The European Union has agreed a set of new rules and amendments intended to speed up the creation and operation of public-private AI supercomputing hubs , often referred to in EU documents as “AI factories” or large-scale AI-optimised supercomputing facilities. These measures expand the remit of the EuroHPC Joint Undertaking and set out legal and operational frameworks to open powerful compute resources to a broader set of public and private actors, including startups and SMEs.
Officials and industry stakeholders say the package is designed to improve Europe’s computational sovereignty, lower barriers for AI R&D inside the bloc, and coordinate investment in energy-efficient, large-scale compute infrastructure. The provisions were adopted through Council-level agreements and subsequent publication steps earlier in 2026 that formally amended the EuroHPC regulation.
Background and regulatory changes
In 2024 the Council adopted a regulation opening EU supercomputing capacity to broader AI development, and in late 2025, early 2026 lawmakers agreed amendments that expand that framework further to accelerate dedicated AI hubs. The updated rules embed AI-focused facilities into the EuroHPC joint undertaking, creating a clearer legal basis for public-private collaboration and cross-border infrastructure projects.
The formal amendment package was taken forward at the Council and published in the EU’s Official Journal in January 2026, a step that put the new rules into force and activated the governance options for creating AI factories and related programmes. This publication also clarified responsibilities for procurement, access rules, and the integration of quantum research activities into the EuroHPC remit.
Those regulatory changes reflect two complementary policy tracks: rule-making for safe and trustworthy AI (e.g., the EU AI Act) and infrastructure-building to ensure European researchers and companies can train and deploy advanced models without depending exclusively on non‑EU hyperscalers. The amendments thus represent both a technical and a strategic pivot for EU industrial policy.
What AI factories and supercomputing hubs will look like
“AI factories” are defined in EU materials as integrated facilities that combine AI‑optimised supercomputers, associated data centres, specialised software stacks, and services for model training and inference. They are intended to serve public research institutions, private firms, and collaborative projects, with tailored access conditions for smaller innovators.
Physical sites already being developed or selected across Europe include established supercomputing centres and new regional hubs. For example, Barcelona’s Supercomputing Center has been named as a host for one of the first AI factories, and several other centres in Italy, France, Finland and other member states are in planning or early deployment stages. The networked approach aims to cover diverse geographies and industrial clusters.
Technically, these hubs will emphasise energy efficiency, modular hardware tuned for both training and inference workloads, and federated interconnects so that compute can be pooled across sites. Projects such as NET4EXA and advances in European interconnect technology are intended to underpin these capabilities and avoid lock‑in to single-vendor or non‑European stacks.
Public-private partnership and funding mechanisms
The updated EuroHPC framework explicitly foresees public-private partnerships to operate and co-finance AI factories. Member states, the Commission and industry partners will pool resources under the joint undertaking model to build, host and run high‑performance AI infrastructure. This structure aims to leverage public funding while attracting private investment and operational expertise.
Alongside governance changes, the EU and legal commentators have highlighted new investment instruments designed to mobilise private capital for infrastructure expansion. Analyses and briefings point to mechanisms such as the proposed “InvestAI” vehicle that could target tens of billions in combined public and private investment to scale compute, cloud and data capacity across the bloc. Those investment plans are framed to help Europe close the gap in large-scale compute relative to other global players.
Procurement and access rules included in the amendments aim to strike a balance: prioritising openness and broad researcher access while permitting bespoke commercial terms for industry partners that contribute funding or specialized services. The regulatory text also sets expectations for transparency, fair access for SMEs, and coordination among national authorities to reduce fragmentation.
What this means for startups and small businesses
One of the central policy goals of the new rules is to lower barriers for startups and SMEs to access high‑end compute for AI training and experimentation. The regulation specifies ad‑hoc access conditions to ensure smaller innovators can use AI factories without the long lead times or costs that typically accompany exascale resources.
By providing subsidised or tiered access tiers, as well as technical support and shared data services, EU policymakers expect many more European scaleups to be able to train large models and prototype advanced applications domestically. This could accelerate product development cycles and stimulate regional AI ecosystems.
At the same time, stakeholders warn that access alone is not enough: startups also need affordable cloud integration, specialist talent, and regulatory clarity about data use and model risk. Complementary programmes such as European Digital Innovation Hubs (EDIHs) and skills initiatives are therefore being emphasised to make compute access effective in practice.
Sovereignty, competition and geopolitical implications
The push to expand AI supercomputing hubs is explicitly framed as an element of Europe’s digital and technological sovereignty. EU documents and analysts argue that having home‑grown large compute capacity reduces strategic dependence on non‑EU hyperscalers and strengthens the bloc’s bargaining and security posture.
Politically, the initiative is intended to make Europe more competitive in foundational model research and industrial AI applications, ranging from healthcare and climate modelling to industrial automation. By coupling regulatory guardrails (for safety, transparency and rights) with infrastructure, the EU hopes to offer an alternative model to both an unregulated open market and highly centralized, state‑led systems used elsewhere.
However, the move will also raise questions about industrial strategy, export controls, and cooperation with non‑EU research partners. Policymakers will need to balance openness for collaborative science against concerns about technology transfer and strategic competition, and the EuroHPC governance model will be a testing ground for those trade‑offs.
Challenges, standards and the road a
Deploying continent‑scale AI compute raises pressing challenges: energy consumption and sustainability, interoperability of hardware and software, workforce shortages in HPC and ML engineering, and the need for shared standards to allow federated operation among sites. Researchers and policy analysts have called for coordinated standards work and investments in green energy for data centres.
Technical initiatives such as European interconnect projects and investments in open‑stack software aim to reduce vendor lock‑in and improve cross‑site schedulers and data sharing. Yet building reliable, secure federations across multiple member states and commercial partners remains a complex engineering and governance task.
Finally, monitoring and evaluation mechanisms will be important: the EU must ensure that promised access levels, procurement transparency and sustainability commitments are met. Early implementation phases (site selection, procurement, first operational AI factories) in 2025, 2026 will be closely watched by both industry and civil society for their effects on innovation, competition and public interest goals.
In sum, the new EU rules mark a significant step to operationalize Europe’s ambition for compute sovereignty and to link regulatory ambition with material capacity. If implemented well, the public‑private AI hubs could become powerful engines for research and industrial innovation across the bloc.
Nevertheless, the ultimate impact will depend on execution: how quickly hubs come online, how access and pricing are managed for smaller innovators, and whether the energy and standards challenges can be met. Policymakers, industry and researchers will need to collaborate closely in the months a to translate the legal framework into functioning, equitable infrastructure.
