Memory lapses: when will the shortage of chips end
Manufacturers of memory chips — Samsung, SK Hynix and Micron — record record profits and soaring capitalization. The rest of the technology sector is forced to pay a kind of "digital tax", as are ordinary users of computer equipment and gadgets. The rapid rise in the cost of memory components threatens the profits of everyone — from manufacturers of budget smartphones to giants like Apple and Nintendo. What is the threat of a combination of shortages in the market and the dictates of monopolists, how long will prices for memory and other components rise — in the Izvestia material.
Memory as a luxury item
According to analytical agencies, the average selling price (ASP) of DRAM (RAM) and NAND (flash) chips in some segments has jumped by 50-80% over the past three quarters. The effect of this leap is clearly visible in corporate reporting. Shares of Dell and Lenovo have fallen by more than 25% from their peaks, while Qualcomm and Honda openly admit that memory shortages limit production volumes and eat away a significant portion of profits. For the end user, this means not only an increase in prices for ready—made devices, but also a slowdown in progress - manufacturers are forced to save on performance in order to meet the previous price tags.
The main culprit of the shortage was the boom in artificial intelligence. However, the problem goes deeper than just rising demand. Learning large language models requires specific high bandwidth Memory (HBM).
Technologically, HBM production is much more complex than standard RAM. It takes three times as many silicon wafers to manufacture one HBM chip as conventional DRAM, while the percentage of usable products is significantly lower. As a result, the largest players have reoriented their best production lines to produce expensive HBM for the needs of Nvidia and other AI giants. This created an artificial vacuum in the segment of standard memory for PCs, smartphones and servers. In fact, the "traditional" memory market has been sacrificed to the AI fever.
Collusion or strategy?
The memory market has historically been a "tripoli": Samsung, SK Hynix and Micron control more than 90% of global DRAM production. This club has a long history of antitrust litigation and accusations of price collusion.
Today, the situation looks like "disciplined supply management." After a difficult year in 2023, when the industry suffered from overstocking and low prices, manufacturers are extremely cautious about increasing capital expenditures. Instead of building new factories, they modernize old ones, artificially maintaining shortages. This is an ideal moment for them: the margin on HBM is off the charts, and the shortage of conventional memory allows yesterday's partners, auto giants and electronics manufacturers, to dictate prices. This is not so much a collusion in the legal sense, as a pragmatic exploitation of the current situation: why produce more if you can sell less, but more expensive?
The Chinese "equalizer"
In this situation, China is the only hope for price stabilization. Beijing, seeking technological sovereignty against the backdrop of US sanctions, is pouring billions of dollars into its champions, CXMT (ChangXin Memory Technologies) and YMTC (Yangtze Memory Technologies).
Will China be able to change the rules of the game? According to experts, CXMT already controls about 10-12% of the world's DRAM production capacity. By the end of 2026, the company plans to increase volumes to 15-18%. This is enough to start putting pressure on prices in the mid-range and low-end segments (DDR4 and early DDR5).
At the same time, it must be admitted that the Chinese are successfully mastering mass production, but they are still lagging behind in top technologies — they are still too tough with the latest technological processes. Equipment supply sanctions (EUV lithography) do not allow them to compete directly with Samsung in the AI segment.
The main threat to the "big three" from China is the saturation of the mass market. As soon as the CXMT reaches full capacity, prices for standard PC and server memory may go down, forcing Koreans and Americans to either return to this segment or raise prices for AI memory even more. In the next year or two, China is unlikely to be able to completely stop the price increase, but it is able to create a "ceiling" for speculation. It is expected that by 2027, the combined share of Chinese manufacturers in the segments of standard memory may reach 20-25%.
Russia is even more expensive
For the Russian market, the situation with the memory supercycle is aggravated by specific local factors. While global component prices have increased by an average of 50-80%, the final price tags on domestic shelves and in the B2B segment have increased even more over the past year. Added to the global prerequisites here is the sharply increased cost of "transactional friction" — the complication of logistics and cross-border payments through friendly jurisdictions.
In the consumer sector, from DDR5 RAM modules to high—speed SSD drives, retail prices have increased by up to 100% or more over the past six months, depending on the brand and volume. Popular memory kits, which a year ago were perceived as an affordable upgrade, have now moved into the category of investment purchases. The situation in the server segment looks even more dramatic: due to the high demand for localization of IT infrastructure and the shortage of server components in the global market, the cost of specialized memory modules in Russia has doubled in some cases.
In fact, the Russian customer today pays a double "risk premium": first, it competes with global giants like Microsoft and Amazon for quotas for chip production, and then pays for the services of a complex chain of intermediaries and insurance for circumventing sanctions barriers. Against the background of the fact that stocks purchased at the old prices from large distributors are coming to an end, it is not necessary to expect price stabilization in the Russian Federation in the near future.
You'll have to be patient
The memory supercycle is unlikely to end in the next 2-3 quarters. Investments in new plants take time (at least 18-24 months before reaching full capacity), and demand from Amazon, Google, Microsoft and others for AI infrastructure continues to grow.
Relief may come only by the middle of 2026, when two factors converge: the introduction of new capacities in Korea and the United States and the large-scale entry of Chinese manufacturers into the international market. Until then, the technology sector will have to get used to the idea that memory has become a strategic and extremely expensive resource rather than a consumable. The electronics market will also either be able to shrink in margins, or shift costs onto the shoulders of the consumer — which is already happening.
Переведено сервисом «Яндекс Переводчик»
