If you have been planning to build a computer recently, or browsed the spec sheets of various pre-installed consoles, you may have noticed something frowning: the price of memory is soaring at an abnormal rate. On some models, the increase is even as high as five times, which directly drives up the cost of the entire machine and greatly reduces the price advantage of self-assembled computers. This is not a short-term market fluctuation, but the result of multiple factors fermenting at the same time, and the core reasons are almost all related to the explosive growth in memory demand brought about by artificial intelligence.
Why will memory be so expensive in 2026? How should computer buyers respond?
Artificial intelligence becomes the main reason for rising memory prices
The rapid popularity of large-scale language models and generative AI has made memory one of the most critical and scarce resources in AI infrastructure. Whether training models or providing real-time inference services, a large amount of high-speed memory support is required. The training phase requires high-speed access to large amounts of data; the inference phase requires saving user conversations, context, and model operation status in memory. Even for short interactions, each user request consumes memory. When such demand expands to millions or even hundreds of millions of users, the overall demand will naturally show explosive growth.
As the size of the model increases, the memory requirements also double.
Demand is increasing not only because there are more users, but also because the models themselves are getting larger. In recent years, the context window of the model has expanded rapidly, jumping from 200,000 tokens in the past to millions today. The larger the context, the higher the amount of data that the model needs to load at one time, and the memory usage during inference also soars. In other words, the stronger the AI model, the tighter the memory.
Consumer memory and AI server memory share the same supply chain
Although general PCs use DDR4 or DDR5, and AI servers use high-bandwidth memory (HBM), the two are technically different, but they share the same production ecosystem: wafers, packaging, factory capacity, and supply chains. AI companies are willing to pay much higher prices for HBM than the consumer market, so memory manufacturers naturally prioritize supplying to enterprise customers. The result is: consumer-grade DRAM production capacity is squeezed, supply drops, and prices naturally rise.
The supply side is highly concentrated and the expansion of production capacity is limited.
Currently, there are only three major memory suppliers in the world: Samsung, SK Hynix and Micron. The highly concentrated market represents a limited expansion of production capacity. Micron has even closed its Crucial consumer brand and invested more resources in the AI memory market; SK Hynix has reported that all HBM production capacity in 2026 has been sold out. When the supply side cannot expand rapidly, the market will naturally be in short supply.
Why won’t memory prices fall back in the short term?
Even though major manufacturers have announced expansions of production capacity, it often takes several years from construction to mass production of new factories. The market generally expects that the time to truly alleviate supply pressure may fall between 2027 and 2028. Prior to this, memory prices will not fall rapidly, but may even double. For consumers who hope to wait until they are cheaper to buy, the short-term prospects are not optimistic.
Early signs of demand easing
While the overall trend still favors strong demand, there are some technological advancements that may reduce memory pressure in the future. For example, the TurboQuant AI compression technology launched by Google claims to reduce model memory requirements by up to six times. If this type of technology can be widely adopted, the memory requirements of data centers may gradually decrease, thereby affecting market supply and demand. However, these technologies are still in their early stages and their long-term impact on the market remains to be seen.
What should PC buyers do now?
It depends on your budget and needs:
- If you are pursuing CP value
Choosing the DDR4 platform is still the most pragmatic option. The price of DDR4 is relatively stable, the second-hand market supply is sufficient, and the overall cost is much lower than DDR5. - If you want to assemble a high-end console with the latest specifications
Waiting in the short term may not necessarily translate into better prices, as the market is still in an upward trend. If you need to use it immediately, it may be more cost-effective to buy it now than to “wait longer”.
Will memory prices return to past lows? Even if supply improves in the future, memory prices are unlikely to return to pre-pandemic lows. As the market becomes accustomed to higher prices, enterprise customers are willing to pay a premium, and AI demand continues to grow, the “new normal” cost of memory may be much higher than in the past. As new production capacity is gradually opened, prices do have a chance to fall from the current peak, but it is probably impossible to return to the past era when “memory is as cheap as cabbage.”
Source: KOCPC Chinese