Samsung memory shortage warning extends to 2028 as AI demand lifts profit

Samsung Electronics said the Samsung memory shortage 2028 outlook is now its base case, warning that AI server demand is tightening supply while the company reports record earnings. The Korean electronics group told analysts that demand from agentic AI and model training is pushing customers to seek more memory for servers and computing infrastructure, while supply for PCs, smartphones and other mainstream devices is being squeezed.

Jaejune Kim, Samsung’s executive vice president of memory, said on the company’s calendar second-quarter earnings call that new fabrication capacity will take time to affect the market. According to Kim, the period from starting construction of a fab to producing wafers is more than three years, limiting the industry’s ability to add meaningful supply before 2028.

“We believe it will be unlikely to see any significant increase in incremental supply through 2028,” Kim told analysts, according to Samsung’s discussion of results for the quarter ended June 30. He said customer requests suggest some demand that goes unmet this year will roll into next year, making supply tighter.

Why does Samsung expect the memory shortage to last until 2028?

Samsung’s stated reason is capacity timing. Memory manufacturers are spending on new capacity, but fab construction, equipment installation and wafer production do not translate into additional bit output quickly enough to relieve the current shortage in the next several quarters.

Kim said Samsung expects supply constraints to worsen in 2027 compared with 2026, supporting the company’s view that the shortage will last through 2028. Visibility beyond that is limited, he said, though higher token processing demand from AI agents is expected to keep pressure on memory requirements over the medium to long term.

The constraint is also being shaped by allocation decisions. Samsung said memory makers are favoring parts used in AI servers, which leaves fewer mainstream DRAM and NAND components available for PCs, smartphones and other devices. That is good for memory pricing and supplier margins, but it puts pressure on enterprise buyers and consumer hardware vendors that do not sit at the top of the allocation list.

Samsung also said customers seeking large AI service infrastructure are asking for multiyear supply commitments. Kim said those agreements fit Samsung’s goal of reducing medium- and long-term risk in a market known for sharp cycles. He added that Samsung is prioritizing customers able to guarantee committed future captive demand. Samsung did not identify those customers or disclose contract values.

The company expects bit output next quarter to rise by the mid-single digits for DRAM and by the high single digits for NAND flash. It said it plans to expand sales of higher-performance memory products, including HBM4, HBM4E, DDR5, SOCAMM2 and enterprise SSDs.

The financial effect is already visible. Samsung reported revenue of ₩171.5 trillion, or about $119 billion, for the second quarter, up 130% from a year earlier. Operating profit rose more than nineteenfold year on year to ₩89.5 trillion, or about $62.4 billion.

TrendForce, a Taiwan-based market research firm, broadly agrees that AI will remain a main driver of DRAM demand in 2027. The firm expects several suppliers to start new DRAM lines in 2027, but says construction schedules and equipment installation mean meaningful production increases are more likely in the second half of 2027, with substantial additional output not arriving until 2028.

NAND may loosen earlier. TrendForce expects NAND flash supply to ease in the second half of 2027 as more advanced layered products and new fabs gradually ramp, producing stronger bit supply growth than in 2026. That creates a split market: DRAM remains tied to AI server demand and delayed capacity, while NAND has a clearer path to added output next year.

This story draws on original reporting from The Register.