Samsung used its third-quarter call to convert a shortage into a calendar. Vice president Kim Jae-joon told analysts that demand keeps outrunning supply and that the gap should hold through 2028, with 2027 high-bandwidth memory already sold at agreed prices and ten customers on long-term contracts.
The mechanism matters more than the headline. New fabs take years to reach volume, and the generation change to HBM4 and beyond eats capacity twice over: each stack is larger, and converting ordinary DRAM capacity into HBM capacity yields fewer usable bits per wafer. Kim described that conversion ratio as a physical limit, not a scheduling problem.
Prices did the rest of the work. Average DRAM selling prices rose about 20 percent over the prior quarter, NAND in the high teens, while bit shipments moved only low to mid single digits. Profit came from price, not from shipping more silicon, which is why the company guided to a richer mix of high-bandwidth memory, server modules and enterprise flash in the fourth quarter.
Samsung also argued that even a sharp rise in industry capital spending in 2027 will not produce output in 2028, because much of the money goes to buildings and rising construction costs rather than tools, and tool lead times are stretching.
The read-through for anyone buying memory is straightforward and uncomfortable. Customers who want certainty are signing multi-year agreements at prices set before the shortage peaked. Those who wait will be buying spot.
