How artificial intelligence and memory demand are making PCs more expensive

BarcelonaThe proliferation of artificial intelligence centers, with their insatiable demand for semiconductor memory of all kinds, as well as the increase in sales of AI-enabled computers and laptops (called AIPCs)—primarily for business use, but also for consumers—is causing a severe shortage of memory, and consequently, a significant shortage in the market. Users looking to upgrade their computers in the coming quarters will find that the equipment is more expensive, largely due to the higher cost of memory.

After many quarters of weak sales of desktop and laptop computers, demand is now steadily rising. One reason is Microsoft's decision to stop updating Windows 10 security tools last week, pushing more users to upgrade to the new version of Windows 11 (it is estimated that there are nearly 500,000 unupdated computers).

The main reason for this increase in demand for computers, however, is the desire to use AI more intensively, and therefore, apart from a more powerful graphics card, more memory is needed: both DRAM (for temporary data) and NAND flash, to store processed data. More traditional hard drives are also being requested, with a corresponding increase in price.

Global demand for AIPCs, both in the laptop and desktop versions – which are better prepared to perform AI tasks – is growing at double digits from quarter to quarter during 2025, according to Enrique Lores, global CEO of HP, who predicts that it will continue to do so throughout 2026. AI sales are in the business environment, but are also being noticed in the consumer segment.

According to the consulting firm Gartner, 31% of all computers sold in 2025 will be AIPC, out of a total market of around 250 million units shipped worldwide. By next year, the percentage will be 54.7%, out of around 261 million total units, estimates Ranjit Atwal, head of analysis at the consulting firm. After two years of poor sales, computer manufacturers are climbing out of the hole.

A high-capacity graphics card is essential for artificial intelligence tasks, as is the processor, but above all, a lot of memory is needed—the more, the better—to output the data read by the graphics card and the processor. This has always been the case with computers, but now it has become essential if you want the equipment to function properly and there are no bottlenecks.

Fortunately, semiconductor memory capacity is constantly increasing, especially now that it's being manufactured at high temperatures (memory chips with more than 200 layers are already being made). However, there's more demand than supply, both for DRAM and NAND flash memory. Shortages are expected to occur at least throughout next year: by 2027, supply and demand are expected to balance, followed by overproduction.

The memory market has always been cyclical, typically lasting five years, with a period of oversupply followed by relative balance and a shortage of supply, before the cycle begins again. The reason is simple: huge investments are required to develop a new generation of memory and to build and deploy the necessary factories, a process that takes years and doesn't always match the increase in supply with the needs of demand at any given time, causing sharp price fluctuations.

This cyclical nature is clearly reflected in the figures from recent years: after growing 37% in 2021, the market fell 13% in 2022 and 34% in 2023, falling below $100 billion. In 2024, it rebounded strongly with 77% growth, and this year it is expected to increase by 17%, reaching $200 billion. According to the consultancy IDC, the peak will be reached in 2026 at $220 billion, before a slight correction in the following two years.

To top it all off, demand growth isn't usually linear, but rather spikes when a new device or application appears: this happened with the emergence of laptops, smartphones, and, more recently, with AI, which is causing major disruptions in the market. Furthermore, supply is highly concentrated, with four large companies: the Korean companies Samsung and SK Hynix, the American company Micron, and Kioxia (formed by the Japanese company Toshiba and the American company Western Digital).

Insatiable AI centers

AIPCs are impacting demand for memory, but the main market disruptor is AI data centers, which are turning everything upside down. These centers are not only causing an unusual and unstoppable increase in power consumption, but they also require a special type of extremely fast memory, called HBM (High-Band Memory), with large capacity and high bandwidth to support the billions of calculations per second processed by graphics cards.

In fact, Nvidia's AI data centers are not so powerful because of Nvidia chips; rather, a considerable part of their large capacity is due to Nvidia's own CUDA software and the HBM memory that SK Hynix practically exclusively manufactures for Nvidia. The development of HBM is relatively recent, and its development has required enormous investments. Consequently, its price is prohibitive for most suppliers, except for Nvidia, which can afford to pay what is necessary to have the best and fastest memory. One tandem which is working divinely, unfortunately for Samsung, which couldn't or didn't know how to take advantage of this opportunity and gave it to a compatriot company.

In any case, the cards for HBM memories are about to be dealt again with a new generation, called HBM4, which promises much more data storage capacity and much more bandwidth, as well as more moderate energy consumption. And the fact is that the current generation HBMs go very fast collecting data, but also consuming energy and heating the chips, which then have to be cooled. Both SK Hynix and Samsung plan to release their devices by the end of the year and have sufficient production capacity by 2026 for the AI wheel to continue turning at full speed.