Deutsche Bank points out that the persistent "Memory Wall" and tight memory supply have begun to impact the product roadmap for next-generation AI chips. Investors are discussing whether some new processors will reduce HBM configurations, while high DRAM prices are also starting to be factored into cost estimates for total AI capital expenditures. Supply constraints are driving the accelerated emergence of new technology routes, including inference chips that rely more on SRAM, custom HBM base dies, HBF, and other HBM variants. Deutsche Bank emphasizes that related engineering solutions are still in a rapidly evolving phase, with low supply chain visibility, but most management believes that the current issues are mainly focused on economics and cost, and overall DRAM demand is likely to remain high.
This also adds a new dimension to the observation of the HBM boom: the tighter the supply and the higher the price, the more beneficial it is for memory manufacturers' profitability in the short term, but it also makes it easier to stimulate customers to look for more memory-efficient architectures. In the next stage, in addition to looking at HBM capacity and prices, it is also necessary to observe the single-chip HBM configuration of next-generation GPUs and ASICs, as well as whether alternative solutions such as SRAM and HBF can truly enter large-scale applications.
