Most people never think about the memory in their computer. It is a quiet, relatively cheap component that comes inside the box. This year it became one of the most sought-after products in the world. Anyone who bought a computer, graphics card or new phone in recent months paid more, and not because the product got better. They paid for a race happening far from them, in AI data centers.
What happened to memory?
Three companies dominate memory chip production: Samsung and SK Hynix from Korea, and Micron from the US. In the past two years they have shifted a large share of capacity to an advanced memory called HBM, which sits next to Nvidia's processors and its rivals' in AI servers. HBM is far more profitable for them, and demand for it looks almost unlimited. The result: fewer production lines for the ordinary memory that goes into computers and phones.
How much did it cost?
According to TrendForce data cited in the media, contract prices for consumer memory jumped by as much as 89% in a single quarter of 2026. According to reports, the average price of a computer in Europe rose by double digits over the summer. Smartphone makers are preparing further increases, and budget models are hit hardest, because memory is a large share of their cost. Some makers are starting to use recycled memory: according to TrendForce, 7% to 9% of phones in 2027 are expected to include memory removed from old devices.
The moment it became a public debate
Earlier this month OpenAI CEO Sam Altman said on a podcast that it is "quite terrible" that home computers have memory, because most of the time it is idle, and that it could be used better in data centers. Technically there is logic to it. But when the CEO of one of the companies leading demand tells consumers their memory is wasted, just as they pay more because of it, the backlash was predictable. Behind the remark is a real debate: will AI run mainly in the cloud, controlled by a few giants, or also on our devices.
Who profits
The shortage is excellent news for memory makers, which reported record profits last quarter. It is good news for equipment makers too: Applied Materials reported this week that DRAM equipment made up about a third of its systems revenue, up from 28% a year earlier. And in Israel, Rehovot-based Nova said its advanced packaging business posted record revenue, partly thanks to expanding HBM production lines. Every new memory fab needs inspection and metrology equipment, and some of it is developed here.
Who pays
On the other side is everyone else: computer and phone makers, who must choose between raising prices and eroding margins, and companies that build memory into their products. Mobileye, for example, said it is passing higher memory prices on to customers. And at the end of the chain is the consumer, who pays more for the same device, or gets a device with less memory at the same price.
When will it end?
Probably not soon. The CEO of Team Group, a Taiwanese memory maker, said earlier this month the shortage is expected to worsen from late 2026 and through 2027. Research firm Counterpoint said the pace of increases has eased but estimated supply will stay tight, because the big new fabs will only start producing in 2028. Building a memory fab takes years and tens of billions of dollars, and cannot be sped up just because prices rose.
What about an AI bubble?
There is one scenario where prices fall faster: if data center investment slows. If cloud companies cut orders, memory makers would move lines back to ordinary memory, and supply would grow. But this week's signals don't point that way: Cisco raised its AI infrastructure forecast, and Applied expects a strong quarter. Anyone waiting for the bubble to burst to buy a cheap computer may wait a long time.
What to do in practice
If you need a computer or phone in the coming months, there is no reason to wait: by every forecast, prices won't fall before 2028. Check how much memory you really need. For a home user, 16GB in a computer is enough in most cases. Those who edit video or run local AI tools need more. A computer that lets you add memory later is an advantage. And it is worth considering last year's model, built when memory was cheap.
The bottom line
The memory shortage is a tangible example of how the AI revolution reaches everyone's pocket, even people who have never used a chatbot. The massive investment in data centers pulls in every resource: power, chips, engineers, and memory too. As long as the race continues, those not in it will pay more for everything else.
