NextFin News - SK Group Chairman Chey Tae-won will travel to San Jose, California, this month to meet with Nvidia CEO Jensen Huang at the GTC 2026 conference, marking a critical juncture in the battle for AI memory supremacy. The meeting, confirmed by industry sources on Thursday, represents the first time the SK chief has attended Nvidia’s flagship developer event. It follows a high-stakes "chicken and beer" dinner between the two leaders in February, signaling that the informal alliance between the world’s dominant AI chipmaker and its primary memory supplier is entering a more formal, strategic phase as the industry pivots toward the next generation of high-bandwidth memory.
The timing of the summit is no coincidence. Nvidia is widely expected to use GTC 2026 to pull back the curtain on its "Vera Rubin" platform, the successor to the Blackwell architecture that has defined the AI boom. Central to the performance of these new accelerators is HBM4, the sixth generation of high-bandwidth memory. While SK hynix has long enjoyed a comfortable lead, controlling roughly 62% of the HBM market as of mid-2025, the landscape is shifting. Samsung Electronics recently claimed a major psychological and technical victory by announcing the world’s first mass shipment of HBM4 in February, reportedly passing Nvidia’s qualification tests with scores that outperformed industry standards in power efficiency by 37%.
For Chey, the mission in San Jose is defensive as much as it is collaborative. Although Nvidia has reportedly allocated approximately two-thirds of its HBM4 orders for the current year to SK hynix, Samsung’s early momentum threatens that dominance. The SK chairman is expected to discuss not just supply volumes, but the deep integration of SK’s custom HBM4 solutions into Nvidia’s roadmap. This includes the potential for "base die" collaboration, where the logic layer of the memory chip is manufactured using advanced foundry processes—a move that could see SK hynix leveraging its partnership with TSMC to counter Samsung’s "all-in-one" internal manufacturing capabilities.
The stakes extend beyond the silicon itself. The two leaders are slated to discuss broader cooperation in AI data center infrastructure and energy solutions. As U.S. President Trump’s administration continues to emphasize domestic energy independence and the build-out of high-tech infrastructure, the massive power requirements of Vera Rubin-class data centers have become a bottleneck. SK Group, through its energy and construction affiliates, is positioning itself as a holistic partner that can provide the cooling systems and power management necessary to keep Nvidia’s hardware running at peak capacity.
Market dynamics suggest that the "HBM gold rush" is entering a phase of diversification. While Micron has carved out a niche in SOCAMM2 memory for standalone CPU systems, the HBM4 race is increasingly a two-horse sprint between the Korean giants. Samsung’s aggressive capacity expansion—aiming for a 50% increase by the end of 2026—puts immense pressure on SK hynix to maintain its yield rates and technological edge. The GTC meeting will likely serve as the venue where the "preferred partner" status for the 2027-2028 cycle is quietly negotiated.
Ultimately, the presence of the SK chairman at a developer-focused conference underscores how the hierarchy of the semiconductor supply chain has been upended. Memory is no longer a commodity bought off the shelf; it is a co-designed component essential to the viability of the world’s most powerful AI systems. As Huang prepares to take the stage in San Jose, the private discussions held with Chey will determine whether SK hynix can protect its fortress or if the momentum has truly begun to swing back toward its cross-town rival.
Explore more exclusive insights at nextfin.ai.
