AMD introduces MEXT memory solution to reduce server DRAM demand using AI

On November 11, 2026, tech outlet Wccftech reported that AMD has launched a memory solution named MEXT for its EPYC server processors. The launch follows AMD's acquisition of MEXT, a memory-optimization technology company, which was announced earlier in June 2026 and aimed at alleviating the memory bottlenecks increasingly faced by data centers.

MEXT is software that manages DRAM and NVMe solid-state drives jointly. Rather than trying to make flash storage as fast as memory, the approach assigns data to the right tier: rarely accessed cold data is moved to NVMe drives, while an AI engine predicts which data will be requested next and pre-fetches it back into DRAM to limit the impact of flash access latency. According to AMD, the solution can reduce server DRAM demand by up to 75% while sustaining near-full-DRAM application performance.

The package consists of a Linux driver, an AI engine, and a memory-monitoring tool called RAM View. The AI engine does not require a GPU and runs on a single EPYC processor core. In tests presented by AMD, a three-node Azure cluster running Spark TPC-DS retained roughly 99% of full-DRAM baseline performance when flash storage took over half of the memory capacity. Another test using the chip-design software Synopsys PrimeTime also showed that halving DRAM still preserved solid performance.

As of the latest available coverage, no general-availability date, licensing terms, or pricing for MEXT have been announced.