Category: Technology | Published: 2026-08-06
If you have noticed that server memory, storage, or business laptops are harder to source or more expensive than they were a couple of years ago, Samsung has some news that will not come as a comfort. The situation is not going to ease any time soon, and the world's largest memory manufacturer says it is likely to get worse before it gets better.
Samsung's memory division head delivered that assessment during the company's second-quarter earnings call, and the timeline attached to it was specific: supply constraints are expected to remain severe through 2028, with 2027 shaping up to be tighter than 2026.
What Samsung Actually Said
Jaejune Kim, Executive Vice President of Samsung's Memory Business, was direct with analysts. Speaking about the global memory supply outlook, he said the company believes it will be unlikely to see any significant increase in incremental supply through 2028.
He went further: unmet demand from this year is expected to carry over into 2027, compounding the shortfall rather than clearing it. The supply constraints are anticipated to become more severe in 2027 than they are now.
That is a striking statement from the company that sits at the centre of the global memory market. Samsung is not a minor player speculating about conditions it cannot control. It is one of the largest producers of the chips in question, and its earnings calls are among the most closely watched in the semiconductor industry. When Samsung says supply will be constrained for two more years, that assessment is based on its own production capacity, its customer order books, and its view of where demand is heading.
The results underpinning that outlook were, for Samsung at least, exceptionally strong. Soaring demand for memory chips used in AI infrastructure drove another set of robust quarterly numbers. The shortage is not a sign of a struggling industry. It is a sign of one being pulled in a very specific direction at high speed.
Why AI Is the Root Cause
The memory crunch traces back almost entirely to AI infrastructure investment, and understanding why helps explain why the problem is so persistent.
Training frontier AI models requires enormous amounts of high-bandwidth memory. But the more immediate pressure on supply right now comes from AI inference — the process of running trained models at scale to serve real users. Every AI assistant, every chatbot, every AI-powered search result requires memory to operate, and the more capable and complex those systems become, the more memory they consume.
Samsung describes what is happening as an unprecedented rise in demand for AI servers and computing infrastructure. The word unprecedented is doing real work there. Memory demand has had boom cycles before, but those were typically driven by consumer electronics — smartphone launches, PC refresh cycles, gaming hardware. What AI is driving is different in character: it is sustained, it is accelerating, and it is coming from cloud providers with very deep pockets who are willing to commit to multi-year purchasing agreements to secure supply.
Samsung confirmed that customers wanting to secure substantial AI service infrastructure are increasingly approaching the company for multiyear supply contracts. Those agreements give Samsung revenue certainty and reduce the volatility that has historically characterised the memory market. They also mean a significant portion of future production is spoken for before it is made.
The Production Problem
The natural question is why manufacturers cannot simply build more capacity to meet demand. The answer is that they can, and they are — but semiconductor manufacturing does not respond quickly to market signals.
A modern chip fabrication plant is one of the most complex and capital-intensive structures ever built. The equipment inside is bespoke, the tolerances are measured in nanometres, and the process of designing, constructing, commissioning, and achieving stable commercial yields from a new facility typically takes upwards of three years. Samsung's own assessment points to this timeline as the central reason additional supply cannot arrive fast enough to meaningfully ease the current shortage.
The investment decisions being made today will produce output in the latter half of this decade. Independent analysts at TrendForce broadly agree, expecting AI to remain the dominant driver of memory demand throughout 2027, with substantial increases in mainstream production unlikely to materialise until 2028 at the earliest.
The Knock-On Effects Beyond AI Hardware
One complication worth understanding is that the shortage is not confined to the specialist chips that AI systems use most intensively. High Bandwidth Memory, enterprise solid-state drives, and advanced server DRAM are where manufacturers are concentrating production because that is where demand and margins are highest. That prioritisation leaves less manufacturing capacity for the standard memory chips that go into laptops, desktop PCs, networking equipment, smartphones, and everyday business hardware.
The ripple extends further still. Analysts report shortages beginning to affect older generations of memory as manufacturers and hardware vendors look for alternatives when newer products are difficult or expensive to source. What starts as an AI infrastructure problem works its way through the supply chain into equipment that has nothing to do with AI.
Samsung's warning sits alongside similar signals from elsewhere in the supply chain. Seagate has indicated that most of its nearline hard drive production is already allocated through 2028, with cloud providers reserving capacity years ahead. Memory, storage, and data centre infrastructure are all under sustained pressure simultaneously, suggesting that AI is reshaping the entire technology supply chain rather than just one specialist corner of it.
What This Means for Business Hardware Decisions
For businesses that buy and manage technology, Samsung's outlook has practical implications that are worth taking seriously.
First, lead times and prices for servers, storage, and business laptops are unlikely to return to pre-shortage norms within the next planning cycle. Organisations that are budgeting for technology refresh on the assumption that the market will normalise soon may find those assumptions challenged.
Second, the businesses best positioned to manage this environment are the ones that are planning further ahead than usual. Longer procurement cycles, earlier engagement with suppliers, and where feasible, extended refresh timelines for existing hardware that is performing adequately, all reduce exposure to tight supply and elevated prices.
Third, and perhaps most importantly for businesses thinking about their own AI strategies, the cost and availability of infrastructure is not a background consideration. It is a central variable. The economics of running AI systems, whether in cloud environments or on-premise, are directly tied to the same supply constraints that Samsung is describing. Understanding that relationship helps organisations make more informed decisions about when to invest, what to commit to, and how to build flexibility into technology plans.
If you want to think through how AI infrastructure decisions fit into a broader managed IT strategy, our Managed IT Services page is a practical starting point.