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

BlockBeats News, June 16th - The "AI Doomsday Report" released by Citrini Research stated that both AMD and Apple are currently advancing AI products with a flash-based alternative to DRAM. AMD acquired MEXT to optimize flash memory, bringing its performance closer to that of DRAM, thus reducing the memory cost in AI data centers; while Apple has implemented a similar optimization at the device level through the "LLM in a flash" technology.

According to Citrini's latest research report, AI inference has a high demand for KV cache in memory, and HBM accounts for 25% of DRAM's capacity in what is referred to as a "memory tax." Flash memory costs only 1/55 of DRAM, and through controller optimization, NAND stacking, and unit mode adjustments, it can provide an alternative solution for edge AI in terms of capacity and bandwidth.

This report provides theoretical support for the recent surge in storage stocks, led by SanDisk.