The forecast that changes the planning horizon
Micron CEO Sanjay Mehrotra told investors on the company's fiscal fourth-quarter earnings call that memory and storage supply-demand conditions are expected to be much tighter in calendar 2027 and 2028 than they were in 2026, directly contradicting earlier industry forecasts that had expected supply and demand to rebalance by 2028. Mehrotra added that Micron has no clear line of sight into when that balance will actually return, a notably blunt admission from a company with as much visibility into the memory supply chain as any in the industry.
That matters for enterprise budget planning specifically because it resets the time horizon CIOs should be using. Treating elevated memory and server costs as a temporary 2026 anomaly to be waited out is no longer a defensible assumption. New fabs will not help quickly either, since Mehrotra noted output ramps gradually and typically starts only a few quarters after initial production begins, meaning any new capacity announced today will not meaningfully ease prices before 2028 at the earliest.
The numbers behind the forecast
Micron's own DRAM prices rose in the high teens percentage-wise in the most recent fiscal quarter, while NAND prices rose roughly 30 percent, according to CFO Mark Murphy. Independent research from TrendForce, published around the same time, forecasts conventional DRAM contract prices rising 10 to 15 percent quarter over quarter and NAND rising 15 to 20 percent, noting that while the rate of increase is slowing, the market remains meaningfully undersupplied relative to demand.
Micron has already committed more than 75 percent of its 2027 output, and most current customer negotiations are now focused on securing 2028 supply rather than 2027, which is itself a signal of how far ahead the largest buyers are now planning. Much of this committed supply sits inside multiyear, take-or-pay strategic customer agreements, and new agreements being signed now reflect current elevated market pricing rather than historical baselines, locking in today's higher costs for years to come.
How cloud providers and server makers are already adapting
TrendForce reports that cloud providers and server OEMs have been shifting server configurations from 96 and 128 gigabyte memory modules down to 32 and 64 gigabyte modules since the first half of 2026, a direct response to constrained supply that trades memory capacity per server for availability and cost control. That shift has real implications for workloads that assume generous memory headroom, particularly memory-intensive analytics and in-memory database deployments that were sized assuming the larger module configurations would remain the default.
Some cloud providers locked in long-term memory supply deals before this squeeze intensified, and those providers are now functioning as the more cost-stable option relative to competitors still buying DRAM at current spot and contract pricing. That divergence means your choice of cloud provider is now meaningfully affected by that provider's own memory procurement timing, a factor that was essentially invisible to buyers eighteen months ago and now deserves a direct question in any vendor evaluation.
Why moving to the cloud does not dodge this
Counterpoint Research analyst Neil Shah makes the point directly: migrating workloads to the cloud will not avoid rising hardware and energy costs, because cloud providers facing the same memory market pressure will pass those costs on to customers through pricing, regardless of whether a workload runs on-premises or in a public cloud. That undercuts a common budget-planning assumption that cloud migration insulates an organization from hardware commodity cycles, when in practice it mostly shifts who negotiates the underlying supply contract.
Lenovo's commercial leadership, including Yuanqing Yang, expects memory supply to stay constrained at least through the end of 2027 and says the company can adjust pricing quickly when component costs rise, which from a buyer's perspective means expecting this pricing pressure to persist, not expecting a vendor to absorb it on your behalf indefinitely. Lenovo is simultaneously reporting a strong server refresh cycle underway even as PC unit volumes are expected to fall, underscoring that enterprise infrastructure demand remains resilient even as consumer demand softens under the same price pressure.
What CIOs should actually do about it
Shah's recommendations, grounded in this forecast, give CIOs four concrete levers. First, lock in multiyear pricing now for computing capacity you already know you will need, since waiting for prices to soften is no longer a reasonable bet given Micron's own stated lack of visibility into rebalancing. Second, be selective about refresh timing: extending lifecycles from three to five years is relatively low-risk for back-office PCs and routine file servers, but delaying refreshes for core infrastructure and engineering workstations can backfire if performance degradation outpaces the savings.
Third, avoid buying legacy DDR4 systems to save roughly 10 percent upfront, since DDR4 supply is becoming scarce as manufacturing shifts toward HBM and DDR5, and systems bought today on legacy memory risk becoming unsupported within two years. Fourth, and most immediately actionable, optimize existing hardware before buying anything new: enterprises typically waste 30 to 50 percent of provisioned memory by sizing for peak loads that rarely occur, and right-sizing virtual machines, quantizing AI models, and batching workloads can meaningfully reduce memory consumption without a single new purchase order.



