KB Securities Says Samsung and SK hynix Inventories Are Below 10 Days, Forecasts DRAM and NAND Demand Growth More Than 10 Points Above Supply in 2027

Seoul Economic Daily reported on September 7 that KB Securities estimates memory inventories at Samsung Electronics and SK hynix are below 10 days of supply. The brokerage also forecasts that 2027 bit-demand growth for DRAM and NAND will exceed supply growth by more than 10 percentage points [1]. This is a sell-side forecast, not an audited inventory disclosure by either producer. Even so, it shifts the risk discussion from a shortage in HBM alone toward a scenario in which conventional DRAM and NAND could also become constrained as AI infrastructure expands.

The report attributes the forecast to cloud companies' AI capital expenditure, HBM4's use of wafer resources, and simultaneous growth in server DDR5 and enterprise SSD demand [1]. The sub-10-day inventory figure, memory's share of AI-infrastructure cost, and the size of the projected gap should all be read as KB Securities' judgment, not as industry statistics independently verified by manufacturers or another research body. The public report does not provide complete inventory definitions by product, customer, or producer, so it cannot establish current pricing, shipments, or revenue.

The demand backdrop, though not those precise figures, has separate public support. In its July second-quarter materials, SK hynix said customer demand exceeded the volume it could supply, that it had concluded long-term agreements with about 10 customers, and that HBM4 mass shipments began in the second quarter [2]. These are company statements, but they show why high-bandwidth products, server DRAM, and NAND need not be isolated supply questions. They do not validate KB's 10-day inventory estimate or its more-than-10-point gap.

John Wood's analysis is that the useful signal is not the phrase “tightest in history,” but whether constraints propagate across product classes. If HBM expansion does consume general-purpose DRAM wafer resources while AI servers add DDR5 and eSSD purchases, procurement adjustments could appear across memory products earlier than an HBM-only view suggests. That is an inference from the public materials, not a forecast of price or company earnings. The countercase is earlier customer stocking, a mix change, or new capacity that narrows the projected gap.

Watch for comparable inventory and mix disclosures from the two producers, delivered cloud capital expenditure, and whether actual deliveries, contract prices, and capacity additions for DRAM, NAND, and HBM corroborate one another. Only that later evidence can turn a sell-side tightness scenario into a verifiable market change.