Inside Nvidia's $4.9 Billion Bet on South Korea and the HBM4 Memory Crisis

• A commitment to double SK hynix's total wafer production capacity over five years. This was a strategic reversal from March of the same year, when SK Group Chairman Chey Tae-won said adding capacity wasn't planned .
• The construction of a $4.9 billion Nvidia-powered AI data center in South Korea, designed to anchor the country's AI infrastructure buildout .

In perhaps the most symbolic gesture of the trip, Huang visited the SK hynix booth at Computex and wrote "Please make more" directly on an HBM4E wafer, creating a viral moment that underscored the desperate urgency for supply .

The partnership is critical because of sheer scale. South Korean firms supply roughly 70% of the memory for Nvidia's AI chips, and Nvidia has deployed over 250,000 GPUs in the country with plans to supply another 260,000 .

HBM4 Qualifications Confirmed for Vera Rubin

The most concrete supply-chain signal came when Huang confirmed that Samsung Electronics, SK hynix, and Micron have all been certified as HBM4 suppliers for Nvidia's next-generation AI platform, Vera Rubin . The platform is now in full production, ending months of uncertainty .

Huang announced the news on arrival at Seoul's Gimpo airport on June 5, stating, "All three vendors have been qualified. All three vendors are in production, and they're competing to deliver HBM4 for shipment in Q3 2026" . Having all three suppliers qualified simultaneously is significant—it diversifies Nvidia's supply base at a time of extraordinary demand and places competitive pressure on the memory giants to deliver.

Samsung's qualification is a particularly notable recovery. After falling behind SK hynix in HBM, Samsung cleared Nvidia's tough qualification tests at the 10 and 11 gigabit-per-second data rates the Rubin design demands .

The Vera Rubin platform, which will combine the Vera CPU with Rubin GPU clusters, represents a massive generational leap. Huang revealed that its supply chain is twice as large as the previous Grace Blackwell generation . The platform will use SK hynix DRAM not just in GPUs but in the innovative Vera CPU itself .

The Memory Shortage: A Multi-Year Structural Crisis

Huang did not mince words about the supply situation. "The whole industry supply chain, everything from wafers to packaging to silicon photonics, everything's in short supply because the demand is so high," he said. "It is going to persist for several years" . He specifically warned the shortage will last "quite a few years" .

SK Group Chairman Chey Tae-won offered an even more precise, alarming timeline, projecting the global memory shortage could last through at least 2030. He explained the shortage is structural, driven by a fundamental lack of wafer capacity. "Securing additional wafers takes at least four to five years," he said, projecting an industry-wide supply shortfall of over 20 percent through the end of the decade .

This warning is part of a broader chorus. Intel's CEO warned shortages could continue until 2028 . The shortage has already pushed Nvidia to reportedly delay consumer GPU launches due to a lack of memory chips . With hyperscalers like Google, Microsoft, Meta, OpenAI, and Amazon locking in GPU contracts through 2026, the pressure on memory supply is structural rather than cyclical .

What This Means for the AI Industry

The convergence of these June 2026 announcements paints a clear picture. The transition to the Vera Rubin generation is proceeding at full speed with a diversified, qualified, and competing HBM4 supply base. Simultaneously, the memory industry's ability to meet that demand is fundamentally strained for years. SK Group's $4.9 billion data center and massive capacity doubling are attempts to close the gap, but Huang's handwritten plea on a wafer says it all: right now, demand is simply outpacing supply.