Despite billions invested in the European Union’s state-backed EuroHPC Joint Undertaking supercomputers, more than 80% of Europe’s top 20 funded AI startups currently train their models on American commercial clouds. Startups bypass the European machines because state infrastructure lacks the proprietary orchestration software and elasticity that commercial providers offer natively.
Where Do Europe’s Top AI Startups Actually Train Their Models?
Most leading European AI companies rely on American computing infrastructure rather than domestic supercomputers like the newly funded €387.8 million LUMI-AI machine in Finland. According to a review of cloud partnership announcements, more than 80% of Europe’s top 20 funded AI startups train and host their models on United States-based hyperscalers, large cloud service providers that offer massively scalable computing infrastructure.
High-profile European champions like France’s Mistral AI rely heavily on Microsoft Azure and Amazon Web Services (AWS) for enterprise model deployments. The hardware often sits in European data centres to comply with local data residency laws, but the software stack running on that hardware belongs to American firms.
Why Do AI Startups Reject State-Backed EuroHPC Compute?
Having thousands of raw graphics processing units (GPUs) is not enough to build a commercial frontier AI model. State-backed machines provide “bare metal” computing power, while building a commercial AI model requires proprietary enterprise orchestration tools, InfiniBand networking optimizations, and containerized environments.
Engineers repeatedly cite the absence of commercial-grade orchestration on state machines as a primary barrier. While EuroHPC systems utilise academic job schedulers designed for scientific workloads, venture-backed startups require the enterprise Kubernetes environments that commercial hyperscalers provide natively.
Adapting a bare-metal environment requires dedicated engineering time that startups cannot spare. Technical documentation shows EuroHPC systems excel at batch-processing scientific jobs, whereas commercial platforms like AWS offer the managed container services and immediate application programming interfaces (APIs) required for rapid AI model iteration.
How Long Does It Take to Access EuroHPC vs American Clouds?
Commercial AI development moves in weeks, while state-funded compute allocation moves in months. EuroHPC’s standard access allocation guidelines require structured application periods and peer-review processes that take months to approve.
In contrast, commercial cloud service level agreements guarantee compute provisioning in minutes. A startup using a commercial provider can provision thousands of GPUs, scale up for a training run, and spin the cluster down within days. Accessing a public supercomputer requires navigating complex grant pipelines. If a model training run fails on a public machine, the startup must re-enter the queue, but in a commercial cloud, they restart the run immediately.
Who Actually Uses the EuroHPC Supercomputer Network?
The EuroHPC supercomputers primarily serve universities, climate modellers, and state research laboratories. According to EuroHPC’s annual activity reports, the systems successfully fulfil their original mandate of supporting academic and state scientific research rather than venture-backed commercial entities.
The European Commission acknowledges this commercial gap. Officials recently launched a new AI Factories initiative, designed to create a commercial fast lane and prioritize compute access for European AI startups and small-to-medium enterprises.
Does Local Data Residency Equal Infrastructure Sovereignty?
Because commercial hyperscalers provide the only currently viable development environment, a significant portion of European venture capital immediately exits the continent to pay for American compute. The AI models are built by European engineers, and the venture capital is raised in Europe, but the infrastructure costs flow to the United States.
Hosting data in a European facility owned by an American corporation fulfils data residency regulations, but it does not achieve true infrastructure sovereignty.
Frequently Asked Questions
What is the LUMI-AI supercomputer?
LUMI-AI is a new €387.8 million artificial intelligence supercomputer funded by the EuroHPC Joint Undertaking and a consortium of European countries, scheduled to deploy in Finland in 2027.
Why do European AI startups use US clouds like AWS and Azure?
European AI startups use US commercial clouds because providers like AWS and Azure offer enterprise orchestration software, instant compute allocation, and managed container services that state-run supercomputers currently lack.
Is EuroHPC failing to support AI development?
EuroHPC successfully supports academic and state research, but its traditional grant-based allocation and academic software stack make it too slow for rapid, venture-backed commercial AI development.
What is the AI Factories initiative?
The AI Factories initiative is a new European Commission programme designed to prioritise supercomputer access for European AI startups and create a faster commercial lane for compute allocation.
