A forecast about allocation, not global production

A warning from Apacer chief executive C.K. Chang has sharpened concerns about the memory market in 2027. Chang’s central estimate is that the volume of DRAM chips available from major manufacturers to independent memory-module suppliers could fall to roughly 30% of the 2026 level. Put another way, the open allocation available to companies such as Apacer could be more than 70% lower year on year.

That is a striking prediction, but its scope matters. It is not a forecast that worldwide DRAM production will decline by 70%, nor that consumers will find 70% less RAM on shop shelves. It concerns the share of supply that leading chipmakers may release to module assemblers and other customers without the purchasing scale or long-term agreements of major cloud and server operators.

The distinction is important because the memory market is increasingly segmented. A chip producer may devote output to high-bandwidth memory for AI accelerators, conventional server DRAM, mobile memory, or consumer and industrial channels. Production can remain high while the amount reaching the merchant market becomes much scarcer.

Why the open market is more exposed

Independent module makers buy DRAM chips, test and package them into products such as memory modules, industrial memory and storage devices. Their role makes them especially sensitive to allocation decisions by the manufacturers that control wafer capacity.

AI data centres have changed the economics of those decisions. High-bandwidth memory commands high values and requires sophisticated stacking and packaging. At the same time, AI servers use large quantities of conventional DRAM for processors, accelerators and associated systems. Large cloud customers can secure supply through direct relationships and multi-year purchase commitments, reducing the flexibility available to smaller buyers.

Apacer has reportedly responded by increasing memory inventory and arranging additional financing to buy chips when supply becomes available. That is a rational defensive move if scarcity persists, but it also carries risk. Inventory bought at exceptionally high prices can quickly lose value if demand weakens, capacity arrives faster than anticipated, or customers postpone purchases.

For a module supplier, the immediate threat is therefore not simply a higher chip price. It is the possibility that there are insufficient chips to support contracted product shipments. In that environment, securing supply can take priority over achieving the lowest possible purchase cost.

DRAM is likely to remain the pressure point

The outlook does not appear equally severe across every category of memory. Industry forecasts still identify DRAM as the tighter market, particularly where advanced server memory and AI-related products compete for capacity. Micron’s recent product and investment updates also underline the intensity of demand for high-performance memory across cloud, data-centre, mobile and embedded applications.

NAND flash, which is used in solid-state storage, faces similar capacity constraints in the near term but may follow a different trajectory. TrendForce expects NAND to remain in deficit during 2026, yet projects that supply growth could exceed demand in the second half of 2027. That scenario would not make storage inexpensive overnight, but it would suggest a gradual easing rather than an ever-deepening shortage.

DRAM supply may take longer to rebalance because building capacity is slow, expensive and technologically complex. New fabrication plants require years of planning, construction, equipment installation and qualification. Process upgrades at existing plants can add output sooner, but manufacturers must balance that work against the need to produce increasingly advanced memory for their most valuable customers.

What it could mean for PCs and devices

The first effects of constrained allocation are likely to be seen in product availability, memory configurations and pricing rather than an absolute absence of consumer RAM. PC and device manufacturers can respond by favouring higher-margin models, trimming lower-capacity configurations, redesigning product plans or passing component costs to buyers.

Analysts already expect elevated memory costs to weigh on the PC and smartphone markets in 2026. The impact is likely to be most pronounced in entry-level products, where memory represents a larger share of the bill of materials and vendors have less room to absorb increases. If module makers face sharper allocation limits in 2027, that pressure could extend beyond premium server hardware to mainstream notebooks, desktops, phones and industrial systems.

Older memory generations may be particularly awkward. Long-lived industrial, automotive and embedded products often depend on mature DRAM types, but chipmakers have incentives to transition capacity towards newer and more profitable technologies. A shortage in legacy parts can therefore coexist with adequate supply in another segment.

A scenario, not a settled outcome

Chang’s estimate should be treated as a company leader’s scenario for the open market, rather than a settled industry forecast. Apacer has a direct interest in securing inventory and highlighting supply risk, while the eventual outcome depends on demand, factory yields, capital expenditure, Chinese suppliers’ progress, and the appetite of cloud companies for AI infrastructure.

There are also countervailing forces. Consumer demand for PCs and smartphones is price-sensitive, and a prolonged rise in memory costs can reduce unit sales. Weaker consumer shipments would lower demand for both DRAM and NAND, potentially freeing supply that would otherwise remain constrained. New capacity and process improvements should also begin to influence the market over time, although not necessarily quickly enough to relieve the 2027 allocation squeeze described by Apacer.

The most credible conclusion is that memory supply is becoming less like a single commodity pool. Companies with direct supply agreements, stronger purchasing power and flexibility across product designs will be better positioned than those relying on spot availability. Apacer’s warning is significant not because it predicts an across-the-board disappearance of memory, but because it illustrates how AI-led demand may change who can obtain it, on what terms, and at what price.

Sources