[News] AMD Taps Flash-Based MEXT amid Memory Crunch, Aims to Cut DRAM Needs by Up to 75%

What happened

In November, AMD unveiled MEXT Predictive Memory Software, a flash-based, AI-powered predictive memory solution that could slash DRAM requirements by 20–75% across EPYC servers while easing costs and supply pressures with comparable performance; the rollout follows AMD's acquisition of MEXT in June. The cause lies in DRAM overprovisioning: enterprises often overprovision memory to prevent performance slowdowns; according to Wccftech, DRAM overprovisioning can reach 50% (half of allocated memory untouched by virtual machines), another 25% remains stranded within instances, and overall 50–80% of DRAM capacity may be underutilized as memory pages become inactive or 'cold'; DRAM reportedly accounts for as much as 75% of total server costs amid persistent supply shortages. MEXT shifts idle data to flash and uses AI to anticipate demand, moving data back to DRAM before applications need it to minimize latency. According to Wccftech, MEXT could effectively triple available memory capacity at a fraction of the cost of adding physical DRAM; AMD claims a 50/50 mix of DRAM and MEXT-enabled flash could boost performance per dollar by 90% in five-agent concurrent workflows and 70% in memory-intensive agent applications (AMD claims). AMD is already putting MEXT to work in its own IT infrastructure: an initial deployment across 145 servers is expected to avoid roughly $4.9 million in DRAM purchases, slashing memory expansion costs by 83% versus an equivalent DRAM-only upgrade (expected, not confirmed as achieved), and AMD aims to expand the rollout to 2,000 servers by 2027. AMD is also teaming up with Hewlett Packard Enterprise (HPE) to bring MEXT to enterprise customers, initially focusing on 'MEXT Ready' systems with ample CPU cores and NVMe expansion capacity. Separately, NVIDIA is taking a related, though distinct, approach in AI inference with its Context Memory Storage (CMX) platform, which adds a flash-based tier beneath GPU HBM and host DRAM for KV cache; according to Wccftech, CMX uses TLC flash connected to Rubin GPU clusters via BlueField-4 DPUs and Spectrum-X Ethernet.

Impact and signal

Affected components and suppliers

Source and time

Source: TrendForce · 11/10/2026 · View original source