If hardware price squeezes make you sad, VMware says it will all end in tiers
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.
- VMware expands memory tiering to offset high DRAM and hardware prices
- Tech shifts unused data from DRAM to fast NVMe storage
- Claims up to 40% lower costs; wider VM support due by 2027
New here? Start with this
VMware makes software called a hypervisor, which lets one physical server run several virtual machines. Its ESX hypervisor is part of VMware Cloud Foundation, a platform businesses use to manage those servers and workloads. Virtual machines need memory, usually supplied by DRAM, which is fast but can be costly.
VMware’s memory tiering moves data that is used less often from DRAM to NVMe storage, a type of fast solid-state drive. This can free DRAM for data that needs it more often, though the approach depends on suitable drives and may not fit every workload. VMware says it could help businesses run more work on existing servers and put off buying new hardware.
Memory prices matter because businesses rely on servers for many kinds of computing, and demand from AI has increased competition for memory and chips. VMware has a commercial interest in customers adopting its technology, so its projected savings are claims from the company. The wider issue is whether shifting some work to storage can help businesses manage hardware costs.
Both sides, in good faith
The strongest fair case each way — we don't pick a winner.
The case for
Memory tiering could help businesses manage unusually high hardware costs by moving less-used data from expensive DRAM to NVMe storage, potentially delaying server purchases and making better use of existing CPUs. If VMware’s performance and savings claims hold in customers’ workloads, the feature could offer practical relief while extending the life of current infrastructure.
The case against
The claimed savings are VMware’s estimates, and actual results may vary with workload, storage costs and the performance impact of moving data between tiers. The approach also depends on fast, durable NVMe drives and may not suit every application; businesses could find that buying additional memory or servers is more predictable than relying on a newer feature whose broader support is still on the roadmap.