UK computer prices are still falling. That should not be happening.
The memory chips inside them nearly doubled in price over three months. TrendForce reported on 1 June that conventional DRAM contract prices climbed 93 to 98 per cent during the first quarter of 2026. Yet the Office for National Statistics price index for personal computers stood at 54.0 in June, down 3.9 per cent on the same month last year.
Dispatch Times pulled the full ONS series — 138 months of it — to work out how long that gap can hold. The answer, based on what happened during the last memory shortage, is that it closes quietly rather than dramatically. And two months in the past year suggest it has already started.
What the UK data actually says
The relevant measure is ONS series L7GT, the consumer price index for personal computers. It runs monthly from January 2015 and is rebased so that 2015 equals 100. Over the 126 months for which a year-on-year comparison exists, it has fallen in 115 of them. The average annual change across the whole series is minus 5.7 per cent.
That near-permanent decline is the first thing to understand, because it is not really about prices. The index is quality adjusted. When a £700 laptop is replaced on the shelf by a £700 laptop with twice the storage, ONS records that as a price cut. So a falling index means buyers are getting more computer for their money, not that the till receipt is shrinking.
The clearest way to see that adjustment at work is to put the computer index next to its neighbour. ONS series L7GU covers accessories for information processing equipment — keyboards, mice, cables, the things that barely improve from year to year. Over the same eleven years that index has moved from 100 to 89.4, an average of minus 0.9 per cent a year. Computers went to 54.0. A mouse is still a mouse, so there is almost nothing for the statisticians to adjust. A laptop is a different machine every eighteen months.
Read that way, L7GT becomes a measure of how fast component improvements are reaching British buyers. And that is exactly what a memory shortage should slow down.
The last shortage left a fingerprint
DRAM prices spiked once before in the life of this series, through 2017 and into 2018. The UK index did not rise then either, in any sustained way. What it did was flatten, and it flattened hard.
Across 2017 the index averaged a fall of just 0.8 per cent a year, against a long-run norm of 5.7 per cent. For twelve months, British computers effectively stopped improving in value. It is the only year in the entire record where that happened.
The monthly detail is starker still. In 126 months of data the index has registered a year-on-year rise on only eleven occasions. Five of those eleven fell inside 2017.
| Year | Average annual change |
|---|---|
| 2016 | −7.9% |
| 2017 | −0.8% |
| 2018 | −5.6% |
| 2019 | −5.9% |
| 2020 | −6.3% |
| 2021 | −3.3% |
| 2022 | −6.0% |
| 2023 | −3.8% |
| 2024 | −10.7% |
| 2025 | −7.2% |
| 2026 (to June) | −5.2% |
So the fingerprint of a memory shortage in British data is not consumers paying more. It is consumers stopping getting more. Manufacturers hold the price and quietly stop upgrading the specification, and the index records that as the decline running out of steam.
Two months have already broken the pattern
Here is the part that has not been reported. Of those eleven year-on-year rises in eleven years, the two most recent are November 2025, at plus 1.5 per cent, and April 2026, at plus 1.8 per cent.
Before November 2025, the index had not risen since August 2023. Before that, March 2021. These events are rare, and two of them arriving within six months of each other is the sort of clustering that last occurred in 2017.
April’s plus 1.8 per cent is the largest year-on-year increase the series has recorded since the last shortage. Only one month in the entire eleven years beats it: August 2017, at plus 2.7 per cent.
Two months do not make a trend, and the monthly figures are noisy — May 2026 came in at minus 10 per cent, one of the sharpest falls on record. The year to date still averages minus 5.2 per cent, close to the long-run norm. On the balance of the evidence the shortage has begun to show up in British data but has not yet worked through to a 2017-style stall.
Why the memory went
Samsung, SK Hynix and Micron between them control more than 95 per cent of global DRAM production, and all three have moved capacity towards the stacked memory that AI accelerators need. High-bandwidth memory earns three to five times the revenue per wafer of the standard chips that go into consumer machines, which leaves little reason to make the cheaper product.
The scale of the pull is hard to overstate. Deloitte estimates the four largest cloud providers will spend around $725bn on capital projects this year, most of it on AI infrastructure. Micron made the direction of travel explicit in December when it announced it would retire Crucial, the consumer memory brand it had run for 29 years.
That last decision matters more than it sounds. Crucial was one of the few routes by which ordinary buyers could purchase memory directly from a manufacturer. Retiring it is a supplier telling the consumer market, in public, that it is no longer the priority.
Where it has already surfaced
Games consoles show the effect more plainly than PCs, because they are sold at a fixed specification and cannot absorb a cost rise by quietly downgrading the parts. There is no equivalent of shipping the same laptop with 8GB instead of 16GB. The price has to move instead.
Sony’s PS5 Digital Edition now lists at $599.99 in the United States, $200 above its launch price, and Microsoft has moved the Xbox Series X to $650. Valve launched its Steam Machine in June at $1,049, roughly $300 above the original target, and blamed memory costs for the difference. Supply was tight enough that launch stock went out through a randomised reservation queue rather than open sale.
The effect is visible in volumes too. IDC expects the global PC market to contract 4.9 per cent this year, with smartphone shipments down about 2 per cent. That is a market where manufacturers have chosen to build fewer machines rather than sell them at a loss.
How long this lasts
A new fabrication plant takes 18 to 24 months to build and longer to reach usable yields. Micron’s newest site is not expected to produce meaningful volume until around the third quarter of 2028, and Deloitte’s July analysis suggests the crunch may not ease until 2029.
Demand is still climbing on the other side. TrendForce has raised its 2026 AI server forecast twice this year, most recently to 31 per cent growth. The same buildout is showing up in the labour market, which we covered in our running list of 2026 AI layoffs.
The 2017 comparison offers one piece of reassurance. That shortage suppressed the index for roughly twelve months, then the decline resumed: 2018 came in at minus 5.6 per cent, almost exactly the long-run average. Shortages in this market have historically been interruptions rather than turning points. The difference this time is that the competing demand is not another consumer product cycle, it is datacentre construction with a multi-year runway behind it.
What it means if you are buying
The practical consequence of a quality-adjusted index is that waiting stops paying. In a normal year, holding off six months gets you either the same machine for less or a better machine for the same money. That is what a 5.7 per cent annual decline buys you.
In a stalled year it gets you neither. The 2017 buyer who waited twelve months received a specification improvement of 0.8 per cent for their patience. If 2026 follows the same path, the machine on sale next spring will be much the same as the machine on sale now, at much the same price.
The corollary is that memory capacity is the specification worth paying for now, because it is the one being rationed. A configuration upgrade that costs £80 today is the part of the machine most likely to be quietly reduced on next year’s equivalent model.
What to watch
The useful signal for British buyers is not the shelf price, which moves for a dozen reasons at once. It is the annual rate on L7GT, published monthly by ONS. If that figure climbs towards zero and stays there across two or three consecutive releases, the shortage has reached UK retail and 2026 is repeating 2017.
If it settles back towards the usual 5 to 6 per cent decline, British manufacturers have absorbed the cost and buyers can carry on as before. The two rises already on the board make the first outcome more likely than it looked six months ago, but they do not settle it.
Dispatch Times will update these figures with each ONS release, as we do with our tracking of UK jobs data on AI displacement. The July index is due next month.
