The bottlenecks are not just financial. Chey stressed that building a new memory fabrication plant requires at least three years from decision to operation, making rapid supply responses impossible . This inherent lag is a core reason he views the shortage as structural rather than merely a passing spike in demand
.
Chey’s outlook extends far beyond the typical two-year memory cycle. At a global media roundtable, he stated, "We still hold to the earlier projection that the memory shortage will persist through 2030," attributing the relentless demand to the proliferation of AI data centers, AI factories, and AI-powered PCs . The HBM segment, critical for Nvidia’s GPU platforms, is at the center of this dynamic. The demand for these specialized, high-performance chips means the memory bottleneck isn't simply about capacity but about producing the right kind of highly complex components at scale
.
The quantitative picture from Goldman Sachs’ June 1, 2026, note underscores Chey's warning with hard numbers. Analysts Giuni Lee and James Schneider project the AI-driven memory squeeze will be tighter in 2027 than in 2026, with undersupply persisting into 2028. Their specific forecasts point to a DRAM undersupply of 5.0% in 2026, worsening to 5.9% in 2027, alongside NAND shortages of 4.4% and 4.6% respectively. HBM shortfalls are expected to remain even more acute .
Earlier in the year, the bank had already been sounding the alarm, describing this as the largest DRAM supply gap in 15 years and a record NAND shortfall . Goldman’s position has since hardened into a “Higher for Longer” thesis for the memory boom
.
Price Targets Raised: In a March 2026 update reflecting the strengthening pricing environment, Goldman Sachs raised its target price for SK Hynix to 1.35 million won from 1.2 million won, and increased its Samsung Electronics target to 260,000 won from 205,000 won, maintaining Buy ratings on both. The analysts called the ongoing memory upturn the strongest on record .
Chey Tae-won’s presence at Computex in Taiwan — home to TSMC, a critical partner in advanced packaging — carries strategic weight. While specific new partnership announcements with TSMC regarding Nvidia’s next-generation Vera Rubin platform were not detailed in his Computex remarks, Chey conducted meetings with executives in Taipei . The broader context is clear: SK Hynix is a dominant supplier of HBM for Nvidia’s current architectures, and maintaining that edge through collaboration with Taiwan’s semiconductor ecosystem is essential as it scales capacity.
Chey also confirmed that SK Hynix has filed to list American depositary receipts in New York this year, signaling ambitions to tap deeper capital markets to fund its massive, multi-year expansion .
Sources did not include Goldman Sachs’ specific estimate for the total HBM market size by 2028, nor detailed commitments tied to the Vera Rubin platform at this event.
Comments
0 comments