If a server quote you signed off six months ago now looks 40% higher, you are not imagining it. The AI build-out has turned memory and flash into the most squeezed components in the data centre: DRAM contract prices have risen sharply through early 2026, and enterprise SSDs are not far behind. This is not a normal cyclical wobble you can wait out in a quarter — supply is committed years ahead. This guide explains what is actually happening, how long it lasts, and the practical moves a UK buyer can make to protect a refresh budget without over-buying.
What's driving the surge
The cause is concentrated demand. High-bandwidth memory (HBM) for AI accelerators consumes roughly three times the wafer capacity per bit of standard DDR5, so every wafer the big three memory makers pour into HBM is a wafer denied to ordinary server and PC memory. With AI projected to absorb around a fifth of total DRAM output in 2026 and HBM lines already sold out for the year, the manufacturers have reallocated capacity toward the highest-margin parts. The result, per TrendForce and Gartner tracking, is DRAM contract prices up dramatically quarter-on-quarter in early 2026 and a projected triple-digit year-on-year rise.
Flash is following the same curve a beat behind. NAND prices have risen sharply quarter-on-quarter and enterprise SSDs — the high-capacity, high-endurance drives that feed AI data pipelines — are seeing the steepest increases as hyperscalers lock up supply. Meaningful new capacity is not expected to land until late 2027 or 2028, so this is a multi-year structural squeeze rather than a blip.
What it means for a server or storage refresh
Memory and storage are now a much larger slice of a server's bill of materials than they were a year ago, and lead times have stretched. A spec that was 'add more RAM, it's cheap' is no longer true — over-specifying memory you don't need is now an expensive habit. The same applies to filling every drive bay 'to be safe'. The discipline that used to be optional is now where the money is.
Crucially, the squeeze is worst at the newest, densest parts (the biggest DIMMs, the largest enterprise SSDs, anything competing directly with AI demand). Mainstream capacities and the previous generation are less affected — which changes the value maths on what to buy.
Five moves that protect the budget
None of these require gambling on a price you can't predict — they reduce exposure to the parts that are spiking.
- •Right-size, don't round up. Spec the memory and flash the workload actually needs (use our calculators), then add planned headroom deliberately rather than padding 'just in case'.
- •Buy the value capacity, not the halo part. Two mainstream DIMMs often beat one of the largest, scarcest modules on both price and availability — and the same for drives.
- •Lock pricing and slots early. Quote and commit ahead of a refresh while you can, and confirm lead times; a credible order beats a spot purchase mid-spike.
- •Look hard at refurbished and the previous generation. Certified refurbished servers and last-gen platforms sidestep the worst of the new-part premium for many workloads.
- •Extend, don't replace, where it's safe. A supported maintenance extension on healthy kit can bridge past the worst of the squeeze — provided the platform isn't a security or capability dead-end.
What we'd advise right now
For most UK organisations the answer is not to panic-buy or to freeze. It's to plan refreshes deliberately: size precisely, favour value capacities and refurbished where the workload allows, lock pricing on what you do need, and use a maintenance extension to buy time on anything that's healthy. The buyers who get hurt are the ones who over-spec out of habit or leave a spot purchase to the last minute.
Servnet quotes new and refurbished side by side, confirms current lead times before you commit, and will right-size a build with you so you're not paying the AI premium on memory and flash you won't use.