Samsung Electronics surprised the market during its second-quarter 2026 earnings call by saying that the global memory shortage will not only continue but is expected to deepen in 2027 and persist through 2028. The statement, delivered by Kim Jaejune, a sales and marketing executive in the company's memory division, cast a spotlight on one of the most important — and least understood — dynamics of the current race for artificial intelligence.
The news arrives at a moment when Samsung just posted a record operating profit of $58.5 billion for the quarter, fueled precisely by explosive demand for AI memory chips. The paradox is only apparent: how can a company break profit records while simultaneously warning that supply cannot keep up with demand? The answer lies in the nature of memory semiconductors, a market that swings violently between boom and bust.
High-bandwidth memory, known as HBM, has become the most coveted input in the technology industry. It is essential for the AI accelerators produced by Nvidia and AMD, which consume ever-larger stacks of these components. The problem is that building new memory fabs takes years — Samsung's executive cited lead times of more than three years for new production capacity. That means even if companies decided today to expand output, the supply-side response would only arrive well after the window of peak demand.
The most visible consequence of this imbalance is price. DRAM and NAND contracts have been rising consistently, and Samsung has signaled it is negotiating multi-year agreements with AI clients to secure supply capacity. It is the same reasoning that leads hyperscalers such as Microsoft, Google and Amazon to reserve supplies years in advance, in a shift that turns memory into a kind of strategic commodity, fought over with the same intensity as energy or technical talent.
There is a detail rarely discussed in this story: the shortage is not confined to cutting-edge HBM. Conventional DRAM, used in PCs, smartphones and traditional servers, is also tight. That is why Samsung sees a curious side effect — while its chip division breaks records, its mobile division faces rising costs and even an operating loss, since DRAM and NAND are inputs to the company's own devices. In short, abundance on the AI side becomes scarcity on the consumer side.
For anyone following the sector, Samsung's warning is a sign that the current cycle is not just another passing peak. AI demand appears structural, not cyclical, which could keep prices elevated for longer than companies would like. The open question is whether manufacturers can ramp up production fast enough, or whether we are heading for a decade in which the bottleneck of computing stops being the processor and becomes the memory that feeds it. How much longer will the end consumer end up paying for this race? The memory crunch also reshapes the competitive landscape. Samsung and SK Hynix are locking in the largest hyperscalers with multi-year agreements, which squeezes smaller buyers who lack the leverage to negotiate. That asymmetry could widen the gap between cloud giants and everyone else, making memory a tool of market concentration rather than just a supply constraint. For the broader industry, the real test will come when AI spending inevitably cools: the same factories built to chase this boom could then flood the market, repeating the downturn cycles that have defined this sector for decades. Watching how companies thread that needle over the next two years will say more about the future of computing than any single earnings call.
Sources: KED Global, Digitimes, Tom's Hardware
✓ Independent sources cross-checked and verified before publishing
