The paperclip problem always seemed too absurd to me. Also known as a paperclip maximizer, this is philosopher Nick Bostrom’s thought experiment that imagines how a superintelligent AI with the goal of maximizing paperclip production could end up destroying the world by directing all available resources into paperclip manufacturing.
While it would be irresponsible to say this is happening, we are starting to run out of some resources. And it’s about to affect your life.
You may have heard about the global memory shortage caused, in part, by the rapid construction of AI data centers. Just as they need semiconductors for data processing and water for cooling, these facilities need memory, or RAM, for short- and long-term data storage. Virtually all consumer electronics, from desktop computers to smartphones, also require memory to function. The problem is that only three companies (Micron, SK Hynix and Samsung Electronics) manufacture almost all the memory on the market. They can’t do it fast enough right now and it’s unclear when they’ll be able to meet demand.
Normally, the scarcity of a computer component wouldn’t lead me to reference an apocalypse thought experiment, but here we are. Memory is a really important component, and as the rise of AI data centers consumes more and more resources, not having enough means that virtually every device with a chip will become more expensive or less innovative, or both. You can think of it in the same way as the dreaded combination of inflation and stagnation made famous in the 1970s and resurrected by the second Trump administration: stagflation. Things cost more and are basically worse.
Prices are already rising and manufacturers are already pointing to memory shortages to explain it. What you can expect in the coming months, and possibly years, is a slowdown in the kind of spec improvements you’re used to seeing in new models. (This year’s iPhone Pro 17, for example, has 12GB of RAM versus the iPhone 16 Pro’s 8GB.) You may even see manufacturers choose cheaper options for components like screens or batteries, in ways that may not be immediately obvious.
“They’re looking for somewhere to cut corners right during this period to offset memory costs,” said Ryan Reith, group vice president at market intelligence firm IDC. He added that some companies simply won’t build the higher-power devices they had planned to build in the near future. Meanwhile, IDC predicts that smartphone sales will decline in 2026 due to memory shortages.
There is also hoarding. There is a veritable alphabet soup of different types of storage, but one that is essential for AI is known as DRAM. You can find DRAM in devices large and small (laptops, game consoles, TVs, cars) and as the big three memory manufacturers directly supply more of that memory to AI data centers, there is less available to device makers. Therefore, some companies are hoarding memory, which has the complicated effect of driving up prices and reducing supply.
The other acronym to know here is HBM, which stands for High Bandwidth Memory. This is a type of DRAM designed specifically to run high-performance processors, like Nvidia’s Blackwell chips, that populate AI data centers. The profit margins on this type of memory are roughly double those of the type of DRAM used in consumer devices, so naturally memory manufacturers are dedicating additional resources to manufacturing it, contributing to the consumer memory backlog.
This situation is going to take a while to resolve. To build more memory chips, memory makers need to build more factories, known as fabs, and that process takes years. Micron, for example, will soon begin construction on a factory in upstate New York that won’t begin producing memory until 2030. The company’s chief commercial officer, Sumit Sadana, told CNBC last week: “We’re sold out by 2026.”
None of this means that if you go to the smartphone store in six months, you won’t be able to buy one, or that it will cost you twice as much. On the contrary, device manufacturers want to avoid the impact of stickers. What you’ll likely see, however, is that the price of the base model remains the same, but the internals aren’t as good as they would have been. If you want the version with more memory, you’ll pay an even bigger premium for those specs than you would last year.
“The price rise is not over yet,” says Reith.
We still don’t know how this ends. On the one hand, the data center boom that is devouring all memory is closely linked to the AI industry, which may or may not be a bubble ready to burst. On the other hand, the upward trend in prices extends to all industries. While the inflation rate has remained stable, things cost more than they did a year ago and are not getting cheaper. If smartphone or laptop makers realize they can sell a worse product at the same price as the better one, they may want to do so, regardless of any shortages.
When we talk about an affordability crisis, we are not exactly talking about a price increase crisis. Affordable means reasonable. It doesn’t seem reasonable that the average consumer should be saddled with crappier products while the AI industry is creating billionaires at a record pace. The world isn’t ending anytime soon, but you’re probably starting to feel the effects of the change in one way or another.
A version of this story was also published in the User Friendly newsletter. Register here so you don’t miss the next one!