From Tokyo: The global memory chip shortage will extend well past the halfway point of this decade. A global shortage of memory chips is likely to persist another four to five years because of endemic constraints in semiconductor production, according to SK Group Chairman Chey Tae-won, whose company controls SK Hynix, one of the world's three dominant memory manufacturers.
Industry-wide, supply of the basic wafers that get made into chips are lagging demand by more than 20 percent, Chey told reporters on the sidelines of Nvidia's developer conference in San Jose. The mathematics are unforgiving: securing additional wafers takes at least four to five years, which means the relief that consumers and manufacturers are waiting for remains years away.
What Australian observers often miss about this crisis is that it is not a temporary disruption but a fundamental reordering of how the world manufactures silicon. Unlike the pandemic-era shortages that stemmed from logistics chaos, this squeeze flows from deliberate choices by manufacturers. This shortage is driven by a structural reallocation of manufacturing capacity toward high-margin products for artificial intelligence infrastructure, creating scarcity in consumer and enterprise PC markets.
The mechanism is straightforward but consequential. Artificial intelligence workloads demand specialised memory chips called high-bandwidth memory, or HBM, which are vastly more profitable than the conventional DRAM that goes into your laptop or smartphone. Every wafer directed toward HBM is a wafer denied to consumer electronics. HBM consumes roughly three times more wafer capacity than commodity DRAM, a gap that could widen to nearly four times as the industry transitions to HBM4 this year and even larger in HBM4E next year.
The consequences ripple across the global economy. Memory now accounts for 15-18 percent of PC production cost, double 2024 levels. Smartphone makers face margin compression. Console manufacturers contemplate delaying new products. The era of cheap, abundant memory has closed, possibly for the remainder of the decade.
Some observers argue this outcome was predictable and that manufacturers should have anticipated it. They have a point. Memory chip makers faced devastating losses when the pandemic boom collapsed around 2023; they cut capacity aggressively to stabilize pricing. They were understandably reluctant to expand quickly when AI demand accelerated. Caution in capital-intensive industries is not irrational.
Yet there is a legitimate counterargument: this situation reflects genuine physical constraints, not merely reluctant suppliers. Building a semiconductor fabrication plant takes years and costs tens of billions of dollars. Semiconductor fabrication expansion takes years, not months. Even manufacturers committing to expansion see new capacity arrive too late to relieve current pressure.
The geopolitical dimension adds another layer of complexity. With HBM manufacturing concentrated in South Korea, US export controls tightening, and China accelerating domestic memory capacity through firms like CXMT, memory had "crossed the threshold from a commercial component to a geopolitical asset" – introducing a category of supply risk that traditional vendor diversification alone cannot mitigate. This is no longer purely a commercial matter; it is becoming a strategic asset.
For the Pacific region, the consequences are mixed. South Korean manufacturers like SK Hynix benefit from pricing power and consolidated demand from hyperscalers. But consumers across the region face higher device costs. Smaller manufacturers in Southeast Asia and Pacific markets will feel the pressure disproportionately, squeezed between rising input costs and limited pricing power in competitive segments.
The structural nature of this shortage is worth emphasising. This is not a temporary supply spike followed by normalisation. This is not just a cyclical shortage driven by a mismatch in supply and demand, but a potentially permanent, strategic reallocation of the world's silicon wafer capacity. Demand for AI infrastructure is likely to remain strong throughout the decade, even if growth moderates from current explosive rates.
The path forward demands intellectual honesty. Manufacturers cannot be faulted for declining to repeat the costly errors of the 2023 downturn. Governments pursuing semiconductor self-sufficiency have good reasons to do so, though realising that ambition takes considerable time and capital. Consumers and businesses will pay higher prices until supply rebalances, which may not occur before 2028 at the earliest.
This is the actual cost of the artificial intelligence infrastructure boom: not just the energy consumed by data centres, but a structural compression of memory supply that will reshape device affordability and capability for years. It is neither a crisis without cause nor an easily resolved constraint. It is a legitimate trade-off that reasonable people can disagree about, but the outcomes are unavoidable.