Chip analyst P Equity argues that hyperscale cloud providers cannot accurately forecast two years out, let alone ten. AI may well grow for a long time, but every technology eventually matures, and capex will sooner or later shift from high-speed growth to a plateau. Storage therefore remains a cyclical industry. This cycle may have a milder downturn than past ones because of long-term agreements (LTAs), prepayments, and price floors.
What is truly extreme in the short-to-medium term is storage. AI data centers are gradually shifting from training toward inference, and inference is more demanding on memory bandwidth, model weights, KV cache, and storage for long-running agent tasks. Institutions differ widely on what share of hyperscaler capex storage will take, but they converge on one conclusion: storage has become one of the largest cost items in AI infrastructure. The note even cites a UBS estimate that storage spend alone could reach $900 billion next year.
