OVHcloud, Scaleway, Hetzner: the European AI compute market in 2026
The compute question that finally has European answers
For European companies building AI workloads at scale, the cloud compute conversation through 2023 was constrained by a difficult set of trade-offs. The cost-effective GPU capacity at scale lived inside the American hyperscalers. The data residency story for that capacity was either complicated or expensive. The smaller European providers had the data story but did not have the GPU capacity for serious training workloads or large inference deployments. A European AI company building a serious application had to choose which constraint to live with.
What changed through 2024 and 2025 was that three European providers reached a level of GPU capacity that made them viable for AI workloads they had not been viable for previously. OVHcloud, the French provider that had been the European data residency answer for traditional cloud workloads, scaled its GPU offering substantially with H100 capacity across multiple European regions. Scaleway, the French provider owned by Iliad, made the most explicit AI compute pitch with its 2024 H100 supercomputer announcement and its 2025 launch of inference-focused GPU instances. Hetzner, the German provider known for its cost-effective dedicated server business, expanded into AI workloads with a more measured approach that focused on smaller deployments at favourable per-hour economics.
The cumulative effect is that a European AI company building in 2026 has options the cohort of 2022 did not have, and the choice between them is now driven by architectural fit rather than by which provider can take the contract.
What the trade-offs actually look like
OVHcloud occupies the slot the European market has been asking for the longest. The company is publicly listed on Euronext Paris, the data residency posture is explicitly European, the GPU capacity is meaningful, and the broader cloud platform supports the surrounding workloads an AI company needs. The 2023 IPO and subsequent operational scaling produced a provider with the institutional shape that European mid-market and enterprise procurement teams find easier to clear than a privately held smaller competitor. The trade-off is that the developer ergonomics and the documentation quality are sometimes behind what the same engineering team has experienced on AWS or GCP. The product is improving year over year and the gap is narrowing.
Scaleway took the most explicit AI positioning bet. The 2024 announcement of the Jeroboam, Nabuchodonosor, and Salomon clusters made the company the most visible European provider for serious AI training workloads. The 2025 launch of the inference-focused GPU instances broadened the offering to companies that needed efficient inference rather than training scale. The product team's public statements on AI infrastructure are the most specific in the European market. The trade-off is that the company is smaller than OVHcloud, and the procurement diligence for a large enterprise contract takes longer than the same diligence for a US hyperscaler.
Hetzner sits in a different position. The company has not made an aggressive AI positioning push and has historically been the cost-effective choice for traditional workloads. The 2024 expansion into GPU instances was measured. The pricing is favourable. The trade-off is that the GPU capacity is smaller than at OVHcloud or Scaleway, and the company's traditional sales motion is less suited to large enterprise procurement conversations. For a European startup or mid-market team running smaller inference workloads on a cost-conscious budget, Hetzner is often the right answer. For a serious training run, the answer is one of the larger providers.
The Schrems II posture that Mistral and Aleph Alpha demonstrated
The structural argument for European AI compute is reinforced by the customer choices the largest European AI companies have made. Mistral, the Paris-headquartered model lab valued at €5.8B in its 2024 round, runs inference for its enterprise customers on a mix of providers that includes European data centres. Aleph Alpha, the Heidelberg-headquartered model lab whose PhariaAI product targets regulated European enterprise customers, has made data residency a core part of its product positioning and runs on European infrastructure by default. These choices are not marketing. They reflect the actual procurement conversations the companies are having with regulated European enterprises that cannot defend a US-headquartered cloud provider to their compliance function.
A real case from the 2025 deployment record. A French insurance company building an AI-powered claims processing system selected Scaleway for the inference workload after evaluating AWS Frankfurt, Azure Europe, and OVHcloud. The decision was made on a combination of the data residency posture, the GPU instance pricing for the specific workload shape, and the Scaleway team's willingness to engage on the AI Act compliance questions in a way the larger providers had treated as a generic compliance conversation rather than a specific architectural one.
What the next year requires
For European AI companies and AI-using mid-market companies choosing compute in 2026, the productive question is no longer "is this possible." It is "which architecture fits the workload shape and which provider's institutional shape clears the procurement conversation cleanly." The architectures are increasingly comparable in raw capability. The institutional fit is where the differences live, and the European providers have closed enough of the capability gap that the institutional fit is now the deciding factor for the customer base that cares about it.
The cohort of European AI companies that built explicitly on European infrastructure in 2024 and 2025 is producing the operational record that the next cohort will reference. The procurement conversations of 2026 are easier than the conversations of 2022 in a way that matters for the next phase of European AI deployment.