Memory market rebounds, but AI demand reshapes recovery

The memory segment offers a particularly clear example of this shift. According to TrendForce, 2026 is shaping up as a period of significant price increases across both DRAM and NAND Flash. In the second quarter alone, conventional DRAM contract prices are expected to rise by 58–63% quarter-on-quarter, while NAND Flash prices may increase even more sharply, by 70–75%.

At first glance, this resembles a familiar pattern. After a prolonged downturn, inventories begin to normalise, demand returns, and pricing recovers accordingly. Yet this time, the underlying drivers suggest that the recovery is not unfolding in a uniform or fully synchronised way.

A large part of the current momentum is being generated by AI-related demand and the continued expansion of data center infrastructure. Hyperscalers are not only increasing their consumption of memory, but are also securing supply through longer-term agreements, which is gradually changing the balance of power in the market. Capacity is no longer simply responding to demand; it is being actively directed towards segments that offer higher margins and greater long-term visibility.

This shift is particularly pronounced in DRAM, where suppliers are increasingly prioritising high-value products such as HBM and server DRAM. As a result, the availability of memory for more traditional applications is becoming more constrained — not necessarily because of an overall shortage, but because supply is being allocated differently than in previous cycles. This distinction is important, as it changes how pricing dynamics should be interpreted.

The effects are already visible downstream. TrendForce notes that demand in the PC segment has been revised downward, yet prices continue to rise. Suppliers are reducing shipments to PC OEMs and module makers, which forces buyers to secure supply through alternative channels and often at higher cost. In this context, price increases are not solely the result of strengthening demand, but also of strategic prioritisation within the supply chain.

A comparable dynamic can be observed in NAND, although the internal balance differs. Here, capacity is increasingly being directed toward enterprise SSDs, again driven by data center demand, while consumer applications remain more sensitive to rising costs. As memory prices feed into the final price of devices, manufacturers are forced to adjust — whether by delaying upgrades, modifying specifications, or scaling back shipment expectations. The recovery is therefore taking place under constraints that are not evenly distributed across the market.

What emerges from this is not a single, coherent recovery, but a market that is moving along at least two overlapping trajectories. On one side, infrastructure-driven demand — closely tied to AI and long-term investment cycles — continues to absorb capacity and support pricing. On the other, more traditional segments, particularly those linked to consumer electronics, remain exposed to cost sensitivity and demand volatility, which limits the pace and extent of their recovery.

This is not just a difference in speed. It reflects a deeper shift in how different parts of the market function.

A few months ago, on Evertiq, Claus Aasholm argued that the semiconductor industry may no longer follow a single, unified cycle. His point was not that cycles have disappeared altogether, but that they are becoming increasingly fragmented, with different segments entering different phases at the same time. When viewed through the lens of the memory market, this observation gains additional weight.

What was once described as a single memory cycle now appears more layered and less synchronised. Infrastructure-driven demand follows a longer and more predictable trajectory, supported by sustained investment and strategic supply agreements. In contrast, consumer-driven segments continue to operate within shorter, more volatile cycles, where pricing pressure can quickly translate into reduced demand.

Both layers remain interconnected, as they draw from the same supply base, but they no longer move in unison. This has important implications for how market data should be interpreted. Strong price increases, as highlighted by TrendForce, do not necessarily signal a broad recovery across all segments. Rather, they point to a concentration of demand in areas that are able to absorb higher costs and secure supply, while other parts of the market continue to face structural constraints.

A similar pattern has already been observed in the foundry segment, where growth is increasingly concentrated in advanced nodes and AI-related applications. The memory market now reinforces this picture, suggesting that the broader semiconductor industry may be evolving away from a single, unified cycle toward a more fragmented structure.

This brings us back to a question raised on Evertiq in December, when Claus Aasholm argued in his guest article that the traditional semiconductor cycle may no longer exist as a single, unified structure. If growth is increasingly concentrated and capacity selectively allocated, then aggregate indicators begin to lose some of their explanatory power, and a more nuanced reading of the market becomes necessary.

This perspective will be explored further in the coming weeks, as Claus Aasholm continues this discussion during his Evertiq Expo appearances — first in Zurich on April 23, followed by Kraków on May 7, and Lund on May 21 — examining how these structural shifts are redefining both growth and risk in the semiconductor market.