PC Sales Fall as Memory Costs Soar
The machines your clients already own are turning into the machines they’ll keep.
Start with Omdia. Global shipments of desktops, notebooks and workstations fell 21 percent year over year in the third quarter of 2026. Demand didn’t disappear. Prices went up. Average selling prices jumped by double digits, and buyers held back. Omdia’s principal analyst, Ben Yeh, puts a number on what changed inside the box. Memory and storage used to be about 15 percent of a PC’s component costs. They’re now close to 40 percent.
That isn’t one firm’s read. IDC counted a 20 percent drop for the same quarter, to about 63 million units. Two research firms, counting separately, landed within about a point of each other. Omdia expects the decline to continue: it forecasts another 24 percent drop in the fourth quarter and 7 percent in 2027. Yeh said the current slide “may be just the beginning of a new downward cycle.” IDC’s Jitesh Ubrani expects some short-term discounting as distributors clear excess inventory, but doesn’t expect prices to fall back to where they were.
Now look at the other side of that sale. Micron reported record DRAM revenue of 39.8 billion dollars for its fiscal fourth quarter, up 343 percent from a year earlier. Its flash memory revenue rose 526 percent. Fewer computers are being sold, while the company selling the memory inside them is posting record revenue.
And the people who build PCs are already planning around it.
DDR4, the memory standard that launched in 2014, is coming back. Gigabyte says two of its motherboard lines will support new processors for an older Intel socket that runs DDR4, expected in early 2027, and AMD is refreshing its DDR4-era platform too. The reason is price. A 32-gigabyte kit of the newer DDR5 memory now runs about 620 dollars. The DDR4 equivalent is about 260. In plain terms, the industry is designing new machines to run on old parts.
So the numbers all point the same way. New hardware costs more, fewer people are buying it, and the companies that make computers are planning for buyers who keep what they have.
The machines are staying. The question is what keeping them actually costs, and where that cost shows up.
Where an Aging Fleet’s Cost Hides
An old computer doesn’t fail all at once. It gets expensive slowly.
That has been documented for years. When Texas’s Legislative Budget Board reviewed replacement schedules for state agencies, it concluded that PCs kept beyond about four years cost more to support and fail more often, which is why the state recommends a four-year lifecycle. That cost doesn’t arrive as an invoice. It arrives as friction: a laptop that takes longer to wake, an app that hangs, a machine that needs one more restart.
Friction is exactly what people don’t report. Buchanan Technologies is an IT services firm, and its report argues for a support model it sells. Its IT Experience Gap survey found that 82 percent of workers had decided an IT problem wasn’t worth reporting. Seventy-two percent had used their own tool or AI instead of going to IT. And 80 percent had seen a ticket marked resolved while they believed the problem was still there. Buchanan’s CEO, Mike Werblun, summed it up: every workaround is a small decision someone made because the official route felt slower.
Now look at what happens to the problems that do get filed. At the Canalys Channel Forum, Omdia research director Robin Ody told MSPs that some of them are already looking at triage and resolution of more than 90 percent of their tickets. At HPE, networking chief Rami Rahim says 70 to 80 percent of tickets in his self-driving networks already need no human, and he expects that to reach all of them within two to three years. For a failed device, he described the system ordering the replacement part, in his words, “without your knowledge.”
Put those together. The aging machine generates friction. The person using it works around most of that friction. What they do report gets triaged, often resolved by software, and sometimes marked closed when it isn’t fixed. Each step is reasonable on its own. Together, they move the cost of an old fleet out of the one place the customer reliably looks, which is the ticket report.
So the cost is real, and it grows. It’s just paid in lost minutes and quiet workarounds, not in anything the client sees.
And that puts the MSP in an awkward spot when it’s time to talk about replacing anything.
Refresh Now or Extend on Purpose
For an MSP, all of this lands in one meeting: the refresh conversation.
You can see the fleet. Your tools know the purchase date and warranty status of every machine you manage. The client sees two different things: a quote that’s higher than the last one, and a support report that looks calm. When the person you’re advising sees rising prices and no pain, “let’s wait” is a perfectly rational answer. The recommendation you used to win with a broken laptop now has to be won with a spreadsheet.
And the people who are supposed to forecast this can’t agree on it. At the Canalys Forums in Barcelona, Steve Brazier, who founded Canalys, told partners to “stock up your warehouse now,” arguing that anything with a semiconductor in it will be worth more over the next twelve months. Minutes later, Omdia chief analyst Jay McBain said it was the first time in nineteen years he’d disagreed with the opening keynote. He called hardware a point-in-time opportunity, pointing to geopolitics, price inflation, supply problems and tariffs. A partner executive in the room summed it up: nobody really knows, even the researchers.
That uncertainty shapes the client conversation. You can’t honestly tell a client that waiting will make the machines cheaper, and you can’t promise it won’t. What you do know is the age of what they’re running, and that keeping it longer doesn’t make the refresh bill go away. It moves the bill, and it changes who decides when it gets paid.
So that’s the choice. Refresh now, at today’s prices, on your schedule, and accept the higher number while you can still plan around it. Or extend on purpose: set a target life for each class of device, put a budget against the support and replacement costs the extra years will bring, and walk the client through that number before a failure does it for you.
What doesn’t work is the default. That’s extension by inaction, where the fleet ages quietly until a dead machine forces an out-of-cycle purchase at whatever memory costs on that day.
Which brings us to the conversation itself.
The fair objection is that walking into a client meeting with a refresh recommendation, right when prices are high and the support reports look calm, sounds like a sales pitch. That’s exactly why the conversation can’t lead with a quote. As the advisor, bring the age of their fleet, the point where each group of machines crosses its target life, and the cost of each path: replace now, or extend with the support and failure risk budgeted. When the client can see both numbers, waiting becomes a decision they made, not one the fleet made for them.
What to Consider
- Lead with age, not tickets. Put the fleet-age view at the top of the next review: device counts by age band and warranty status. The ticket report is the document telling the client nothing is wrong. Keep the ticket metrics, but present them after the client has seen what they’re running.
- Price both paths, including the one you don’t recommend. Put the refresh quote next to a costed extension: the target life for each device class, the extra support time older machines take, and a reserve for failures out of cycle. An advisor who prices the cheaper path honestly is the one the client believes on the expensive one.
- Surface the friction nobody files. Before the meeting, pull the signals that never become tickets for the oldest group of machines: boot times, crash and restart counts, warranty status. That turns “your machines are old” into specific counts from the client’s own devices, which they can weigh.
If this trend continues: If Omdia’s forecast holds and shipments keep falling through 2027, the MSPs that put a costed lifecycle plan in front of every client before then will be setting the refresh schedule. The ones that wait will be setting it from failures.