High-performance computing in Europe: how to distinguish investment, projects and operational capacity

The expression advanced computing can encompass several technologies, but high-performance computing, known by its English acronym HPC, has a prominent place in European digital-infrastructure policy. It does not simply mean a fast computer: it describes systems designed to perform intensive calculations and workloads that may need to process large datasets or distribute operations across many components. The European Commission presents investment in supercomputing as part of the Union’s digital strategy, although that policy direction alone does not establish how much capacity is installed or who can use it (European Commission: advanced computing).

The distinction matters because digital infrastructure is not defined solely by the advertised speed of its machines. Its purpose, its current stage and the conditions under which the service is provided also matter. So, when you encounter a reference to advanced computing, ask whether it describes a policy priority, a particular system or capacity available to users. Each description answers a different question and cannot be used interchangeably as proof of deployment.

EuroHPC JU—the European High Performance Computing Joint Undertaking—is one of the central instruments for coordinating resources from the Union, participating countries and private partners. Its stated objectives include developing pan-European infrastructure and strengthening technological capabilities. These statements should be read as institutional objectives, not as an up-to-date inventory of systems already in service. Moreover, “advanced computing” is not synonymous with HPC alone: AI factories and quantum computing feature in the programme’s activities, but they have different needs, uses and levels of maturity. Grouping them under one label may help describe a strategy; it does not make their capacity or results directly comparable.

To assess an initiative, follow its administrative and technical progress rather than treating every announcement as if it described an installation ready to use. A strategy priority expresses intent; a call or funding decision allocates resources under certain conditions; signing a contract formalises a purchase; and installation and testing precede regular service. Even when a system is operational, that does not mean every interested party has immediate access, or that all its capacity is available for any task.

This sequence makes it possible to identify precisely what has been confirmed and what remains outstanding. A mention in a strategy does not establish that funds have been committed; funding, in turn, does not prove that procurement has been completed. Similarly, a contract confirms an acquisition in progress but does not automatically establish when installation and testing will be finished. And confirmation that a system works does not, by itself, determine who may apply to use it, under what criteria or for which kinds of work.

A practical verification sequence is as follows:

• Objective: What need does the initiative say it will address, and over what timeframe?
• Funding: Has a budget or contribution been approved, and who is providing it?
• Procurement: Is there a published procedure and a signed contract?
• Installation and commissioning: Does the responsible organisation confirm that the system works, and under what conditions?
• Access: Is there a current policy, calls and criteria for obtaining computing time?

This distinction is illustrated by EuroHPC’s announcement of the contract for the MareNostrum-Ona quantum computer, planned for Spain: the signature documents a specific procurement step, but on its own it does not confirm that the system has been installed, validated and made accessible to users (EuroHPC JU, 28 January 2025). The example also shows why the correct technology category must be preserved: a quantum computer should not be counted as if it were a conventional HPC supercomputer. It is therefore not enough to add up announcements under a general label: record both the type of technology and the verifiable status of each initiative.

Computing power is often expressed in floating-point operations per second, or FLOPS. This is useful for describing certain calculations, but it does not, by itself, sum up the performance an organisation will achieve on a real workload. Architecture, memory, storage, communication between nodes, available software, energy efficiency and the system’s suitability for the specific task also matter. A theoretical peak figure should not be confused with performance measured in a benchmark, and neither value automatically corresponds to computing hours that a user can request.

The distinction between capacity and practical performance is particularly important when comparing systems with different characteristics. A measurement may be relevant to one task and less representative of another; an isolated figure therefore cannot predict the outcome of every application. Nor does it show how much computing time remains available for external requests. Answering that requires information about the system’s operation and access rules, not just a power specification.

Rankings provide an additional point of reference, with limitations. In November 2024, EuroHPC reported the positions of some of its systems in the TOP500 and Green500 lists. The first ranking focuses on performance achieved in a benchmark; the second ranks systems by energy efficiency according to its methodology. These are comparable indicators within those frameworks, not a comprehensive assessment of availability, total cost, ease of access or performance for every application. A careful reading should note the publication date, the system’s exact name, the version or module included and the metric used, and should avoid treating a position in one edition as permanent (EuroHPC JU, November 2024 lists). A ranking measures one aspect; it does not replace operational status or access policy.

Installed capacity becomes valuable to researchers and businesses when there is a clear way to use it. EuroHPC publishes an access policy and frequently asked questions that help explain how applications are handled, although the page included in the research is provided through machine translation. To interpret detailed requirements, deadlines or conditions, it is preferable to consult the original document and its latest version. It is also worth distinguishing access granted from access requested: an open call does not prove that a particular organisation has received resources or is using them (EuroHPC JU access policy).

These stages can be described as a chain: a system exists, a call is published, an application is submitted and resources are ultimately allocated. These are different facts. When assessing an infrastructure’s usefulness, establish which one supports each claim. The policy and frequently asked questions help locate the process rules, but they do not replace confirmation of a specific allocation and do not, on their own, report how much the system has been used.

AI factories illustrate another dimension of infrastructure. EuroHPC’s Spain page describes an initiative associated with AI services and computing infrastructure enabled for that activity. This description documents the announced purpose, but it is not enough to quantify capacity already in service or results for businesses or research centres (EuroHPC JU: Spain). Assessing impact would require published indicators with clear definitions: applications received and served, users and sectors, hours actually allocated, availability, workload types, verifiable results and costs. An institutional account of potential benefits cannot replace those measurements. For indicators to support comparisons between periods or initiatives, it is also necessary to know what each figure counts and which interval it covers; otherwise, the data may describe different activities and not be directly comparable.

The available documentation supports a limited conclusion: the European Union has established an institutional structure to promote supercomputing and other forms of advanced computing, and EuroHPC reports on procurement, systems, access mechanisms and related initiatives. It also makes it possible to identify specific milestones, such as the signing of a contract, or consult the reported position of particular systems in rankings for a given date. To turn these milestones into an assessment of real capacity, commissioning, access conditions and the latest operational figures must be checked separately.

This way of reading the evidence does not diminish the value of announcements, contracts or rankings; it defines what evidence each one provides. A strategy helps identify priorities, a contract confirms a procurement step, and a ranking offers a comparison based on a particular test and methodology. To claim that infrastructure is available and usable, the question changes: confirmation is needed about its operation, the rules governing access and the capacity actually allocated. Keeping these dimensions separate makes a project’s status easier to understand without turning a partial milestone into a broader conclusion.

It would not be rigorous to infer from the strategy that Europe has already achieved technological autonomy, that an investment has produced a specific scientific result, or that all announced infrastructure is available. The Commission’s page sets out the policy framework, while EuroHPC news and fact sheets describe milestones or programmes from the responsible body’s perspective. Assessing outcomes requires current, comparable data, as well as independent sources when evaluating impact or efficiency. The best approach separates what is proposed, what is contracted and what can already be used. That discipline avoids both overlooking real infrastructure and presenting plans and announcements as fully deployed capacity.