The rising cost of memory has become one of the most noticeable side effects of the artificial intelligence boom. What started as a race to secure HBM for AI accelerators has ended up affecting conventional DRAM, desktop DDR5 modules, servers, SSDs, and gradually the final prices of laptops, workstations, and consumer electronics. But the question is no longer just how much memory prices can go up, but who will have the capacity to break the cycle.
Kye-hyun Kyung, former head of Samsung Electronics’ Device Solutions division and current senior advisor to the company, has focused on China. According to his remarks gathered after the forum of the National Academy of Engineering of Korea, memory prices could start to fall in the second half of next year if China’s significant investments in production capacity translate into more wafer availability. This is no small forecast: Kyung led Samsung’s chip business until 2024 and knows the industry from within, accustomed to cycles of shortage, overinvestment, and sharp corrections.
The thesis is simple, though not certain. Today, prices are high because AI demand has absorbed capacity, major cloud providers are signing long-term contracts, and manufacturers are prioritizing higher-margin products. But if China manages to add a large volume of DDR5 DRAM in 2027, that shortage could quickly turn into oversupply. In memory, history often repeats itself: when everyone invests simultaneously, the market can shift from euphoria to correction in just a few quarters.
DDR5 Price Has Gone Off Track
The best indicator of current tension comes from the German market. According to 3DCenter tracking cited by Wccftech, retail prices for DDR5 have surged to 414% above July 2025 levels. This reflects store and specific market data, not a global industry measure, but it clearly illustrates the impact on consumers and PC builders.
The explanation lies in capacity reallocation. Memory manufacturers cannot ramp up production overnight. When AI demands HBM, server DDR5, high-capacity RDIMM modules, and enterprise NAND, part of the available capacity shifts away from the consumer market. As a result, products seemingly unrelated to AI, like a DDR5 kit for a home PC, end up facing pressure originating from data centers.
TrendForce already warned that manufacturers are prioritizing server applications, with contractual prices for conventional DRAM rising in 2026 and an increasing gap between cloud buyers, enterprise servers, and consumers. The consultancy also forecasted significant increases for DRAM and NAND early in 2026, driven by demand from North American cloud providers and capacity competition.
| Factor | Market Effect |
|---|---|
| AI demand | Absorbs HBM, server DDR5, and enterprise NAND |
| Long-term cloud contracts | Leave less capacity available for other buyers |
| Production reallocation | Reduces availability of conventional DRAM |
| Preemptive purchasing | PC manufacturers and distributors place early orders |
| Chinese entry into DDR5 | Could increase supply and pressure prices in 2027 |
| Overcapacity risk | If demand cools, prices could correct quickly |
This is already affecting the downstream supply chain. Laptop makers, PC integrators, and electronics brands must decide whether to absorb costs, raise prices, or cut base configurations. A system that previously shipped with 16 GB of RAM might revert to offering 8 GB to contain costs. Higher-capacity SSDs may become premium options. Memory, once taken for granted, returns to being a central variable.
China Enters When the Market Is at Its Highest
China’s role is what could shift the balance. ChangXin Memory Technologies, better known as CXMT, has become the leading Chinese player in DRAM. The company has announced DDR5 chips and modules reaching speeds up to 8,000 MT/s and high-performance LPDDR5X, signaling that the technical gap with Samsung, SK Hynix, and Micron is narrowing in conventional memory.
Furthermore, CXMT aims to fund its expansion through an IPO in Shanghai. Reuters reported that the company plans to raise approximately 29.5 billion yuan (around $4.22 billion) to upgrade facilities, fund R&D, and expand its portfolio, including HBM for AI applications. Although its overall market share remains small compared to the Big Three, its growth is significant because DRAM is highly sensitive to any increase in supply.
Kyung himself cited estimates from analysis firms suggesting memory production capacity could reach about six million wafers per month by late 2027. If this capacity hits the market with acceptable yields, the impact on prices could be profound. But he also issued a key warning: investments could slow down if major tech companies see diminishing returns on their AI capital expenditures.
