Aug. 25, 2026 Industry Insights
According to TrendForce's latest memory industry research, to accelerate the construction of AI infrastructure, the capital expenditure of major global cloud service providers (CSPs) is projected to increase by 98% year-over-year in 2026 and grow by another 50% in 2027. This substantial growth is partly attributable to the sharp surge in memory contract prices and the strong growth in procurement demand. TrendForce estimates that the combined share of DRAM and NAND Flash in CSPs' capital expenditure will expand from 47% in 2026 to 68% in 2027.
TrendForce notes that the significant increase in memory contract prices since the second half of 2025 has driven a substantial rise in the proportion of memory in CSPs' capital expenditure. Taking Server DRAM, a key procurement target for CSPs, as an example, contract prices accumulated an increase of approximately 64% in the second half of 2025 and are expected to surge by about 270% cumulatively in 2026. NAND Flash contract prices have shown a similar trend, with Enterprise SSD prices rising by about 35% cumulatively in the second half of 2025, and the cumulative increase is projected to reach 235% in 2026.
Although some long-term agreements (LTAs) signed since the second quarter of 2026 stipulate price ceilings and limit the room for growth, considering that HBM contract prices could still rise by 70-140% in 2027, memory contract prices are estimated to remain generally at high levels in 2027, continuing to be a crucial factor driving up the share of memory in CSPs' capital expenditure.
Beyond pricing, the memory bit demand from CSPs has also grown substantially, prompting original manufacturers to concentrate their limited supply on server-related applications. TrendForce estimates that in 2026, HBM and RDIMM combined will account for 51% of the DRAM supply bits. Driven by process node advancements and the capacity ramp-up of some new fabs in the second half of 2027, the growth rate of supply bits for Server DRAM and HBM will reach 27%.
TrendForce points out that the rising memory contract prices, significant price increases in HBM, coupled with the growth in bit supply, will drive the proportion of memory in CSPs' capital expenditure to 68% in 2027, bringing two major impacts to the AI industry ecosystem. First, the high memory costs provide a reasonable rationale for server and AI chip manufacturers like NVIDIA to raise product prices, while CSP clients may further increase their capital expenditure to meet the procurement volume of AI chips. Second, CSPs will more actively adjust the memory architecture of AI systems by reducing the equipped capacity to cope with the high costs or insufficient supply quotas of DRAM and NAND Flash, in order to achieve the shipment targets of AI chips or servers.
The aforementioned solutions for adjusting memory architecture include, but are not limited to, adjusting RDIMM specifications, modifying the HBM integration specifications for future AI chips, and attempting to use AI ASICs that hardcode model architectures into the chips.
A Global High-Tech Industry Research Institution
As a global high-tech industry research institution, TrendForce is committed to gaining insights into the entire AI industry chain and assisting enterprises in strategic decision-making. Its research landscape focuses on the growth engines of the AI industry, covering: key memory products such as HBM, DRAM, and NAND Flash; AI computing chips like GPU and ASIC; wafer foundry, IC design, and packaging and testing; as well as infrastructure and terminals including AI servers, display panels, LED, and AR/VR. It also extends to "AI+" vertical industry clusters such as autonomous driving, embodied AI, photovoltaic and energy storage, and low-altitude economy, while providing price forecasts and databases for core components. With years of deep cultivation, it provides industry research reports, corporate strategic consulting, and integrated brand marketing services for government and enterprise clients as well as investors.