Someone had ordered a laptop and asked for 64 gigabytes of memory. They paid a deposit. Then, in July of 2026, before it shipped, before they had done anything at all, the order changed by itself. The website now said 32. The price had even dropped a little, as though the smaller machine were a gift.
The email that followed was polite and short. "We've had to update your memory configuration from 64GB to 32GB, and have lowered the price on your pre-order as a result." There was an exit, if they wanted it: cancel for a full refund of the deposit. One reviewer, browsing the brand's fan forums, wrote that it felt grim to watch even its most devoted fans say they were being priced out and cancel their orders. A buyer put the unease plainly: component costs can move, that's understandable, but charging for a pre-order, sending a confirmation, and then changing the configuration afterward doesn't sit right.
I have been turning this over for days, and I keep failing to find the villain the story seems to want.
Because the company that sent that email had built its whole name on the opposite promise. It sold a laptop you were supposed to open, add to, and keep: memory you could swap, a screwdriver in the box, little codes on the parts that linked to repair guides. Its pitch, more or less, was that when you place an order you get the product you ordered at the price you ordered it at. It seems to have meant it. What happened was that its memory supplier came back with a quote more than double the batch before, and the company decided it could not quietly eat that and stay alive. So it shrank the orders it could not fill, lowered the prices to match, and offered everyone a way out.
The buyers were not chasing a luxury number. One person, moved from 32 down to 16, wrote that 16 was not enough for their work, and that they had not paid to reach 32 "for the fun of it." Another, dropped from 64 to 32, wrote that they had intentionally not chosen 32. They had bought a specific machine to do a specific thing, and the thing got cut while their money sat in the company's account.
Here is what I did not expect to be true. The first bill for the AI boom that reaches ordinary people is not landing on the ones who use AI. It is landing on people who wanted nothing to do with it, as a worse deal on a thing they thought they already owned.
A preorder got smaller after the money changed hands. Electronics are the one corner of ordinary life where we have been trained to expect the reverse. For a long time the same money kept buying more: the cheapest recorded memory fell from over a million dollars a terabyte around 2000 to about a thousand dollars a terabyte by 2023. We built a quiet faith on that slope. Wait, and you get more. Buy later, pay less.
So walk the boring explanations down, one at a time, and see how far each one gets.
Greed? The company lowered the price when it cut the memory, and handed back deposits on request. That is not the shape of a squeeze.
A single weird part? This is the strongest of the boring answers, and it is partly right. The memory in question was a new, still-rare kind of upgradeable module, and rare parts spike. That is real, and it explains why this one company got hit so hard. But it does not explain the rest of the aisle. Within the same stretch of weeks, a console maker raised prices worldwide and said, out loud, that its storage and memory now cost more than 2.5 times what they had, with another doubling expected by the fall of the next year. Months before that, a headset maker had raised prices and named "memory chips" directly. And a large PC maker was reported, through a leaked internal document, to have warned its business buyers that placing an order today would not lock today's price. One odd module explains one company's bad week. It does not explain a whole shelf moving at once.
That leaves the explanation I could not make go away, and it is not really about gadgets. It is about who you are standing next to when you buy one.
The memory in a data center rack and the memory in your laptop are not the same part. But they are made by the same handful of companies, drawn from the same balance sheets and roadmaps, subject to the same allocation decisions, competing for the same year of finite factory capacity. And right now, nothing else on the market wants memory the way the machines being built for AI do. A single top-end accelerator can carry 141 gigabytes of it, more than twice the 64 the biggest laptop in this story was going to have; one rack built for AI can hold more than thirteen terabytes, hundreds of times what sits on your desk. The buyers of those racks can pay far more than you can, because for them the memory is not a line item on the build. It is the product. So when one constrained supply base faces two kinds of buyer, the scarce capacity goes where the money is loudest. The laptop maker gets a worse quote, or the same price for less. You get 32.
