The surge in memory prices is now directly impacting the capital expenditure decisions of Cloud Service Providers (CSPs), as clearly demonstrated by OVHcloud's recent announcement.
European cloud company OVHcloud revealed it is strategically front-loading its purchases of memory and storage for the second half of the year to combat what it calls “exceptional” cost inflation. Consequently, it has raised its full-year capital expenditure (CAPEX) guidance from 30-32% of revenue to 33-35%. This decision shows that the pressure to secure components, even at high prices, has translated into real cash deployment. The company did, however, emphasize its commitment to maintaining positive free cash flow, suggesting that this accelerated spending pace may not be sustained in the latter half of the year.
This event is part of a larger narrative: 'AI Infrastructure Memory Inflation'. The causal chain leading to this situation is clear and has unfolded over several months.
First, the immense demand for AI accelerators created a supply bottleneck for high-bandwidth memory (HBM). As memory manufacturers like Micron and SK hynix prioritized HBM production, which was fully booked for 2026 as early as late 2025, they reallocated manufacturing capacity away from general-purpose server memory like DDR5 and enterprise SSDs. This strategic shift by suppliers began to squeeze the availability of essential, non-HBM components.
Second, this supply reallocation triggered a rapid price escalation. Industry data from Q4 2025 and Q1 2026 showed dramatic price hikes, with NAND flash prices reportedly soaring by as much as 85-90% in a single quarter. For CSPs, memory is not optional; it's a critical component needed alongside GPUs to expand service capacity. The risk of being unable to secure memory became a major operational threat.
Third, CSPs were forced to react. To avoid a situation where expensive GPUs sit idle due to a lack of memory, companies like OVHcloud adopted a proactive procurement strategy. They began buying components earlier than planned and accepted higher CAPEX-to-revenue ratios as a necessary cost of doing business. This shift confirms that bargaining power has moved firmly into the hands of the three major memory suppliers, who are now setting the terms with pre-payments and long-term contracts.
- CAPEX (Capital Expenditure): Investments made by a company in physical assets, such as servers, networking equipment, and data centers, to maintain or expand its operations.
- CSP (Cloud Service Provider): A company that offers cloud computing services, such as data storage, computing power, and software, to other businesses or individuals. Examples include Amazon Web Services, Microsoft Azure, and OVHcloud.
- HBM (High-Bandwidth Memory): A high-performance type of computer memory used alongside GPUs to accelerate AI and high-performance computing tasks. Its high demand is a key driver of the current market dynamics.