Your Laptop and Your AI Tools Just Got a Memory Problem

A shop owner ordering a replacement laptop in August 2026 opened two browser tabs, one from a quote saved in spring and one live, and found the same machine had gone up by a chunk that made no sense for a year old model. Nothing about the laptop had changed. What changed sits two steps back in the supply chain, in the memory chips inside it, and the reason those chips cost more is the same reason your AI subscription might cost more next year. Both draw from one pool, and artificial intelligence is draining it fast.

This is the least glamorous AI story of the year and one of the few that reaches into a small business bank account directly. It has nothing to do with a new model or a clever prompt. It is about the physical chips that hold data while a computer works, and a squeeze that the companies making them now describe in years.

What is actually happening to memory prices?

Memory makers have redirected their factories toward the high margin chips that AI accelerators need, and that has starved the ordinary memory that goes into laptops, phones and standard servers. The result is a price surge that analysts call the sharpest on record. According to The Register's report on TrendForce forecasts, contract prices for DRAM, the working memory in nearly every device, were revised in a single week from a 55 to 60 percent quarterly jump up to a 90 to 95 percent quarterly surge, with PC memory expected to roughly double from the previous quarter. Low power memory of the kind used in phones and thin laptops was pegged at around a 90 percent rise, described as the steepest in its history.

It is not only working memory. The same report put NAND flash, the memory that stores your files on a solid state drive, up 55 to 60 percent in the same quarter. So a shop planning to add storage to a server, buy new drives for a backup rig, or refresh a fleet of tablets faces the increase twice over, once in the RAM and once in the storage. For a business that treats its computers as three year assets, an increase of this size resets the math on when a machine is worth replacing.

Those are contract prices, the rates that device makers pay, not the retail sticker. The retail effect arrives later and more quietly, folded into the price of the next model. Looking across the whole period, J.P. Morgan Global Research estimates DRAM prices will have climbed more than 400 percent from the start of 2024 to the end of 2026. The bank has a name for what this does downstream, chipflation, and it lists the consumer targets plainly: phones, computers, smartwatches and TVs. Its figures show the import price index for computers and parts already up 37 percent, and the consumer price index for software and accessories up 23 percent since the end of 2024.

Here is the chain in five steps, from a data center you will never see to the invoice you will.

Why is one product line eating everyone else's supply?

A handful of companies, Samsung, SK Hynix and Micron, make almost all of the world's memory, and they can only make so much at once. When AI accelerators need vast amounts of a premium memory type, and that memory sells at a far better margin than the standard chips in a laptop, the factories tilt toward the premium line. Building a new factory takes years and costs billions, so supply does not simply expand to meet the new demand. The chips that lose out are the everyday ones.

A quick vocabulary helps, since the reporting throws these terms around without defining them. The premium memory that AI accelerators need is called HBM, short for high bandwidth memory. The ordinary working memory inside a laptop or a register is DDR5. Both come off the same production lines, so every time a memory chip maker commits a wafer to HBM, that is a wafer no longer making DDR5. That single trade is the whole mechanism behind the AI memory shortage. It is why surging demand for Nvidia AI chips shows up first as higher AI server prices, and only later as higher laptop and cloud costs for the rest of us.

The size of the AI appetite is easy to underestimate. Modern AI server racks are memory monsters. When a single rack of the latest systems carries tens of terabytes of memory, and cloud providers are installing them by the thousands, the draw on the supply is enormous. This is the same dynamic that makes running your own models expensive, which is worth understanding if you have ever priced self hosting against an API. We broke down the moving parts of that decision in our look at why GPU rental prices vary more than tenfold for the same chip, and memory is now one of the loudest variables in that spread.

None of this is a temporary blip caused by a bad quarter. One equity analyst quoted by J.P. Morgan put it bluntly, saying the industry will stay in shortage for multiple years and that a timetable for resolution is difficult to set. When the people selling the chips tell you the squeeze outlasts the year, that is a planning input, not a rumor.

Will this reach the AI tools I actually pay for?

Short answer, yes, with a delay. The AI tools a small business rents, the chatbots, the copywriting assistants, the image generators, all run on servers packed with exactly the memory now in short supply. The cost lands first on the companies buying the hardware. The Decoder, reporting on Bloomberg's findings, wrote that servers built around Nvidia's newest AI chips are set to cost more than 15 percent more in many cases, for shipments beginning in early 2027, with the systems that need the most memory hit hardest. The outlet named the buyers who absorb that first: Amazon, Microsoft, Google and Meta, plus AI labs including OpenAI and Anthropic.

Those companies do not eat a 15 percent hardware increase quietly forever. It shows up eventually in the price of cloud compute and in the per use rates of the AI services built on top. You have already seen the opposite trend for a couple of years, where the price of calling a model fell steadily as competition heated up. We wrote about what that earlier drop changed for small sellers in the piece on falling AI API prices. The memory crunch is the first serious force pushing the other way. It may not reverse the decline outright, because software efficiency keeps improving, but it removes the easy assumption that AI only ever gets cheaper.