You have felt this before, somewhere far from a checkout page. A town gets a new employer that pays triple. You never apply. But the plumber costs more now, because the plumber can get the new money too, and your rent climbs, because someone who values a roof more than you just moved in down the street. You did not lose a bidding war. You found out, after the fact, that you had been in one the whole time. That is the mechanism, and it is old: a buyer who wants the shared thing more than you do can reprice it for you without ever once touching your copy of it.
You do not have to use AI to pay for it. You only have to want the same scarce thing as someone who wants it far more than you.
There was a smaller, stranger proof of this sitting inside the story the whole time. One publication's review unit of that same laptop was meant to cost $2,299 when it landed on the reviewer's desk. By the time the review went out, the very same machine cost $3,099. Nothing about it had changed but the number.
The console story rhymes with the laptop one. Consoles are usually sold at a loss, the machine kept cheap so the games can be sold dear, which means component costs go almost straight to the sticker. So the price went up: $100 on one model, $150 on another, the biggest model dropped entirely. And alongside the increase, the company pointed customers toward financing, toward trade-ins, toward its own certified used machines at a discount. The old, cheaper box became a feature it advertised. When a company starts selling you last year's hardware as the deal, something in the ground has shifted.
There used to be a sound.
If you upgraded the memory in an older laptop, you did it with your hands. You shut the machine down and, if it had been running, you waited ten minutes for the inside to cool. You turned it over, took out ten screws, touched bare metal to bleed the static out of your body. You pushed two small levers and the memory rose to meet you at an angle. You held the new stick by its notches, kept your fingers off the gold, lined it up, and pressed until it clicked. That click was the entire bargain in one sound. The machine was not finished when you bought it. It was yours to add to, later, when you had the money or the need.
Then the industry soldered the memory down for thinner machines, and the click went away, and the advice became: buy all you will ever need now, because there is no later. A few stubborn companies kept fighting for the click. The very laptop in this story used a new kind of module meant to bring it back, low-power memory you could still unscrew and replace. And mechanically, it worked. The joke, the one the reviewers caught, is that the swap arrived exactly as the module to swap in climbed to $800, when you could find one for sale at all. The machine stayed repairable. The repair became the expensive part.
The strongest case against everything I have said is simple, and I think it is partly right. Memory has always been cyclical. It spikes, the makers overbuild, it crashes, and the crash is as reliable as the spike. A major supplier says so in its own filings: prices have swung by nearly half in a single year, and if this demand ever cools, an oversupplied market could send them right back down. New factories are being built. So maybe this is just an old cycle in a new costume, and by 2028 memory is cheap again and this reads as a small panic about a passing thing.
Maybe. But two things keep me from letting the cycle argument close the case. The first is that "temporary" here is measured in years. The people who watch this closely do not expect meaningful new storage capacity until late 2027 or 2028, and the new memory factories are on the same slow clock. That is several back-to-school seasons, a couple of holiday console runs, the whole useful buying life of a machine bought today. The second cuts deeper. Even when the cycle turns, the mechanism will still be sitting there. The real lesson was never "memory got expensive in 2026." It is that the cheapness we treated as a law was only ever a truce, holding for as long as nobody richer wanted the same thing badly enough to bid. AI is simply the buyer big enough to break that truce where you can see it happen, on your own order page. It will not be the last.
I cannot tell you how much of any single price rise is AI demand and how much is ordinary shortage, or a currency, or a pricier chip, or a company guarding its margin. Nobody outside the supplier can; those numbers are not public. The direction is not in question. The size of any one arrow is.
So the next time you buy something with a memory in it, which is now nearly everything, notice what you are actually paying for. The machine, yes. Also a place in a line you did not know you were standing in, behind buyers you will never see, who can outbid you for the one part you both need. The thing used to get quietly better while you owned it. Now it can get quietly smaller before it even arrives, hand you a small refund for the difference, and call the missing part a courtesy.
There used to be a click. Now there is a number on a screen, and the new knowledge that it can go down as easily as up, and that whether it does was never really yours to decide.