Cloud Capex Rewritten as Memory Costs Set to Consume Two-Thirds of Budgets

The fundamental cost structure of cloud computing is on the verge of a seismic shift, with the price of memory set to become the dominant factor in infrastructure spending. According to analysis from market-watcher TrendForce, the combined cost of DRAM and NAND flash memory is forecast to account for a staggering 68 per cent of cloud service provider (CSP) capital expenditure by 2027. This represents a dramatic increase from an already elevated 47 per cent this year. The price inflation driving this change is severe, with the firm projecting that server DRAM component prices will have grown by 270 per cent year-on-year by the end of 2026, while enterprise solid-state drive prices are expected to be up 235 per cent over the same period.

This surge is not happening in a vacuum. It is a direct consequence of the voracious appetite for artificial intelligence, which is cannibalising the global supply of high-performance memory. TrendForce estimates that high-bandwidth memory (HBM), essential for AI accelerators, and server-grade RDIMMs will together consume more than half of all DRAM bit supply this year. This prioritisation of production capacity for high-margin server applications is creating scarcity and driving up prices across the board, with knock-on effects already being seen in rising PC costs and falling smartphone shipments. For cloud providers, total capital expenditure is expected to nearly double in 2026 and grow a further 50 per cent in 2027, with soaring memory bills being a primary cause.

The implications for the finance function are profound and immediate. The long era of assuming deflationary cloud pricing, where compute becomes cheaper over time, appears to be over. Finance leaders must prepare for significant and sustained price increases from their cloud providers, a reality already previewed by European operator OVHcloud, which warned of price rises up to 87 per cent to cover its own rising memory costs. This is not a risk confined to those running large AI models; as CSPs pass on their higher underlying component costs, the price of all cloud services will inevitably rise. This creates a challenging inflationary cycle, as chip suppliers like Nvidia are also reported to be increasing prices, citing their own elevated component costs as justification.

For the CFO, this necessitates a fundamental reset of technology budgeting and forecasting. Financial models based on historical cloud pricing trends are now obsolete. It is imperative to re-forecast cloud spending with the assumption of significant inflation and to build this new reality into the financial framework of the entire organisation. The cost of inaction is high, as projects approved based on old assumptions could become rapidly unprofitable. Scrutiny of cloud contracts, active management of consumption, and a strategic review of which workloads are best suited to the public cloud are no longer just good practice; they are essential measures for financial survival in this new, high-cost compute environment.