AI demand is pushing DRAM prices up 171.8% and relief may not come until 2028
量子位 · wechat · 2026-07-25
A WeChat article argues that the current memory price surge is not a normal cycle, but a structural shift driven by AI demand.
In 2025 Q3, DRAM contract prices reportedly rose 171.8% year over year. The article says the real driver is not simple shortage, but a major reallocation of advanced manufacturing capacity toward HBM and data-center DRAM. As a result, consumer markets like PCs and phones are facing slower, pricier, and less stable supply.
Three takeaways stand out:
- Capacity is being diverted: major memory makers are reportedly sending 70–80% of advanced-node output to HBM and data-center demand.
- Price transmission is uneven: PC memory sees the pain first, server memory is often locked in through contracts, and smartphones absorb the cost more gradually through product pricing and spec changes.
- Relief may be years away: several forecasts in the piece point to high prices through 2026–2027, with a clearer normalization only around 2028.
The article concludes that this is a structural opportunity for the memory supply chain, not a short-lived spike.
More from Infra
- 80% of the DIY LLM inference hype posters have already quit — it's brutally hard systems work — abhijithneil · 2026-09-11
- Hugging Face's Ultra Scale Playbook: a free book on training LLMs on GPU clusters — mdancho84 · 2026-09-11
- Is inference latency becoming the biggest bottleneck for production AI agents? — Euphoric_Sea632 · 2026-09-11
- LLM Serving Metrics Thread: Why TPOT and Uptime Make or Break User Experience — abhijithneil · 2026-09-11
- PlanetScale launches sharded Postgres: 768 servers acting as one, 1PB scale — dhruv2038 · 2026-09-11
- Can a 7900 XTX 24GB run Qwen locally? Reddit seeks ROCm tok/s benchmarks — thenomadexplorerlife · 2026-09-11