Cloud Infrastructure
OVHcloud's Price Hikes Expose Infrastructure Vulnerability as AI Demand Diverts Component Supply
European cloud provider faces steep price increases through 2028 as AI-driven demand for specialized memory starves broader infrastructure, while hyperscalers cushion themselves through long-term contracts and chip design.
InfoQ
OVHcloud announced significant price increases rolling out this autumn, with the sharpest hits—up to 87 percent for recent gaming servers—reflecting a broader crisis in component availability rather than any change to the company's own operations. The root cause: global RAM manufacturers have redirected fabrication toward high-bandwidth memory for AI accelerators, leaving standard server components scarce and costly.
The Cost Cascade#
Founder Octave Klaba detailed the situation on X, revealing the trajectory of OVHcloud's purchasing costs indexed from June 2025. Memory prices have already climbed to 6x their baseline, with projections reaching 9x by September 2026 and 12x by early 2027. NVMe drives sit at 7x, while standard hard drives at 3.5x show less pain. CPU, motherboard, and network card costs are expected to rise 15 to 20 percent.
The increases land unevenly across OVHcloud's catalog:
- Gaming servers (2026 editions): 87 percent increase
- Other recent servers: 40 to 59 percent increases
- 2024-era equipment: Three to six times smaller increases at renewal
- Older ranges (Kimsufi, Rise, earlier Advance/Scale): unchanged
From October, OVHcloud is restructuring charges rather than simply raising them—storage and IP addresses become separate line items on Gen3 instances, charged hourly. The company is also discontinuing one-, six-, and 24-month savings plans, keeping only 12- and 36-month options to lock pricing.
Why AI Is Repricing Everything Else#
Three major RAM suppliers have reallocated their production toward high-bandwidth memory that feeds GPUs at better margins. This squeeze has nothing to do with OVHcloud's demand; it reflects the asymmetric purchasing power of AI hyperscalers. Klaba was blunt about the timeline: "The situation is exceptional and will last until 2028."
Absorbing these costs was not viable. Klaba explained: "The main concern we want to avoid is running out of parts and not being able to deliver the Cloud to you." OVHcloud, like other non-hyperscaler providers, buys components month to month with no guaranteed pricing and no visibility into customer demand.
The Hyperscaler Moat#
Amazon Web Services faces the same market conditions but has barely moved prices. AWS raised its EC2 Capacity Blocks for ML—a reserved GPU product—by roughly 20 percent in July after a 15 percent increase in January, leaving most of the catalog untouched, including instances running Trainium, its own custom accelerator. The company announced the change only after it appeared on the pricing page.

