Is sovereign compute Europe’s best shot at breaking the AI monopoly?

Why is Europe betting on sovereign compute to compete with U.S. and China in foundational AI infrastructure?

Europe has formally entered the exascale era with the commissioning of JUPITER, a supercomputer capable of exceeding one quintillion operations per second. Located at the Jülich Supercomputing Centre in Germany, this system is the most powerful ever deployed in Europe and the fourth most powerful worldwide. More importantly, it is the cornerstone of a much larger ambition: to establish sovereign control over the infrastructure required to build, train, and deploy frontier artificial intelligence.

Funded through a €500 million public-private collaboration between the European Union and Germany under the EuroHPC Joint Undertaking, JUPITER embodies the EU’s push to ensure that the next generation of large-scale AI models—whether in language, science, health, or defense—are built on infrastructure owned and operated within its borders. It reflects a new vision for Europe’s digital autonomy, one that goes far beyond regulation and into the domain of compute capability itself.

At the inauguration ceremony on September 5, 2025, German Chancellor Friedrich Merz described JUPITER as a generational infrastructure asset that would allow Europe to “catch up and keep pace” in the global AI arms race. The machine is powered by 24,000 NVIDIA GH200 Grace Hopper Superchips and employs Eviden’s patented Direct Liquid Cooling architecture. This not only places it at the top of the Green500 list for energy efficiency but also positions it as the central node in Europe’s sovereign AI infrastructure roadmap.

What are the key advantages of sovereign compute infrastructure for the European Union’s AI agenda?

The push for sovereign compute rests on the argument that Europe can only lead in AI innovation if it owns and controls the hardware infrastructure on which AI models are trained. For policymakers, research institutions, and emerging AI startups, control over compute is no longer optional—it is foundational.

One of the most important benefits is data localization. JUPITER and its upcoming companion systems allow AI developers to train models using sensitive datasets that are kept entirely within European jurisdiction. This reduces compliance risk under the EU’s General Data Protection Regulation and avoids the need to transfer data to foreign clouds, which can pose privacy, security, and geopolitical risks.

The infrastructure also supports access democratization. Unlike commercial hyperscalers where access to compute is often cost-prohibitive or skewed in favor of large corporations, JUPITER allocates at least half of its compute time to public research institutions and European startups through the EuroHPC resource allocation mechanism. This policy gives universities, small labs, and deep-tech companies a credible chance to build competitive models and test scientific hypotheses without being constrained by commercial pricing or availability.

Another crucial advantage is sustainability. JUPITER runs entirely on renewable energy and recycles its waste heat to serve the heating needs of nearby buildings. With its number one ranking on the Green500 list, it sets a benchmark for energy-efficient AI compute globally, aligning with the EU’s goal of a climate-neutral digital economy.

Finally, the modular design of the infrastructure, based on Eviden’s Modular Data Center architecture, allows for scalable expansion. This enables additional sovereign nodes to be built across Europe using interchangeable components, ensuring both technical flexibility and cost efficiency as demand increases.

How do JUPITER and Europe’s AI Gigafactories fit into the long-term sovereign AI strategy?

JUPITER is the first operational node in a larger federated system that includes AI Factories and AI Gigafactories, a network of large-scale AI compute hubs being built across EU member states. These facilities are designed to handle the growing demands of generative AI, digital twin modeling, climate simulations, and public-sector decision support systems. Each Gigafactory will be capable of training foundation models with hundreds of trillions of parameters and will serve as a high-trust, high-performance alternative to centralized U.S. and Chinese cloud platforms.

The scale of interest in this initiative has been staggering. By mid-2025, the European Commission had received 76 expressions of interest from 16 member states seeking to host AI Gigafactories. These proposals range from national compute clusters to sector-specific training centers focused on healthcare, mobility, or climate. All are linked through the EuroHPC Joint Undertaking and governed by shared standards for access, energy efficiency, and data security.

