Deutsche Bank says a persistent 'memory wall' and tight memory supply are
reshaping next‑generation AI chip roadmaps. Investors are discussing some new
processors cutting HBM configurations as elevated DRAM prices are being built
into AI capex calculations. Supply constraints are accelerating alternative
approaches, including more SRAM‑dependent inference chips, custom HBM base dies,
HBF and other HBM variants. Engineering plans remain fluid with limited
supply‑chain visibility; most management teams view the problem as primarily
economic—cost and affordability—while overall DRAM demand may remain high. Key
market signals to watch: HBM capacity and pricing, per‑chip HBM allocations on
next‑gen GPUs and ASICs, and whether SRAM, HBF and other substitutes achieve
scalable deployment.
