PC Shipments Fall 20.1% as Memory Shortages Raise UK Hardware Risks

PC shipments chip shortage 2026: the headline numbers

According to IDC data reported by Tom's Hardware, global PC shipments dropped 20.1% year-over-year in Q3 2026 — a fall of 15.8 million units and a sharp acceleration from the 3.8% decline recorded just one quarter earlier. Shipments were also down 9.1% quarter-on-quarter, breaking the usual seasonal pattern where Q3 normally outperforms Q2.

The top three vendors absorbed the worst of it: Lenovo shed 22.6% of shipments, HP fell 30.9%, and Dell contracted 25%. Apple (-11.3%) and Asus (-8.6%) fared comparatively better, while the rest of the market dropped 14%. Despite the losses, Lenovo, HP and Dell retained their market-share lead at 23.8%, 16.5% and 12.1% respectively.

IDC's Jitesh Ubrani attributes much of the drop to vendors and channels over-ordering inventory in the first half of 2026 to get ahead of price rises — a pull-forward that contributed to the Q3 result by distorting normal seasonality, and which has now left channel partners cautious about carrying excess stock into a market where high prices are already suppressing demand.

Why AI datacentre demand is starving client and edge hardware

AI-infrastructure demand is a major contributor, alongside inventory effects, elevated prices and weaker demand. Hyperscalers' spending on HBM to feed AI GPUs has pulled memory manufacturers away from standard DDR5 production. Tom's Hardware reports that Micron retired its Crucial consumer RAM and SSD brand in late 2025, which it links to a redirection of capacity toward data-centre output — a motive not confirmed by a direct Micron announcement.

Tom's Hardware reports increases of up to 500% in some memory products, including a reported $3,399 price for 128GB of DDR5. Separately, TechRadar reports DRAM prices up 171% year over year and says relief is not expected until 2027–2028.

This isn't confined to client devices. Server shipments actually grew 15.4% year-on-year in Q2 2026, but GPU-accelerated server shipments fell 10.8% in the same period — an indication that supply constraints are affecting GPU-specific deployments, even for organisations with budget already approved.

Server and GPU availability risk for UK enterprise buyers

For UK procurement teams, the asymmetry matters. GPU-heavy AI server builds carry the highest price and allocation risk: average selling prices for GPU-accelerated systems rose nearly 44% to $170,200 in Q2 2026, and The Next Platform estimates that a four-GPU node built on cheaper PCIe cards such as Nvidia's RTX Pro 6000 costs roughly $90,000–$125,000, while eight-way B200/B300 nodes run to about $400,000–$500,000. GPU-accelerated systems represented approximately 53% of server-market revenue in the same quarter.

Conventional servers, storage arrays and edge systems face a different but still material exposure — through shared DRAM, NAND and SSD supply rather than direct GPU scarcity. Enterprise-grade NAND is facing tighter availability and longer lead times as AI workloads consume more of the available capacity.

Tracking these shifts in real time is now a procurement discipline in its own right; buyers comparing vendor quotes should consult a server lead-time tracker alongside pricing data before committing budget to a specific refresh window.

When might the chip shortage ease — and who disagrees

Not every industry voice agrees on the timeline. Acer CEO Jason Chen has pushed back against the gloomier forecasts, arguing that memory-chip makers are overstating the crisis to protect margins. He expects PC prices to fall by the latter half of 2027 as Chinese memory capacity comes online, though he concedes that some high-end memory and specialised components may remain constrained for longer.

That's a more optimistic read than some other forecasts. Some forecasts extend the disruption to 2029, while other reporting expects meaningful relief in 2027–2028. For UK buyers, the practical takeaway is the same either way: plan for elevated pricing and allocation constraints through at least 2027, with the possibility — not the certainty — of easing on standard parts after that.

Procurement strategy: new hardware, refurbished, or cloud capacity

With new-build lead times stretching and prices holding well above last year's levels, UK buyers are increasingly weighing three routes to capacity rather than defaulting to a straight refresh. Refurbished and redeployed hardware may reduce exposure to current component pricing, though it limits access to the newest GPU silicon — a trade-off worth working through via a new versus refurbished servers comparison before locking in a 2026 refresh cycle. This is practical procurement guidance rather than a sourced market finding.

Cloud and GPU-hour contracts are emerging as a reported alternative to capital purchase for AI workloads specifically. A five-year Oracle–Tencent arrangement covering around 100,000 advanced AI chips was valued at roughly $7bn, reportedly working out to about $1.60 per GPU-hour — a reported estimate that compares, indicatively rather than like-for-like, with the roughly $2.80 per H100 GPU-hour cited for one-year contracts elsewhere.

Whichever route is chosen, the underlying advice from industry analysts is consistent: shift from just-in-time refreshes to advance capacity reservation. Buyers assessing whether to commit now or hold out for better pricing should weigh the buy now or wait trade-offs against their own refresh deadlines, and check current memory and SSD price surge data before finalising a spec with Dell, HPE or Lenovo configurators.

UK-specific bottlenecks beyond the chip shortage

Even once hardware is secured, UK deployments face additional scheduling risk that has nothing to do with silicon. Grid-capacity upgrades for new data-centre sites can, according to industry reporting, involve manufacturing lead times of up to five years for high-voltage transformers and switchgear in some cases, while high-density GPU racks have reportedly faced waits of up to a year for liquid-cooling plumbing and manifolds in some cases.

That stacks compounding delay on top of compounding delay: a GPU-accelerated server that clears procurement may still sit idle awaiting power infrastructure or cooling capacity. Buyers configuring new builds through Dell, HPE or Lenovo server configurators should factor these downstream constraints into project timelines now, rather than discovering them after hardware has shipped.