Cloud costs: OVHcloud raises prices by up to 87% over AI demand — what to check before October 1

Since September 1, 2026, OVHcloud has applied a new price list to its dedicated (Bare Metal) servers: an average of +28% on the 2024 generation, +51% on the 2026 generation, and up to +87% on the “Game” range. OVHcloud explains the increase by the soaring cost of memory (RAM) and storage, driven by global AI compute demand: the price the French host pays for its RAM is said to have multiplied sixfold between June 2025 and June 2026, and could reach ninefold by September, possibly twelvefold by early 2027 if the trend continues. A second wave hits, from October 1, 2026, servers already in production on the Advance, Scale, Game and High Grade ranges, with increases of 20% to 40% on memory and storage options — while a full pass-through of the cost increase would have required +127%, according to figures the company published.

What actually changes

• →New dedicated server orders since September 1, 2026: +28% on average for the 2024 generation, +51% for the 2026 generation, up to +87% on the “Game” range and +74% on some “Advance” configurations.
• →Example reported by trade press: a Scale-A9 server (dual AMD EPYC Turin, 128GB DDR5 RAM, two 1.92TB SSDs) goes from €1,349.99 to €2,299.99 per month — a 70% increase.
• →Servers already in production (Advance, Scale, Game, High Grade): memory and storage option adjustments from October 1, 2026, of +20% to +40% depending on generation.
• →Entry-level ranges (Kimsufi, So You Start, Rise) are not affected by these increases.
• →OVHcloud says it is not passing on the full cost increase to customers — the actual price rise stays below the increase in its own component costs.

What it means for your SME

If your SME hosts its ERP, e-commerce site, backup servers or a self-hosted AI model at OVHcloud on one of these ranges, your September or October bill will differ from August’s — with no change in your own usage. But the stakes go beyond a single invoice: the underlying cause of this increase, a memory and component shortage tied to AI infrastructure demand, affects every hosting provider, not just OVHcloud, and is expected to keep pushing cloud and hardware prices up through 2027. For an SME weighing whether to self-host an open-source AI model rather than use an API, this is a clear signal: the cost of the GPU and memory hardware required is now volatile and trending upward, while the price of an AI agent built on a managed API (Anthropic, Mistral, OpenAI…) stays set by usage rather than underlying hardware — an API provider absorbs that volatility in its own P&L, not yours.

Concrete steps

• →Identify, in your OVHcloud contract (or any similar host’s), which server ranges and generations are affected, and the exact effective date: September 1 for a new order, October 1 for a server already in production.
• →Ask for an updated quote before your contract renews, rather than discovering the increase on the next invoice.
• →For any upcoming AI project, compare the total cost of a self-hosted GPU server (exposed to the RAM/GPU shortage) against an agent built on a managed API, billed by usage.
• →Set up vendor monitoring — a monitoring agent can automatically track your cloud providers’ pricing announcements and alert you before an increase takes effect, rather than at invoice time.
• →Set aside an IT budget margin for 2027: the memory shortage is expected to persist, with a cost factor that could reach 12x on OVHcloud’s own projections.

Dependence on a single infrastructure provider is not a new risk, but this episode makes it visible. A landmark study published in the Journal of Cloud Computing by Opara-Martins, Sahandi and Tian, Critical analysis of vendor lock-in and its impact on cloud computing migration: a business perspective, surveyed 114 organizations that had migrated to the cloud and found that migration increases, rather than reduces, exposure to vendor lock-in — businesses that document their data formats, exit clauses and alternatives before signing are far better positioned to absorb an imposed price increase than those who discover their dependency at invoice time.