Refurbished Servers vs New: How DRAM Flipped the 2026 Maths

The 2026 memory supercycle: why buying a server got expensive overnight

The root cause is demand, not a factory fire or a shipping delay. TrendForce data shows DRAM contract prices — the agreements between memory makers and large buyers, not retail spot prices — climbed 90-95% quarter-on-quarter in Q1 2026, with a further 58-63% QoQ rise projected for Q2. NAND flash then overtook DRAM's rate of increase for the first time in this cycle, forecast to jump 70-75% QoQ in Q2 2026.

The driver is AI infrastructure. Avnet and EnkiAI estimate AI data centres will consume roughly 70% of high-end DRAM output in 2026, an inversion of previous cycles where enterprise servers and PCs set the pace. Manufacturers are responding rationally: wafer capacity is being reallocated toward higher-margin HBM and server-class DRAM, squeezing the conventional DRAM that ordinary enterprise servers depend on. Gartner puts the combined effect at a 130% surge in DRAM and SSD prices by the end of 2026 — and TrendForce doesn't expect meaningful new fab capacity before late 2027 or 2028. This is not a blip UK buyers can wait out.

New servers in mid-2026: premium price tags, longer queues

Major OEMs implemented 15-20% system price hikes in Q1 2026, according to ReluTech, directly passing DRAM cost inflation to buyers. Lead times have stretched to 4-8+ weeks as vendors ration memory allocation across their order books. A single DDR5 64GB RDIMM — the kind of module that fills a mainstream two-socket server — moved from roughly $873 in Q1 2026 to around $1,350 by June 2026, according to reports from ServerMonkey and Citi data at the time. That's one memory module, not a whole system, but it illustrates why bills of materials are moving.

The UK retail channel shows the same pressure lower down the stack: DDR5-6000 32GB kits rose more than 250% in late 2025, from around £80 to roughly £350, per Bacloud.com. DDR5 is becoming the default for new deployments aimed at AI, big-data and dense virtualisation workloads, per Wecent — but that's precisely the memory generation currently absorbing the steepest price rises.

Refurbished servers: where the DRAM shock flips the maths

Refurbished systems typically carry memory that was populated before this price cycle took hold — DDR4, or earlier-generation DDR5 — so the buyer isn't paying 2026 memory prices for that capacity. That's the mechanism behind the widened saving. Hivenet puts potential savings at up to 70% versus new; Comtek's more conservative estimate is 30-50%, but notes the differential grows even wider during periods of DDR5 pricing volatility — exactly the conditions of mid-2026.

Availability matters just as much as price. Where new-build OEM orders sit in a 4-8+ week queue, certified refurbished stock is generally available ex-stock or within days. For workloads that don't need bleeding-edge DDR5 bandwidth — file, ERP, steady-state virtualisation, backup targets — Wecent notes DDR4 "remains a solid option for legacy systems and budget-constrained environments," which is a fair description of a large share of UK enterprise estate. If you're weighing this up in detail, our guide to buying refurbished enterprise hardware covers the sourcing side; you can also configure a refurbished Dell server to see current specced-out pricing.

TCO: what the maths actually look like for a UK buyer

A proper total-cost view has to include more than the invoice. On acquisition, new hardware carries both the 15-20% OEM hike and the compounding memory cost shown by the RDIMM figures above; refurbished hardware sidesteps both, landing in the 30-50% (up to 70%) saving band already discussed. On timing, the 4-8+ week new-build lead time is itself a cost — a delayed go-live, a stalled project milestone, a longer bridge on rented capacity — even before any figure is attached to it.

Refresh planning matters too. Servers generally follow a four-to-five year refresh cycle, but 2026's extended lead times mean that cycle now needs to start earlier, not later, if new-build is the plan. Licensing is a separate lever entirely: Windows Server 2012's final year of Extended Security Updates ends on 13 October 2026, which is a forcing function for any buyer — new or refurbished — still running that estate. Spreading either acquisition route is worth exploring through server finance options, and extending cover on existing kit via third-party maintenance for your hardware is often cheaper than a forced early refresh driven purely by memory pricing.

Warranty and reliability: separating fact from fiction

The instinctive worry with refurbished hardware is coverage — will anyone stand behind it if a DIMM or a drive fails eighteen months in? The honest answer is that warranty quality varies significantly by supplier, which is exactly why due diligence is the differentiator, not the asset class itself. New OEM kit comes with manufacturer-backed cover as standard; reputable refurbished suppliers back their stock with their own warranty and support commitments, and the difference in practice comes down to how rigorously that supplier tests before it ships.

This is also where the EU's Right to Repair directive, arriving in July 2026, should shift the picture for the better across the estate lifecycle: it requires manufacturers to provide spare parts and repair guides, which materially improves the long-term serviceability of hardware that's already in the field — new or refurbished.

Vetting a refurbished supplier: the UK due-diligence checklist

Not all refurbished stock is equal, and the risk sits almost entirely in supplier selection rather than in the concept of buying refurbished itself. Before committing budget, UK buyers should be asking a specific set of questions.

• •Ask for the grading standard used and what physically distinguishes each grade
• •Request evidence of pre-sale testing — burn-in runs, stress tests, and SMART data on any drives
• •Confirm the warranty term and what it actually covers: parts, labour, DOA replacement window
• •Check the supplier's return policy and turnaround time for faulty units
• •Verify firmware and support status on the platform to avoid buying into a dead-end EOL model
• •Use tools like a server end-of-life checker or storage end-of-life checker before locking in a platform for the next refresh cycle

The UK sustainability angle — and the decision framework

There's a genuine ESG case alongside the financial one. The UK refurbished electronics market was valued at roughly £7.6 billion in 2025 and is projected to more than double by 2032; refurbished PC sales volumes nearly doubled between Q4 2024 and Q4 2025, with the UK overtaking Germany to become Europe's largest refurbished PC market. For any organisation with a stated sustainability commitment, refurbished procurement is now a mainstream, well-evidenced route rather than a niche one.

Putting it together: buy new where the workload genuinely needs current-generation DDR5 bandwidth — AI-adjacent, big-data or very dense virtualisation — and where the 4-8+ week lead time fits your project timeline. Choose refurbished where the workload is standard business computing (ERP, file, steady virtualisation, DR targets), where budget discipline matters, or where you need capacity now rather than in two months. Given TrendForce doesn't expect real relief in DRAM supply before late 2027 or 2028, mid-2026 is a sensible window to lean refurbished for the bulk of the estate and reserve new-build spend for the workloads that truly justify it.

Sources

Every figure in this article traces to the sources below.

• •TrendForce — DRAM contract price rises Q1/Q2 2026 and fab capacity timeline
• •Avnet, EnkiAI — AI data centres' share of high-end DRAM consumption
• •GreenNode, EnkiAI, Polaris Market Research — wafer capacity reallocation to HBM/server DRAM
• •Hivenet — refurbished server savings up to 70%
• •Comtek Network Systems — refurbished server savings 30-50% and DDR5 volatility
• •ReluTech, Comtek Network Systems — OEM lead times 4-8+ weeks
• •ReluTech — OEM system price hikes 15-20% Q1 2026 and Gartner combined DRAM+SSD forecast
• •Wecent — DDR5 as standard for AI/dense workloads, DDR4 for legacy and budget systems
• •Enterprise IT Lifecycle Policy — four-to-five year server refresh cycle
• •US Cloud — Windows Server 2012 ESU coverage ends October 2026