According to the latest memory industry research from TrendForce, major cloud service providers (CSPs) are projected to see capital expenditures surge by 98% year-over-year in 2026, followed by another 50% growth in 2027, as they accelerate AI infrastructure development. This substantial increase is largely attributed to sharp rises in memory contract prices and robust growth in procurement demand. TrendForce estimates that the combined share of DRAM and NAND Flash within CSP capital expenditures will expand from 47% in 2026 to 68% by 2027.

TrendForce notes that memory contract prices have seen significant growth since the second half of 2025, driving a notable increase in memory's proportion of CSP capital spending. Taking Server DRAM, a key procurement item for CSPs, as an example, contract prices accumulated a rise of approximately 64% in the second half of 2025, with projections indicating a cumulative increase of around 270% in 2026. NAND Flash contract prices are following a similar trajectory, with Enterprise SSD prices climbing about 35% in the second half of 2025 and an estimated cumulative gain of 235% expected in 2026. Although certain long-term agreements (LTAs) signed from the second quarter of 2026 onward impose price ceilings that may temper growth, considering that HBM contract prices could still rise by 70-140% in 2027, memory contract prices are expected to generally remain at elevated levels. This is anticipated to continue serving as a key factor boosting memory's share of CSP capital expenditures.

Beyond pricing, CSP memory bit demand is also growing substantially, prompting manufacturers to channel limited supply into Server-related applications. TrendForce projects that HBM and RDIMM combined will account for 51% of DRAM supply bits in 2026. In 2027, driven by process node transitions and capacity ramp-ups at some new fabs in the second half of the year, Server DRAM and HBM supply bit growth is expected to reach 27%.

TrendForce points out that rising memory contract prices, notable HBM price increases, and enhanced bit supply will collectively push memory's share of CSP capital expenditure to 68% in 2027, which will have two major implications for the AI ecosystem. First, high memory costs provide server and AI chip manufacturers such as NVIDIA with reasonable justification for raising product prices, and CSP clients may subsequently increase their capital budgets to secure the required volume of AI chips. Second, CSPs will become more aggressive in adjusting AI system memory architectures, including reducing memory capacity configurations, to cope with high DRAM and NAND Flash costs or supply quota shortfalls, thereby meeting AI chip or server shipment targets. These memory architecture adjustments may include, but are not limited to, modifying RDIMM specifications, adjusting HBM configurations for future AI chips, and exploring AI ASIC designs that hard-code model architectures into chips.