This nuance is important. Price declines are not guaranteed. For China to truly influence the global market, it isn’t enough to just install equipment. Quality production, yield improvements, module validation, platform compatibility with Intel and AMD, acquiring clients beyond the domestic market, and navigating trade restrictions are all essential. The memory industry rewards not just announced capacity but useful, stable volume.
Samsung, SK Hynix, and Micron Are Not Out of the Game
A simple narrative would frame China as the actor about to crush prices and end Korean and U.S. dominance. The reality is more complex. Samsung, SK Hynix, and Micron still hold significant advantages in scale, technology, process control, OEM relationships, HBM, packaging, and enterprise validation. Particularly in HBM, China still lags behind.
What could happen is increasing pressure on conventional DDR5 and segments where price sensitivity outweighs top certification. If CXMT and other Chinese manufacturers supply enough memory to PCs, servers, local integrators, and emerging markets, the large players will have less margin to sustain extreme prices on certain products.
This scenario isn’t new. The memory industry has gone through phases of new entrants, capacity excess, and price crashes. Japan once dominated part of the market, South Korea took over, Micron established its US presence, and now China is attempting to build a nationalist alternative. Each wave involved industrial policy, massive investment, and painful cycles.
The current difference is AI. Demand for data center memory could sustain higher prices longer than in past cycles, especially if hyperscalers continue investing hundreds of billions in infrastructure. But this demand could also become more selective. If AI returns are delayed, models become more efficient, or some investments are postponed, the new capacity could come online just as the market begins to stabilize.
Consumers Might See Relief, But Not Immediately
For PC users, content creators, gamers, and small businesses, Kyung’s forecast offers a reasonable hope: memory prices could stabilize and begin correcting by 2027. However, immediate normalization isn’t guaranteed. Supply contracts, inventory buildup of expensive stock, and server priorities could keep pressure high for several quarters.
Segment differences will also persist. HBM remains a strategic, scarce product. High-capacity server DDR5 could keep prices elevated if data center purchases continue. Consumer DDR5 might see faster correction if Chinese supply increases and traditional manufacturers reinstate production of standard modules. NAND and SSDs will have their cycles too, closely linked to AI and enterprise storage trends.
For equipment manufacturers, devising strategy will be more complicated. Overpaying now could result in inventory that’s difficult to sell if prices fall by 2027. Underordering risks missing out on market demands for laptops, servers, or PCs still under tension. Inventory management will again be a competitive advantage.
For South Korea, Kyung’s warning carries an industrial message: if China grows its mainstream DRAM capacity and the US maintains its fabless design advantage, Korea must strengthen its position in system chips, packaging, HBM, foundry, and design. Excelling only in memory isn’t sufficient if the market commoditizes or Chinese competitors learn to produce at scale.
Memory is now in a tricky phase: signals point in conflicting directions. In the short term, scarcity, AI, and high prices dominate. Mid-term, massive investment, China’s acceleration, and correction risks loom. End users only see higher RAM prices; industry insiders see a race to control the next cycle when AI euphoria meets new wafer-filled factories.
Frequently Asked Questions
Why has DDR5 prices surged so much?
AI infrastructure demand has absorbed significant memory capacity, especially HBM and server DDR5. As production shifts toward higher-margin products, less supply remains for conventional and consumer memory.
Can China drive memory prices down?
Yes, if investments lead to useful production with good yields and volume. Kyung believes this effect might start in the second half of 2027.
What role does CXMT play?
CXMT is China’s leading DRAM manufacturer, advancing in DDR5, seeking funding to expand capacity, and being closely watched by Samsung, SK Hynix, and Micron.
Will consumer prices drop soon?
Not necessarily. Some relief could occur in 2027 if supply increases, but throughout 2026, high demand from AI, cloud contracts, and capacity shortages will keep prices high across many segments.
via: wccftech