TrendForce's latest analysis of the memory chip market reveals that major global cloud service providers are dramatically accelerating their artificial intelligence infrastructure investments. Capital expenditure among these cloud operators is projected to surge by 98% year-on-year in 2026, with growth remaining robust at 50% in 2027.

Memory is playing an increasingly dominant role in this spending spree. The research firm estimates that DRAM and NAND Flash combined will account for 47% of total cloud service provider capital expenditure in 2026, climbing further to 68% by 2027. In practical terms, for every 100 yuan spent on capital expenditure by cloud operators, nearly 70 yuan will be directed toward storage by 2027.

This rapid increase in the storage share is driven by substantial jumps in memory contract prices.

Explosive Price Hikes: Server DRAM Cumulative Gains Could Triple in Two Years

TrendForce notes that memory contract prices have risen sharply since the second half of 2025, significantly boosting the storage portion of cloud provider spending. Server DRAM contract prices accumulated a 64% increase in the second half of 2025, with a further estimated surge of approximately 270% expected in 2026. Enterprise-grade SSD (NAND Flash) prices rose about 35% in the latter half of 2025, with an estimated cumulative increase of 235% in 2026. HBM contract prices may still climb by 70% to 140% in 2027.

The firm indicates that although some long-term agreements signed from the second quarter of 2026 include price caps that may limit further upside, overall memory contract prices are expected to remain elevated through 2027, continuing to be a significant factor pushing up the share of storage in cloud service provider capital expenditure.

Supply Side: HBM and RDIMM to Represent 51% of DRAM Bit Supply in 2026

Beyond price increases, structural shifts in demand are equally noteworthy. TrendForce estimates that HBM and RDIMM combined will account for 51% of DRAM bit supply in 2026, as suppliers prioritize limited capacity for server applications. By 2027, as process node migrations advance and new wafer fabs ramp up in the second half of the year, combined bit supply for server DRAM and HBM is expected to grow by 27%.

The research firm points out that rising prices combined with increased supply will jointly drive storage to reach 68% of cloud service provider capital expenditure by 2027.

Two Major Ripple Effects: Chip Price Hikes and Architecture Overhaul

TrendForce believes that high storage costs will have two significant impacts on the AI ecosystem. First, it provides justification for AI chip price increases. The firm states that rising memory contract prices, especially for HBM, will give server and AI chip suppliers such as Nvidia more substantial grounds to raise product prices. Cloud service providers may then need to further increase capital expenditure to maintain their planned AI chip procurement volumes.

Second, it pushes cloud providers to optimize storage architectures. An alternative response path involves cloud operators more aggressively optimizing their AI system storage designs, reducing storage capacity per system to alleviate the pressure from high DRAM and NAND Flash costs or supply allocation constraints, while still achieving their AI chip and server shipment targets. The firm outlines several potential adjustment approaches, including modifying RDIMM configurations, reducing HBM capacity integrated into future AI chips, and exploring AI ASIC solutions that embed model architectures directly into chips.