AI's memory crunch: DRAM prices may jump 130%, and consumers pay first

APPSO · wechat · 2026-11-06

Sam Altman's podcast remark that idle PC RAM should be pooled into cloud datacenters sparked mockery — he briefly renamed himself "Ram Altman" and Musk posted a meme. The piece argues AI's resource race has shifted from GPUs to memory: long-context and agent inference make HBM and server DRAM the new bottleneck, with Samsung, SK Hynix and Micron controlling supply.

Capex keeps ballooning: Microsoft, Amazon, Google and Meta are projected to spend over $730B combined in 2026, funded increasingly by debt; OpenAI is co-designing 10GW of custom accelerators with Broadcom and building Stargate, while reportedly discussing cutting its 2027 datacenter capacity forecast by 40%.

For consumers, Gartner expects DRAM/SSD prices 130% above 2025 levels, PC prices +17%, smartphones +13%. Cloud AI users don't need more RAM, but local-model enthusiasts and creators will feel the pinch — a textbook "productivity J-curve" where infrastructure costs land before the benefits.

Original post →

AI's memory crunch: DRAM prices may jump 130%, and consumers pay first

More from Venture