At a systems level, these compute hubs are not intended to operate in isolation. The EU’s broader objective is to create a distributed but interoperable compute fabric across the continent. This would allow AI models to be trained collaboratively across borders while ensuring that sensitive data remains within national boundaries. Such a federated training model would serve as a European alternative to the centralized AI stacks currently dominated by U.S. tech giants.

What challenges threaten Europe’s ability to scale sovereign compute as a viable alternative to global cloud platforms?

Despite its progress, Europe’s sovereign compute model faces multiple systemic constraints. One of the most immediate is hardware dependency. Although JUPITER is controlled and managed in Europe, it is built almost entirely on NVIDIA hardware. The GH200 Grace Hopper Superchips that power the system are designed and manufactured in the United States. This raises concerns about supply chain vulnerability, especially in the context of escalating U.S.–China tech tensions and increasing scrutiny of semiconductor exports.

Europe also lags significantly in foundational AI model development. In 2024, European institutions contributed just three notable large language models, compared to 40 from the United States and 15 from China. While initiatives like OpenGPT-X and Aleph Alpha are trying to close this gap, the lack of large-scale pretraining datasets and training expertise remains a bottleneck. Without an expanded ecosystem of researchers, data engineers, and ML ops teams, sovereign compute risks being underutilized.

Another issue is scale. Even with JUPITER’s exascale compute power, Europe remains behind the aggregate AI compute capacity of the major U.S. cloud providers. Amazon Web Services, Google, and Microsoft each operate more exaflops of compute dedicated to AI training than all EuroHPC systems combined. Without further investment, Europe will struggle to offer sufficient compute availability to meet demand from emerging foundation model developers.

Finally, Europe’s talent pipeline in machine learning and AI engineering is still developing. Many top researchers migrate to the U.S. for better funding, career progression, and access to commercial-grade infrastructure. The European AI strategy must therefore go beyond hardware and include aggressive talent retention measures—ranging from competitive grants to public-sector fellowships and startup incubation programs.

Could sovereign compute infrastructure unlock a new generation of open, ethical, and multilingual AI models?

Despite the challenges, the strategic direction of Europe’s AI policy suggests that sovereign compute could enable a new class of AI models built on European values. These would include open-source large language models, multilingual transformers tuned for European languages, and domain-specific models for energy, finance, agriculture, and healthcare.

One example is OpenGPT-X, a multilingual open-source LLM being trained on JUPITER with a focus on high-quality German-language outputs. Similar initiatives are underway in France and Finland, where national labs are fine-tuning models for climate forecasting, digital twins, and scientific discovery.

The ability to train such models within sovereign infrastructure is a game-changer. It means that datasets, training runs, and model weights can all be audited, documented, and governed under EU law. It also opens the door to secure inference APIs hosted entirely within EU territory, enabling AI services for healthcare, transportation, and public administration without sending data overseas.

If executed at scale, this strategy could allow Europe to define a third path between the centralized, private AI platforms of the U.S. and the state-controlled AI ecosystem of China. It would be a vision of AI that is open but secure, powerful but green, and globally competitive while remaining accountable to democratic values.

What is the outlook for Europe’s sovereign compute strategy and what role will JUPITER play going forward?

JUPITER is the opening act in a decades-long play to establish a European AI infrastructure built on openness, energy efficiency, and digital sovereignty. The EuroHPC Joint Undertaking is expected to finalize several AI Gigafactory locations in the first half of 2026, with construction and onboarding to follow throughout 2027.

In the meantime, JUPITER will continue operating as the primary exascale compute hub for high-impact applications. These include kilometer-scale climate simulations using the ICON model, neuron-level brain modeling with Arbor, and AI model training across natural language, genomics, and material science.

Startups, universities, and government institutions will be able to access JUPITER and its successors through structured allocation programs. These programs will prioritize public interest use cases such as foundational model research, vertical AI for regulated industries, and training under-resourced languages.

As the broader AI Gigafactory network comes online, Europe will not only expand its compute footprint but also strengthen its geopolitical and economic position in the global AI race. For the first time in two decades, Europe is not merely reacting to shifts in U.S. and Chinese tech leadership. With JUPITER and sovereign compute, it is actively redrawing the map.