AMD and Apple Simultaneously Advance Flash-Based DRAM Alternatives for AI, Cut Memory Costs 55-Fold

According to Citrini Research, on June 16, AMD and Apple are simultaneously advancing flash-memory-based alternatives to DRAM in AI products. AMD acquired MEXT to optimize flash storage performance close to DRAM levels, reducing data center memory costs; Apple deployed "LLM in a flash" technology to achieve similar optimization at the device edge.

Citrini's latest research indicates that KV cache demands in AI inference and HBM consuming 25% of DRAM production capacity create significant memory cost pressures. Flash storage costs only 1/55 that of DRAM, and through controller optimization, NAND stacking, and cell mode adjustments, it offers viable capacity and bandwidth alternatives for edge AI.