A technology chain spanning models, datacentres and customer sites
OVHcloud’s November 2026 announcements present three connected initiatives. First, the provider says it is extending its sovereign-AI strategy from infrastructure into models and applications. Its stated path includes GPU services, AI Endpoints with access to more than 40 open-source models, an AI Lab, the acquired OVHai LLM and Gladia capabilities, and the forthcoming beta of OVHai Workspace. Second, OVHcloud argues that AI infrastructure should be evaluated by useful output per unit of energy, including tokens or samples per joule, alongside PUE and WUE. It reports a group-average PUE of around 1.24 and WUE of 0.34 litres per kWh for FY2025, while attributing further cooling reductions to direct-to-chip water cooling and AI-based control. Third, it has certified Scality RING for a defined configuration of OPCP Core, enabling documented deployment of S3-compatible object storage in customer-controlled and potentially disconnected datacentres. These are provider-published claims and product descriptions rather than independently validated results.
Sovereignty is becoming an architectural proposition
Taken together, the initiatives indicate that OVHcloud wants sovereignty to mean control across the technology stack rather than data residency alone. The strategic direction stretches from European datacentres, self-designed servers and cooling operations to on-premises provisioning, storage, encryption, inference services and provider-developed models. Scality’s certification also suggests that OVHcloud is trying to make OPCP an extensible foundation for regulated and sensitive environments instead of a closed infrastructure appliance. For Western European organisations, this positioning is relevant to public-sector, industrial, critical-infrastructure and defence workloads that may require local operation or an air-gapped design. Sustainability is being incorporated into the same proposition: OVHcloud’s argument is that infrastructure efficiency affects both AI economics and customers’ Scope 3 reporting. The emerging theme is therefore a vertically coordinated European cloud and AI stack, with selected partners filling specialised layers while OVHcloud retains control of the underlying operational framework.
Defined compatibility and efficiency claims require careful scrutiny
The proposition has important boundaries. Scality RING’s certification covers compatibility with a specified OPCP Core hardware, network and system configuration; OVHcloud explicitly says it is not a regulatory qualification or certification of the complete platform. Licences remain a separate commercial matter, Cloud Store integration is not yet available, and deployment, updates and support for disconnected environments must be designed for the chosen architecture. The sustainability comparison also mixes OVHcloud figures with external benchmarks from different countries and years, so buyers should not treat the table as a like-for-like assessment of every site or workload. Tokens per joule and samples per joule are useful concepts, but the article provides no workload-level results that would allow buyers to compare OVHcloud services directly with alternatives. Similarly, the sovereign-AI roadmap contains announced, acquired, previewed and beta-stage elements. Claims concerning European hosting, legal jurisdiction, predictable costs, model sovereignty and native encryption therefore need to be tested against contracts, service locations, operating responsibilities, model terms, support arrangements and the actual availability of each component.
Decision guide and source: OVHcloud’s sovereign-AI roadmap
For IT decision-makers in Western Europe, OVHcloud merits attention where data control, European operation, on-premises deployment and AI infrastructure efficiency are joint requirements. Its direction is more coherent than a standalone AI-service launch: the provider is connecting models and inference to physical infrastructure, cooling, storage and deployment in customer datacentres. That coherence does not remove the need for workload-specific validation. Procurement teams should request site-level PUE and WUE data, relevant tokens- or samples-per-joule measurements, full cost models, service maturity details, contractual data-location commitments and a responsibility matrix for OPCP and Scality support. They should also verify whether required AI components are generally available rather than announced or in beta. The underlying provider material is available in OVHcloud’s account of its sovereign-AI strategy, its discussion of output-based datacentre efficiency metrics, and the documented announcement of Scality RING certification for OPCP Core.