UK IT leaders configuring servers in 2026 face a market that is short on exact specifications, not necessarily short on boxes. CDW UK expects the current memory-driven shortage to run for at least 18 months, into mid-2027, while public procurement guidance says any component quoted above 26 weeks should be treated as a shortage risk and multi-sourced immediately. That changes the configure-to-order (CTO) versus off-the-shelf question from a taste preference into a genuine cost-and-risk decision. This analysis sets out what's actually driving delays, what UK buyers should specify differently right now, and when a stock or refurbished servers route beats waiting for a bespoke build.
View the data behind this chart
| Layer | Detail |
|---|---|
| AI demand for global memory chips | Up to 70% of 2026 memory chips — Accuris |
| HBM demand vs fabrication capacity | ~70% of fab capacity for HBM — CDW UK |
| Micron core-customer fulfilment | Only 55–60% of demand met — Arctiq |
The 2026 landscape: allocation tightening, not a simple outage
Softcat's March 2026 update frames the current environment precisely: this is allocation tightening rather than a global product outage. In practice that means UK buyers can usually still get a server — but not necessarily with the exact CPU stepping, memory density, SSD size, or RAID/NIC combination they specified in the original request for quote.
CDW UK's assessment adds a timing dimension that should shape every procurement plan made this year: the shortage is expected to last at least 18 months, running into mid-2027. That is long enough that 'wait it out' is not a credible strategy for anything with a hard deployment deadline in 2026.
Why your CTO build is taking longer — and costing more
The single biggest force behind CTO delays and price inflation in 2026 is memory, and the cause is AI, not general server demand. Accuris reports that up to 70% of all memory chips produced globally in 2026 will be consumed by AI data centres — a separate but reinforcing data point from CDW UK's observation that AI demand for high-bandwidth memory has absorbed around 70% of global fabrication capacity. These are two distinct measures (chip output versus fab capacity) but they point the same direction: standard server DRAM is competing for the same wafers as AI infrastructure.
Even large buyers aren't immune. Arctiq reports that Micron has publicly stated it can fulfil only around 55–60% of core customer demand — meaning allocation shortfalls are hitting established accounts, not just smaller resellers. The knock-on cost effect is visible in Arctiq's account of HP's early-2026 disclosure that memory rose to roughly 35% of its PC bill of materials, up from 15–18% just a quarter earlier. That is a single OEM's bill-of-materials ratio, not a market-wide average, but it illustrates how fast memory-driven cost inflation can move through a build sheet — and CTO servers, where buyers actively choose memory density, are directly exposed to that swing.
UK component lead-time snapshot, mid-2026
There is no single published UK figure for 'CPU lead time' or 'GPU lead time' in weeks — what does exist, and what UK buyers can act on today, is category-level data on the semiconductor and controller parts that sit inside every server: power-delivery boards, baseboard management controllers, and discrete logic. GlobX's 2026 European OEM guide and SimplyTronix's 2026 procurement guide both publish concrete ranges for these categories, and Accuris has an independent, more recent benchmark for the broader semiconductor market.
The pattern below is what a CTO configuration is actually waiting on when a supplier says 'component allocation' rather than 'assembly time.'
How shortages bite differently by server build type
Not all CTO builds carry equal risk in 2026 — the shortage's biggest driver, memory, hits AI-oriented and storage-dense configurations far harder than general-purpose compute nodes. CDW UK links the current shortage explicitly to 'a global DRAM and NAND crunch' driven by AI demand for high-bandwidth memory, which means GPU-heavy AI servers (dependent on HBM and high DRAM density) and high-density storage servers (dependent on NAND) are drawing on the same constrained supply from different angles.
For GPU-heavy AI server builds, the exposure is most direct: Accuris reports that up to 70% of all memory chips produced globally in 2026 are being consumed by AI data centres, and CDW UK puts HBM's share of global fabrication capacity at around 70% too. CTO configurations built around AI workloads should expect the tightest allocation and the most price volatility on memory line items.
High-density storage servers sit on the NAND side of the same crunch CDW UK describes, so bulk-capacity storage specifications should be treated with the same shortage-risk lens as memory-heavy compute — don't assume storage-only builds are insulated just because they're not GPU-bound.
