SK Hynix and Micron have tripled their revenues in a year as AI datacentre demand strips out global memory capacity, with Samsung roughly doubling its own. For UK infrastructure buyers, that means DDR5, HBM and NAND costs are staying high well into the refresh cycle, not falling as historically expected.
View the data behind this chart
| SK Hynix | Micron | Samsung | |
|---|---|---|---|
| Revenue growth multiple | x3 | x3 | x2 |
Why memory vendors are suddenly flush
The maths behind this boom is stark: SK Hynix and Micron have both seen revenue triple year-on-year, and Samsung's has roughly doubled, all on the back of AI datacentre buildouts devouring HBM, DDR5 and NAND flash supply. For UK buyers used to memory being a predictable, occasionally cheap commodity, this is a structural break from the normal pattern, not a temporary blip.
Historically, memory has been a boom-bust business — vendors overbuild capacity during good times, prices crash, and buyers wait out the cycle. That cycle has been suspended. AI infrastructure is consuming every wafer that comes off the line, and it shows no sign of easing before new capacity lands.
The capacity build won't help you this year
All three major vendors are now spending enormous sums to catch up. South Korea's president announced a $576 billion investment led by SK Hynix and Samsung to shore up chip production and AI supply chains, while Micron has committed up to $3 billion to strengthen the US semiconductor supply chain and is separately working to expand output across its Singapore, Taiwan and Japan sites.
None of this changes procurement timelines for 2026 buyers. Building a new DRAM or NAND fab involves years of financing, permitting, cleanroom construction and equipment validation before yields are even usable at scale. Anything greenlit today is realistically three years or more from coming online, and longer still before it meaningfully increases supply.
IDC's 2028 warning and what it means for budgets
A recent IDC report flags that meaningful relief on memory pricing may not arrive until at least 2028. For UK IT leaders building 2026 and 2027 capital plans, that's a signal to stop assuming prices will soften on the usual cyclical timetable. Anyone planning server refreshes, HBM-heavy AI clusters, or bulk DDR5 procurement should track DDR5 server memory prices rather than budgeting against last year's numbers.
This is also a good moment to understand the 2026 memory price surge in full — the shortage isn't confined to server-grade parts. It's hitting consumer electronics too, which tells you how thin the entire supply chain has become.
Buy now, phase deployment, or wait?
There's no clean answer, and the brief is honest that the outcome hinges on whether AI demand keeps justifying these prices or whether startups run out of runway first. UK buyers weighing new server orders should decide whether to buy servers now or wait based on their own refresh urgency rather than hoping for a near-term correction.
Teams specifying GPU-dense nodes should also learn why AI servers cost more due to HBM specifically — the HBM squeeze is a separate, tighter bottleneck from general DDR5 supply and it's pushing accelerator-based configurations up disproportionately.
- •Lock in DDR5 and HBM pricing early where contracts allow, given no relief is expected before 2028
- •Separate urgent refreshes from discretionary upgrades to avoid overpaying for capacity you don't need yet
- •Revisit financing structures, since higher unit costs change the maths on buy-versus-lease decisions
- •Build in longer lead times for HBM-heavy AI server configurations specifically
Practical steps for procurement teams
Given the scale of capital vendors are pouring into new capacity, prices are unlikely to spike further indefinitely, but they're also not coming down soon. UK buyers should optimise your IT procurement strategy around longer lead times and locked-in pricing windows, and where budgets are stretched, it's worth exploring how to discuss IT financing options to spread the cost of a necessary refresh rather than delaying it and risking further increases.
Whatever your platform choice, it's worth comparing configurations directly — teams can Dell server configurator, HPE server configurator or Lenovo server configurator to see how current memory pricing is shaping quoted lead times and specs before committing budget.