Citrini: AMD and Apple Advance Flash-to-DRAM Replacement in AI Products

June 16 — Citrini Research’s "AI Doomsday Report" reveals that both AMD and Apple are developing AI products using flash-based alternatives to DRAM. AMD acquired MEXT to boost flash memory performance, bringing it nearly on par with DRAM and cutting memory costs for AI data centers. Apple has rolled out a similar device-level optimization through its "LLM in a flash" technology. Per Citrini’s latest research, AI inference workloads have high memory demands for KV cache, with HBM making up 25% of DRAM capacity in what’s called the "memory tax". Flash memory costs just 1/55 of DRAM, and through controller tuning, NAND stacking, and unit mode adjustments, it provides a viable alternative for edge AI applications in terms of both capacity and bandwidth. This report offers theoretical support for the recent rally in storage stocks, with SanDisk leading the gains.