After its 2026 technology conference, Deutsche Bank said AI compute demand shows
no clear sign of cooling. Investor expectations for hyperscaler capex in 2027
have risen to $1.2–1.5 trillion versus about $800 billion in 2026. Demand is
broadening beyond large cloud providers to neoclouds, enterprises, sovereign AI
projects and new AI labs. The bank identifies memory (HBM/DRAM) and, to a lesser
extent, power supply as the current industry constraints. Conversations about
next‑generation processors are shifting from simply adding compute to securing
more memory, optimizing chip architecture and cutting per‑token inference cost.
Deutsche Bank says the next phase of incremental AI spending may favor HBM/DRAM,
power infrastructure and next‑gen architectures; if capex stays elevated,
markets should watch memory prices, delivery capacity and power availability —
potentially more market‑relevant than a pure GPU demand peak.
