Goldman Sachs Sees AI-Driven Memory Chip Shortage Lasting Through 2028, Raises Samsung and SK Hynix Targets

Demand for high-bandwidth memory and advanced chips surged with the expansion of AI training, but a recent pullback in memory shares raised concerns that the industry cycle may be weakening. Morgan Stanley argues the retreat is a temporary disruption within an AI supercycle, as the rise of agentic AI and inference workloads makes memory an increasingly critical bottleneck for computing capacity.

Morgan Stanley’s semiconductor analyst has turned bullish on the memory sector after adopting a bearish view in July, saying the positioning correction is nearing an end. In the latest report, the analyst estimated that both Samsung Electronics and SK Hynix offer potential share-price upside of more than 60%, supported by growing memory requirements from agentic AI systems and inference workloads.

Demand for high-bandwidth memory, or HBM, and data-center chips is accelerating as technology companies expand the computing infrastructure needed to train artificial intelligence models. Supply growth remains constrained by advanced manufacturing, packaging capacity and lengthy customer qualification cycles. Samsung Electronics, one of the world’s largest memory producers, is therefore closely watched for its ability to improve HBM competitiveness and convert the AI investment boom into stronger earnings.

Goldman Sachs raised its 12-month price target for Samsung Electronics to 490,000 won in its latest research report, citing robust demand for HBM and data-center memory driven by AI training. The bank expects tight conditions across the global memory market, with demand continuing to outstrip supply, to persist through at least 2028. That outlook points to sustained pricing and earnings support for Samsung over the medium term.

The rapid buildout of generative AI infrastructure has made advanced memory a critical constraint for servers and data centers. Samsung Electronics expects supply growth to remain behind demand, leaving chip availability and pricing exposed for years. The outlook matters because a prolonged shortage could raise deployment costs and slow the expansion of computing capacity planned by the world’s largest technology companies.

Samsung said on its latest earnings call that some memory demand has already been pushed into 2027 and that tight supply could persist through 2028. The company has signed contracts lasting more than five years with the world’s five largest data-center operators, securing commitments for about two-thirds of its capacity. Investors, however, remain concerned about whether technology giants can sustain the massive capital spending required for AI infrastructure.

Kioxia is a major Japanese manufacturer of NAND flash memory, with products widely used in data centers and enterprise solid-state drives. Generative AI is fueling demand for high-performance storage, but memory producers are prioritizing capital spending on DRAM, limiting additions to NAND capacity. The resulting supply-demand gap has become a key factor affecting prices and Kioxia's earnings.

Goldman Sachs recently raised its Kioxia price target for the second time, to ¥116,000 per share, while maintaining a Buy rating. The report forecast that continued growth in demand for enterprise SSDs from AI data centers could extend the NAND shortage and price-increase cycle through the end of 2027. Tight supply may not begin to ease until 2028.

The boom in generative AI and data-center construction is driving demand for high-bandwidth memory, or HBM. Samsung Electronics, SK hynix and Micron are shifting resources toward higher-margin products, squeezing supplies of conventional DRAM and NAND. HBM is a critical component in AI accelerators, and shortages could affect server delivery times, electronics costs and the semiconductor industry cycle.

Micron recently reported better-than-expected results, while the market estimates SK hynix’s revenue could reach $54.8 billion. Samsung and SK hynix have said their main memory capacity is fully booked through 2027. The industry estimates that even with manufacturers accelerating HBM capacity expansion, the supply shortfall will still be about 40% at the end of 2027. New capacity may not begin gradually easing the pressure until after 2027 at the earliest.

The generative AI boom and wave of data-center construction have driven demand for high-bandwidth memory, or HBM, and DRAM, making memory a critical bottleneck in the AI chip supply chain. SK hynix is a major NVIDIA supplier, while TSMC provides advanced manufacturing and packaging. Cooperation among the three companies will directly affect global AI server capacity and costs.

SK Group Chairman Chey Tae-won predicted at NVIDIA GTC 2026 that the memory supply shortfall caused by expanding AI demand could persist through 2030. He also called TSMC an indispensable partner in building AI solutions. SK hynix is preparing a plan to stabilize DRAM prices and is considering issuing American depositary receipts, or ADRs, in the United States, but has not disclosed the size or timing of a potential offering.