To feel where the money goes, it helps to know what a single AI request actually costs the provider in the first place. If you have never seen that math, our explainer on what a language model call costs in tokens shows why the hardware bill underneath it matters. When the servers get 15 percent more expensive, the floor under every token rises with them.

How much of my costs are we really talking about?

For a solo business the direct hit is not catastrophic, but it is real and it compounds across several lines at once. The table below pulls the reported figures into one place so you can see which pressures are already here and which are still arriving.

Read that column of timing carefully, because it is the practical part. The hardware you already own is not affected. The squeeze shows up at the moment of purchase and at the moment a subscription resets. That gives you two levers, and both are about timing rather than avoidance.

What can a small business do about it right now?

You cannot argue with a memory shortage, but you can stop it from catching you at the worst moment. A few moves cost nothing and buy you room.

First, treat any hardware you were going to buy in the next year as a decision to make sooner rather than later, while stock bought at older prices still sits on shelves. That applies to laptops, tablets used as registers, and any on site server. It does not mean panic buying gear you do not need, since idle hardware is its own waste, but a machine you were already going to replace in six months is cheaper to replace now than after the next contract reset flows through to retail.

Second, look at your recurring software costs and lock the ones you can. Many AI and software tools offer an annual rate that freezes today's price for a year. If a tool you depend on lets you prepay a year at the current rate, that is a hedge against a mid 2027 increase, and the saving is often the same as the monthly to annual discount you would take anyway. Do this only for tools you are sure you will keep, since prepaying for something you abandon is a loss, not a saving.

Third, when you choose an AI tool, weigh how predictable its pricing is, not only its headline rate. A per use meter that floats with the provider's own costs will drift upward as the hardware underneath gets pricier. A plan with a fixed monthly allowance you can plan around behaves better under this kind of pressure. This is one reason we price the MaShop builder in a flat credit system you can read in advance rather than an opaque meter, and you can see exactly how that works on our pricing page for the AI store builder. Whatever tool you pick, the test is the same, can you predict next quarter's bill from this quarter's usage.

Fourth, review the storage and cloud spend you rarely look at. Solid state storage is rising alongside memory, so the cost of keeping years of old backups, unused virtual machines, and duplicate copies of your catalogue is quietly going up. A short audit that deletes what you do not need and moves cold archives to a cheaper tier can offset a good part of the price drift, and it is work worth doing regardless of the shortage. The cheapest byte is the one you stop paying to store.

Fifth, do not over rotate. The memory shortage is a cost headwind, not a reason to stall your plans. AI still does more for a one person business than it did a year ago, and the productivity it buys usually dwarfs a few percent of price drift. The point is to plan around the increase, not to freeze because of it.

Is this the same as the old chip shortages?

It feels familiar, but the cause is the opposite of the one most people remember. The shortages of a few years ago came from disruption, factories idled and logistics tangled, so supply fell while demand stayed roughly normal. This one is a demand story. The factories are running hard and making record volumes of memory. The problem is that a single, hungry customer, the AI build out, is buying up the most profitable share and pulling capacity away from everything else. That difference matters for how you plan, because a supply disruption tends to snap back once the disruption clears, while a demand surge only eases when either the demand cools or years of new factory capacity finally come online.

The practical consequence is that you should not wait for a sudden return to normal. There is no blocked port to reopen here. The people building memory factories are moving as fast as capital allows, and that speed is measured in years. A merchant who understood the earlier shortage as a one time event should read this one differently, as a slow, structural change in what computing costs while AI demand runs this hot. The same underlying pressure is what makes renting compute so variable, a subject we get into when comparing the true cost of running your own AI, and memory is now a leading line in that bill.

How long does the shortage last?

Longer than anyone selling you a device will admit at the counter. The consistent message from the companies and analysts closest to the supply is multiple years, driven by the simple fact that new memory factories take years to build and every maker is prioritizing the AI lines first. J.P. Morgan's own analysis frames the resolution timetable as difficult to set. That uncertainty is itself the useful signal. When the end date is unknown and the direction is up, you plan as if the higher prices are the new baseline, and you are pleasantly surprised if relief comes early.

For a merchant, the takeaway is calm and concrete. AI made the memory that runs your business scarce, that scarcity has a price, and the price is now visible in laptops, cloud bills and the AI tools you rent. You cannot change the supply, but you control your timing, your contracts and your choice of tools, and those three levers are enough to keep the crunch from landing on you at full force. Automation and good tooling still pay for themselves, and if you want a sense of where those wins are largest before peak season, our guide to what to automate before the busy months is a good next stop.