those examples are computational achievements, not guaranteed policy or medical outcomes. A higher-resolution climate model still requires validation and careful interpretation. A digital heart or brain is not a patient and does not replace a clinical trial. A materials simulation narrows possibilities before laboratory and industrial testing. Supercomputing increases the questions Europe can ask and the detail with which it can test them; evidence, regulation and human expertise still decide what becomes useful.
How this works in practice
Access is part of the infrastructure. EuroHPC runs calls for scientific users, industry and public administrations, with modes ranging from smaller development allocations to extreme-scale projects. Proposals are checked for technical readiness and, where appropriate, ranked by peer review for excellence, innovation, impact and implementation quality. The current extreme-scale call includes JUPITER, LUMI, Leonardo and MareNostrum 5 and grants successful projects allocations for defined periods rather than selling the machines as ordinary cloud instances.
This European allocation model is worth celebrating because scarce capacity becomes a continental scientific instrument. It also has limits. Demand can exceed available hours, technically weak applications can waste resources, and a strong proposal may wait months between a cut-off and access. Commercially sensitive, personal or classified data need an appropriate system and legal basis. Training, porting code and expert support can be as important as the hardware, especially for smaller countries, public bodies and companies new to high-performance computing.
EuroHPC therefore funds a network of national competence centres and wider skills programmes. Their job is to help users understand which machine, software, access mode and expertise they need. Shared capacity without this human layer would mostly reward organisations already fluent in supercomputing. A federal approach becomes real when an Estonian company, Portuguese research group or Croatian public authority can find local guidance into a European resource rather than being told that a distant machine technically exists.
The funding structure is deliberately European. EuroHPC is an EU legal and funding entity based in Luxembourg, governed by the Commission and participating states with advice from scientific and industrial representatives. Its updated 2021–2027 budget is at least €8.2 billion: the EU contribution is matched by participating countries, with an additional target for private-member contributions. Procurement and grants follow open calls, while hosting countries co-finance and operate machines under the agreed model.
JUPITER itself is fully owned and half co-funded by EuroHPC. Germany's federal research ministry and North Rhine-Westphalia fund the remaining half equally. This is not Brussels buying a German computer or Germany donating a machine to Europe. It is joint ownership and financing attached to a European access mission, delivered by a host with decades of computing expertise. Similar combinations let national centres retain responsibility while European funding buys more scale and broader access than a fragmented set of national procurements.
Energy is the unavoidable physical test. A supercomputer turns vast electrical power into calculation and heat. JUPITER ranks among the world's most efficient systems per operation and uses direct warm-water cooling designed to feed waste heat into the Jülich campus heating network. In 2026 Jülich reported that it required less energy per computational operation than the machine ranked first globally at that time. Efficiency and heat reuse are achievements, but they do not make total electricity demand trivial or automatically carbon-free.
Responsible pride means measuring the whole system: electricity source, power-use effectiveness, cooling, utilisation, embodied hardware, replacement cycles and whether the computation is worth its cost. An idle efficient machine is wasteful; a fully used machine can still draw substantial power. The right comparison is not a slogan about green computing but useful scientific and industrial output per unit of energy, with transparent accounting and continuous pressure to improve.
What this means for a shared Europe
EuroHPC is also connecting classical supercomputers to a diverse European quantum programme. By June 2026 it had procured six quantum computers and co-funded two more, with several systems accessible through a common pilot call. The technologies include superconducting circuits, photonics, trapped ions, neutral atoms and annealing. This diversity lets users test algorithms and hybrid workflows without pretending that experimental quantum machines have already replaced exascale computing.
The CERN feature shows an earlier European lesson: difficult science needs permanent institutions, shared facilities, competitive access and distributed computing, not only brilliant experiments. EuroHPC extends that institutional idea beyond one laboratory. The Copernicus feature shows the data side: satellites and services produce evidence that increasingly needs large-scale processing. European science becomes stronger when instruments, data, networks and computing capacity are designed as parts of one ecosystem.
Technological sovereignty here is capability, not autarky. JUPITER's accelerated partition depends on NVIDIA technology from the United States, global semiconductor manufacturing and supply chains beyond Europe. The system also relies on European Eviden integration, Jülich design and operations, EuroHPC procurement, a growing European processor effort and software communities across the continent. Europe does not gain sovereignty by mislabelling foreign components; it gains it by owning the mission, maintaining integration skills, diversifying suppliers and developing parts of the stack it cannot responsibly leave to chance.
AI factories are the next institutional layer. They connect selected EuroHPC machines with data, expertise and services intended to help European researchers, startups and SMEs develop trustworthy AI. By June 2026 EuroHPC described 19 AI factories and 13 antennas under implementation. Those numbers are capacity being built, not proof that every promised service is mature or that Europe has closed the commercial AI gap. Judge them by accessible compute, useful support, credible users and applications that reach deployment.
Security and openness must also coexist. Shared supercomputers can support public research, industrial innovation and administrations, yet not every dataset or result should be public. Access conditions, export controls, intellectual-property terms, cybersecurity, data location and research-integrity rules matter. A European public infrastructure should make ordinary scientific access simpler while offering appropriately governed routes for sensitive work and refusing projects that cannot protect people, systems or lawful restrictions.
For Federal Europa, EuroHPC is federation expressed as compute time. Countries host machines, train experts and choose investments; European institutions create common procurement, ownership, access and long-term strategy. A laboratory in a smaller member state can reach infrastructure that no small national budget could reproduce. Host centres gain scale and specialisation without turning every other country into a passive customer. That is subsidiarity with cables, cooling pipes and peer-reviewed allocations.
Europe built a top-tier supercomputing ecosystem in less than a decade, crossed the exascale threshold with JUPITER, opened high-end machines across borders and linked them to skills, AI and experimental quantum capacity. It still needs more European components, easier access, sustainable power, stronger software and evidence that public investment produces broad results. Those are the next tests of a real achievement—not reasons to diminish it. Europe now possesses shared computing power that it chose, funds, governs and can improve together.