Map/Compute hardware/Wave 2

Commodity memory crowded out by HBM

Moving DRAM wafers to high-bandwidth memory starves ordinary server, PC and phone memory, which raises AI server costs and delays deployments.

brakeBuilt today Open on the map →

Facts

Lane: Compute hardware
Wave: 2 of 12 (build order, not a date)
Needs: 2 boxes all the way back through hard links
Unlocks: 0 boxes downstream through hard links
Cluster: Around datacentre capex
Found by: bill of materials lens

Evidence as of October 2026

Conventional DRAM contract prices rose 90-95% quarter on quarter in Q1 2026, and Goldman projects 2026 deficits of 4.9% for DRAM and 4.2% for NAND, the largest since 2011, with normalisation base-cased for 2028.

Around it

Needs · 2

Memory bandwidth and supplycapability · Built today
Datacentre capexcapability · Built today

Unlocks · 0

Nothing further on the map.

Brakes

This box works against

Hardware cost per tokencapability · Built today
Compute supplycapability · Built today
Neocloud compute providerscapability (detail) · Built today
On-device modelscapability · Built today

Loops and off-ramps

Payback and collateralbalancing loop
Capex pullbackoff-ramp · slowed
Taiwan chip disruptionoff-ramp · slowed

Openings researchers' judgement, October 2026 scouting

Crowding4 Lab overlap4.5 Small-team fit4 Buyer today4 Research openness3

Depth: light
Players: Samsung; SK Hynix; Micron; Nvidia (Dynamo, Groq); Astera Labs; Marvell (CXL); XCENA ($135M B, May 2026); Tensormesh ($20M); Cerebras; vLLM/LMCache
Wedge: Inference software that cuts DRAM/HBM per token (KV-cache tiering to cheaper memory or flash, compression, memory-aware scheduling), sold to neoclouds and inference providers facing doubled memory bills. startup