Buying a PC or a smartphone will cost more in the second half of 2026, and the reason remains one word: AI.
According to the Taiwanese research firm TrendForce, memory manufacturers are prioritizing shifting their production capacity toward AI servers, leaving consumer markets with tighter supply and prices trending upward.
This shift is part of the surge in investments by the major CSPs that continue to lift the entire AI server market.
On this segment, TrendForce now estimates growth at nearly 31% year over year, up from 28% previously.
Three factors explain this rebound.
First, a clear strengthening of demand for rack-scale NVIDIA GB/VR platforms among hyperscalers and second-tier data center operators. Then, the expected ramp-up in production at Google and AWS of their next generation of internal ASICs in the second half of 2026. Finally, the acceleration of deployment of domestic AI solutions by Chinese CSPs to support their large language model services.
AI Servers: +31% in 2026
According to TrendForce’s estimates, the cumulative capital expenditures (CapEx) of the nine largest CSPs worldwide (Google, Amazon, Meta, Microsoft, Oracle, ByteDance, Tencent, Alibaba and Baidu) are expected to exceed $886.7 billion in 2026, up by about 90% year over year.
The five North American hyperscalers alone account for nearly 90% of this total. This uptick reflects continued investments in AI data centers, GPU clusters, liquid cooling infrastructure and the full range of equipment needed to meet the rising demand for generative AI and large-scale models.
Google continues to favor deploying its home-grown TPUs in 2026 and 2027, with volumes expected to grow rapidly in 2027.
AWS is betting on the NVIDIA GB300 GPU platform as its main solution in 2026, while gradually developing its own ASICs, whose volumes are expected to keep growing in 2027.
Meta, for its part, will rely mainly on rack-scale NVIDIA GB/VR systems and AMD Helios in 2026, before significantly accelerating the deployment of its own ASICs in 2027 in a move to reduce infrastructure costs and improve inference efficiency.
China accelerates its investments
TrendForce expects that the combined spending of the four leading Chinese CSPs ( ByteDance, Tencent, Alibaba and Baidu ) will rise by more than 80% year over year in 2026.
ByteDance is expected to show the strongest increase, with investments focused on large-scale AI data centers, the development of proprietary ASICs and the deployment of GPU clusters.
Looking ahead to 2027, TrendForce anticipates the cumulative CapEx of the nine largest CSPs to reach around $1.3 trillion, i.e., about 50% year-over-year growth. The slowdown in growth would be mainly explained by a higher base effect, and would not signal a tapering of AI investments.
Progress expected in AI inference, AI agents, custom ASICs and next-generation models should continue to sustain strong demand, with investments extending beyond GPUs to cover AI servers, liquid cooling, advanced packaging, high-speed interconnects, electrical infrastructures and memory.
TrendForce forecasts that DRAM will remain tight into the third quarter of 2026, but with more moderate price increases: 13–18% for DRAM, 10–15% for NAND Flash.
Persistent tightness, moderate price gains
This demand remains largely driven by AI inference and the deployment of data centers at scale. Memory manufacturers continue to reorient production toward higher-margin server and AI applications, at the expense of consumer segments:
- Server DRAM: General-purpose servers based on x86 processors and RDIMM modules remain the memory platform of reference for AI agent workloads, thanks to their multitasking capabilities. Server shipments are expected to stay solid through 2027 as CPU availability improves, but the market will remain undersupplied in the third quarter.
- LPDRAM (mobile memory): Suppliers continue to prioritize AI-related applications in their production allocation, keeping LPDRAM supply tight and fueling further price rises as smartphone makers become more cautious amid weaker consumer demand.
- Enterprise SSDs: CPU shortages have curbed system deliveries, prompting buyers to restock, while NAND suppliers allocate more capacity to enterprise SSDs, notably driven by the gradual deployment of the NVIDIA Vera Rubin platform.
- Graphics DRAM (GDDR): Demand tied to the NVIDIA RTX PRO 6000 Blackwell card remained below expectations, but suppliers continue to reallocate capacity to other products, keeping GDDR6/7 supply tight and prices on the rise.
By contrast, consumer segments (PCs, smartphones, storage) show signs of fatigue. Demand remains weak for televisions and set-top boxes, while high inventories among PC manufacturers prevent any further rise in SSDs.
Additionally, memory price records are starting to push against buyers’ tolerance, helping moderate the increases seen in the third quarter.
