AMD’s MEXT is predictive memory-optimization software for data centers. A November 11, 2026, report by Tom’s Hardware said more than 100 servers were running it, with AMD planning to expand the deployment to 2,000 servers in 2027. The report date is not a deployment date.
How MEXT is described to work
AMD announced that it had acquired MEXT on June 15, 2026, describing the technology as a way to expand usable memory and reduce data-center costs in its acquisition announcement.
El Comercio described MEXT as predictive memory tiering: software predicts which data will be needed, moves infrequently used data from DRAM—the fast memory a server uses directly—to flash storage, and retrieves it when needed. The approach is meant to make more of a server’s memory capacity usable without keeping every piece of data in DRAM.
MEXT’s practical promise depends on prediction: data moved out of DRAM must return when needed, while the workload still meets its performance target.
What has been reported about the deployment
Tom’s Hardware reported in its November 11 headline that more than 100 servers were running MEXT and that AMD planned to expand the deployment to 2,000 in 2027. Separately, El Comercio reported AMD’s figure of an initial deployment across 145 of its own servers.
El Comercio also attributed two results to AMD: a chip-design program used half as much memory while retaining 95% of performance, and the 145-server deployment avoided $4.9 million in memory purchases. The test result applies to that chip-design program; it is not a general performance guarantee. The cost figure refers to memory purchases AMD said it avoided, not MEXT’s price.
What the server claims mean for consumer devices
AMD’s stated aim is to reduce memory costs in data centers. Any effect on laptop or phone prices would depend on whether MEXT materially eases data-center demand for memory, El Comercio reported.
