AMD said it has acquired MEXT, a startup that builds software to wring more usable memory out of data‑center storage. The price was not disclosed. AMD described the technology as strategically important for AI infrastructure, though it hasn’t shared financial details of the deal.
MEXT’s approach pushes flash closer to RAM in behavior — i.e., making SSDs act more like volatile memory so servers can present more addressable memory without buying more DRAM. In practice that lets teams run bigger AI models or heavier computations on lower‑cost media (e.g., NVMe flash) instead of relying solely on expensive memory modules.
According to AMD, the plan is to fold MEXT’s code into its processors and platforms to improve efficiency for AI workloads across cloud and HPC environments and to trim infrastructure spending where memory is the bottleneck. Whether the trade‑offs around latency, consistency, and real‑world manageability pay off at scale remains to be seen; vendors and operators will be watching implementations closely.