OVHcloud is about to raise prices on some of its servers by as much as 87%, and the company’s CEO says memory costs deserve the blame, not the GPU shortage everyone’s already tired of hearing about. It’s everything else that goes into a server.
According to Octave Klaba, OVHcloud’s memory costs had already climbed sixfold by June and ninefold by September, with a twelvefold jump expected by next year. Hard drives cost 3.5 times more than before. NVMe storage is up sevenfold. Even CPUs, motherboards, and network cards have risen 15% to 20%. None of that is AI chip pricing. Hyperscalers buying massive amounts of AI infrastructure are also snapping up memory, storage, and networking gear at scale, and suppliers are filling those orders first, which drives up the cost of the ordinary hardware sitting around the chips.
New server pricing kicks in September 1, with renewals on existing systems following from October 1. The increases vary by hardware generation and product line. Newer 2026-generation equipment sees the steepest jumps, averaging 51% overall, with Game servers hitting that headline 87% figure. Older 2024-generation hardware gets a smaller average bump of 28%, which creates a genuinely odd incentive: older, less efficient servers are suddenly the cheaper option, even when newer hardware performs better.
OVHcloud is trying to soften the blow for existing customers specifically. Someone already running a system will see RAM costs rise 20% to 40%, compared to the 127% increase Klaba says would reflect true current market pricing. Existing contracts stay locked in until they expire, and customers can lock in current rates for up to 48 months by paying upfront before October 1, assuming they act in time.
The bigger implication reaches past OVHcloud itself. AI servers consume the same memory, storage, and networking components that ordinary cloud infrastructure needs, so companies running plain databases or web servers with zero AI involvement could still end up paying more, simply because the supply chain doesn’t distinguish between an AI workload and a normal one.
Whether this stays an OVHcloud-specific story or turns into a wider pattern across the cloud industry depends largely on whether component prices actually follow the trajectory Klaba is predicting.
