Why Did Server Prices Go Up in 2026? Memory, Storage and CPUs

Server prices went up in 2026 because the parts inside them did, above all memory and flash. AI data centers are absorbing a large share of the world's DRAM and NAND output, contract prices for both have multiplied, and memory now makes up the majority of a typical server's parts cost. Dell, HPE and Lenovo responded with list increases, much shorter quote validity and new repricing terms, while Intel raised Xeon prices too. You cannot negotiate your way back to 2024 pricing, but you can control a surprising amount of the cost through how you configure and when you order.

Why are server prices going up in 2026?

The root cause is memory. Suppliers have moved capacity toward high-bandwidth memory for AI accelerators and toward server DRAM for hyperscalers, leaving everyone else short. TrendForce estimates server DRAM contract prices rose a cumulative 64% in the second half of 2025, with a further increase of roughly 270% expected across 2026, while enterprise SSD prices are projected to rise a cumulative 235% in 2026 (TrendForce). Gartner's Jon Forest told Network World that memory prices had risen 50% to 200%, with some server prices up more than 125% (Network World). CPUs, hard drives and smaller parts such as power management and BMC chips added further pressure.

How much of a server's cost is memory now?

More than half, according to HPE. On its March earnings call, CEO Antonio Neri said DRAM and NAND now make up over half of the bill of materials of a traditional server, and that the share would keep rising (The Register). Greyhound Research's Sanchit Vir Gogia told Network World that memory historically represented roughly a quarter to about 40% of server material costs, depending on configuration (Network World). The consequence for buyers is that DIMM count and SSD capacity now move the price of a server more than the processor, chassis or support term.

Module prices show the scale. Fusion Worldwide's September report cites direct Samsung Q3 RDIMM pricing of $928 for 32GB and $1,632 for 64GB, up to $7,883 for 256GB, with allocation to non-hyperscale accounts limited (Fusion Worldwide). Those are reported market figures, not anyone's list price, but they explain why a memory-heavy virtualization host now costs several times what it did. Our guide to when server RAM prices will drop covers the memory market in depth.

Did Dell, HPE and Lenovo raise prices?

All three did, along with the CPU makers:

• Dell raised server prices on December 10, 2025, and COO Jeff Clarke said in February that Dell's quotes were valid for the shortest period ever (Channel Dive). In May he said Dell was "repricing it feels like every day" (Dell Q1 FY27 call transcript).
• HPE began DRAM-related price increases in November 2025 and in March amended its quoting terms to allow repricing of existing orders for commodity cost increases between quote and shipment (Network World).
• Lenovo told partners in a February 2 letter that server quote validity was cut to 14 days in its internal system and 30 days on its partner platform, and that select larger server deals in backlog had already been repriced (Tom's Hardware).
• Intel raised Xeon list prices in July, including the flagship Xeon 6980P from $12,460 to $13,955, about 12% (TrendForce News).

The results show up in vendor revenue. In its quarter to July 31, Dell reported Traditional Servers and Networking revenue of $10.5 billion, up 122% (Dell), and HPE said strong average selling price growth in traditional servers offset supply-constrained unit volumes (HPE Q3 FY26 call transcript). Much of that growth is price, not more servers.

Why are server quotes only valid for a few days?

Because the cost of the parts can change between the day a quote is issued and the day the server ships. Vendors have protected themselves with short validity windows and, in some cases, repricing clauses. For buyers this changes the internal approval process: a quote that waits two weeks for sign-off may be void before the purchase order is raised. Before you approve any quote, check:

• Whether the price is fixed at purchase order or at shipment.
• Whether the terms allow repricing for component cost increases, and whether there is a cap.
• How long the quote is valid, and whether a written price hold is available for a deposit or firm order.
• Whether a partial shipment can be invoiced at the quoted price if some parts are delayed.

Will server prices come down in 2027?

Unlikely, based on what suppliers say. Micron CEO Sanjay Mehrotra said on September 30 that memory and storage supply-demand conditions will be much tighter in 2027 and 2028 than in 2026, that Micron has no line of sight to balance, and that more than 75% of its 2027 output is already committed (Network World). TrendForce forecasts conventional DRAM contract prices up another 10% to 15% and NAND up 15% to 20% in Q4 2026, with server DRAM still undersupplied (TrendForce). The rate of increase is slowing, and TrendForce expects server DRAM prices to keep rising each quarter through the second half of 2027 at a more moderate pace (TrendForce). Even when prices ease, expect a high plateau rather than a return to 2024 levels.

How can you configure a server to control cost?

• Right-size memory from real usage data. Counterpoint Research's Neil Shah told Network World that enterprises often waste 30% to 50% of memory by provisioning for peaks that rarely occur (Network World).
• Fill memory channels with smaller modules where capacity allows. TrendForce reports cloud providers and OEMs moving some servers from 96GB and 128GB modules to 32GB and 64GB modules. Keep channels balanced, as explained in our guide to DDR5 DIMM population.
• Size flash for the working set, not the whole data set, and keep bulk data on hard drives or another tier.
• Match core count to licensing. Per-core software licensing can make fewer, faster cores cheaper overall.
• Approve more than one platform. If one vendor's configuration is repriced or delayed, a comparable Dell, HPE, Lenovo or Supermicro build may be available.
• Be careful with legacy platforms bought only to save money up front. Shah warned that older systems may lack serviceable parts within two years, and DDR4 is no longer cheap (Network World).

How should you budget for servers in 2026 and 2027?

Treat hardware pricing as a moving number. Gartner's Forest recommends rolling 12 to 24 month forecasts, early engagement with suppliers on constrained parts, and quarterly or monthly buys rather than one large annual order (Network World). Build a contingency line into approved budgets, agree with finance who can approve a revised quote quickly, and extend the life of back-office servers where the risk is low. TD SYNNEX's Phillip Privett also pointed to vendor financing offers on compute as an opportunity. If cash flow is the constraint, compare leasing and buying, and our AI hardware shortage playbook ties the constraints together.

Nexus Compute supplies new hardware only, sourced through authorized distribution channels with the full manufacturer warranty, and checks compatibility against your platform before dispatch. Tell us the workload, the platform you prefer and any alternates you can accept, and we will quote a new configuration within 48 business hours, with purchase orders and net terms available for qualified organizations.