Europe Is Building 35 NVIDIA AI Supercomputers and the Race for Compute Is On

Europe has spent the last few years watching the United States and China pour money into AI infrastructure while debating regulation, sovereignty, and the ethics of it all. That conversation is not going away, but it is now happening alongside something more urgent: a genuine hardware build-out. NVIDIA announced at ISC High Performance 2026 that a record 35 AI supercomputers are currently in development across Europe, and the scale of what is being built deserves serious attention.

Thirty-five systems. That is not a handful of flagship projects in a few capital cities. That represents a broad geographic push across European research institutions, national labs, universities, and emerging AI factories. Each of these systems is built on NVIDIA's latest HPC and AI GPU architecture, making Europe collectively one of the largest markets for high-performance AI compute outside of hyperscale US cloud providers.

Why This Moment Matters for European AI

For years, European researchers and companies complained about a structural disadvantage: when you needed serious compute, you ended up renting it from Amazon, Microsoft, or Google, often on servers located outside the EU, which created both cost and data sovereignty problems.

The push to build sovereign AI infrastructure is not just about national pride. It is about ensuring that European researchers can train large models on sensitive data, that public institutions are not dependent on US corporations for critical scientific capability, and that European AI companies can compete on something closer to a level playing field.

The timing is also notable. The EU's AI Act is now fully in force, meaning organisations building high-capability AI systems in Europe need to comply with transparency, safety, and documentation requirements. Having compute infrastructure on European soil, under European jurisdiction, makes that compliance significantly more straightforward.

What These Systems Actually Look Like

NVIDIA has not published a full list of all 35 systems, but the ones that have been announced publicly give a sense of the ambition involved. Several are built around NVIDIA's Blackwell architecture GPUs, which represent a significant generational leap in AI training and inference performance. NVIDIA claims Blackwell delivers up to four times the training performance of the previous Hopper generation for large language model workloads.

Some of the systems are purpose-built AI factories, designed not just for traditional high-performance computing tasks like climate simulation and protein folding, but specifically for training and running AI models at scale. This reflects a shift in how research institutions are thinking about compute: not as a shared resource to run batch jobs on, but as persistent infrastructure for ongoing AI development.

Countries with confirmed or announced systems include Germany, France, Finland, the Netherlands, and several Central and Eastern European nations that are positioning themselves as emerging tech hubs.

The Geopolitical Dimension

It would be naive to look at this purely as a science story. The race to build AI compute is geopolitical, and Europe knows it. The European Commission has made sovereign AI infrastructure a priority under its AI Continent Action Plan, committing funding to reduce dependence on non-European cloud providers and ensure that EU institutions and companies have access to world-class compute.

At the same time, there is a real tension here. Every one of these 35 supercomputers relies on NVIDIA chips, which means European AI sovereignty is still deeply intertwined with a single American chipmaker. Until European chip initiatives like RISC-V-based European processors or the European Chips Act funded fabrication capacity mature, that dependency is not going away.

That said, having compute on European soil, operated by European institutions, and available to European researchers is a meaningfully different situation from renting time on a US cloud. The data stays where policymakers say it should. The capability is under European control.

What Comes Next

The 35 systems currently in development will not all come online at once. Some are already operational, others are in procurement or construction phases. Over the next 18 to 24 months, European researchers should have access to significantly more AI compute than they do today.

Whether that translates into European AI companies and research labs closing the gap with their US counterparts will depend on more than just hardware. Talent, funding, and organisational culture all matter. But you cannot build world-class AI without world-class compute, and for the first time in a long while, Europe is actually building it.