General-purpose compute builds are comparatively less exposed to the memory squeeze but are not risk-free: they still depend on the same power-management ICs (40–52+ weeks per GlobX) and management controllers/MCUs (26–40 weeks) that sit inside every server chassis, regardless of workload type.
Strategic CTO: configuration choices that actually beat the delay
None of the mitigation tactics below are guesswork — they follow directly from the lead-time and allocation data above. Applied at spec stage, they reduce both wait time and the chance of a mid-project price change.
- •Treat any quoted lead time above 26 weeks as a shortage-risk flag per GlobX's 2026 guidance, and get the same spec priced through at least one alternative channel before committing.
- •Build a 15–25% cost buffer into any CTO quote for memory-heavy or storage-heavy configurations, as Arctiq recommends — mid-quarter price revisions are now a normal feature of quoting, not an exception.
- •Ask for allocation-confirmed pricing, not just an estimated lead time. Given Micron's reported 55–60% core-customer fulfilment rate, a quote without confirmed allocation is a forecast, not a delivery date.
- •Where a design can use a general-purpose controller instead of an automotive-grade or legacy-node part, specify it: SimplyTronix puts general-purpose MCUs at 12–20 weeks against 40–55+ weeks for automotive-grade equivalents.
- •Plan and budget from the expected shipping date, not the quote date — Softcat's 'allocation tightening' framing means the spec you approve today may need a substitution decision before it ships.
View the data behind this chart
| Phase | Starts (week) | Duration (weeks) |
|---|---|---|
| Safe lead time | 0 | 26 |
| Shortage risk | 26 | 52 |
New CTO vs off-the-shelf vs refurbished: the 2026 calculus
UK resellers and distributors are consistently flagging the same trade-off in 2026: stock servers are available, but often only with compromises on CPU stepping, memory density, SSD capacity, or RAID/NIC options, and on delivery timing that's outside your control. A configure-to-order build secures the exact specification you need, but that precision comes with more exposure to allocation delay and mid-project price movement.
The practical rule that follows: reserve CTO for workloads that genuinely need a fixed specification — compliance-mandated hardware, long-lifecycle platforms that must match an existing fleet, or performance-critical builds where substitution isn't acceptable. Use stock configurations for refreshes that can tolerate a substituted CPU stepping or a different memory module without breaking anything. And where the workload allows it, quality-assured refurbished servers are a genuine third lane — see our detailed breakdown of new vs refurbished server options for how the DRAM shortage specifically changes that maths in 2026.
Engaging suppliers in a volatile allocation market
The most useful question to ask any UK distributor or reseller in 2026 isn't 'what's the lead time' — it's 'is this allocation confirmed, and against which order?' Given that even Micron's core customers are only getting 55–60% of requested volume, an unconfirmed lead time is essentially a placeholder.
It's also worth documenting lead times per component rather than per finished system. A single quoted delivery date can hide a power-management IC sitting at 40–52+ weeks inside an otherwise 'in-stock' chassis. Buyers who ask suppliers to itemise the constrained parts get a much clearer picture of real risk — and a stronger negotiating position if a like-for-like substitute exists. If your team doesn't have the internal bandwidth to run this level of component-level scrutiny across every order, that's exactly the gap that dedicated IT procurement services and structured specification support — see how to spec a server — are built to close.
Beyond 2026: what changes by mid-2027
CDW UK's 18-month shortage window points to mid-2027 as the earliest realistic point at which memory allocation for standard server builds might loosen. Until then, the safest planning assumption for UK IT leaders is that the current allocation-tightening environment — not a return to pre-shortage stock availability — is the baseline for any CTO project scheduled this year.
Sources
Every figure in this article traces to the sources below.
- •GlobX — 2026 European OEM shortage risk thresholds and IC lead times
- •Accuris — semiconductor lead times and AI memory consumption, March 2026
- •CDW UK — HBM demand, fabrication capacity, and 18-month shortage outlook
- •SimplyTronix — 2026 MCU lead-time procurement guide
- •Arctiq — memory BOM inflation, Micron fulfilment, and cost buffer guidance
- •Softcat — March 2026 global component shortage update