VMware is expanding its memory tiering technology to help customers cope with soaring memory and hardware prices, betting that businesses will want to delay server purchases rather than pay inflated costs. The tool, built into VMware's ESX hypervisor and part of Cloud Foundation, shifts "cold" data out of expensive DRAM onto fast NVMe storage, freeing up capacity so CPUs don't sit idle waiting for memory. VMware argues this matters because DRAM now often costs several times more than the servers housing it, a squeeze driven by the AI boom's demand for chips and memory.

At VMware Explore, technical marketing architect Dave Morera outlined a two-to-three-year roadmap, with the next update likely arriving around May 2027, aiming to extend support beyond the roughly 75 percent of workloads currently covered to include Microsoft's Radius platform and "monster VMs" running up to 960 vCPUs and 16TB of memory. VMware recommends a 1:1 ratio of DRAM to NVMe storage and claims the approach can cut CPU cycle consumption by 30 percent and total cost of ownership by 40 percent by letting firms run more workloads on fewer servers. Using the feature requires fast, durable NVMe drives, and while VMware naturally has commercial reasons to push adoption, the firm says both conference sessions on the topic drew capacity crowds of 500 attendees.