Jensen Huang just told the world that Nvidia will be the first company to get its hands on HBM4 memory chips. SK Hynix, the South Korean memory giant, has reportedly locked up roughly 70% of Nvidia’s HBM4 orders for the upcoming Vera Rubin AI platform. That share significantly exceeds earlier estimates that pegged it closer to 50%, making SK Hynix the dominant supplier in what’s shaping up to be the most important memory technology cycle in years.
What HBM4 actually means
SK Hynix achieved the world’s first 12-layer HBM4 development back in September 2025. The company is now gearing up for mass production, with Nvidia positioned as the launch customer.
In June 2026, the two companies formalized a multi-year co-development agreement extending through 2030, complete with advance payments from Nvidia.
This isn’t Nvidia’s first rodeo with SK Hynix. The memory maker previously supplied HBM3E chips for Nvidia’s Blackwell GPUs, which became the backbone of the current AI training infrastructure. While Samsung and Micron have both been qualified by Nvidia to produce HBM4, SK Hynix holds the largest market share by a comfortable margin.
What this means for investors
The Nvidia-SK Hynix partnership signals something broader about AI infrastructure spending. Companies are no longer making one-off procurement decisions. They’re signing multi-year supply agreements with advance payments, treating AI compute components the way energy companies treat long-term fuel contracts.
For crypto investors, this creates a two-track investment thesis. The first track is straightforward: Nvidia and SK Hynix are cementing their positions as the picks-and-shovels plays of the AI revolution. SK Hynix’s 70% share of HBM4 orders makes it arguably the most leveraged pure-play on AI memory demand.
The consolidation of memory supply around SK Hynix also introduces concentration risk into the AI hardware stack. A single production disruption at SK Hynix’s facilities could cascade through Nvidia’s product timeline and, by extension, every industry that depends on cutting-edge GPU availability.