SK Group and Nvidia announced on July 24, 2026 what both companies called a $500 billion-plus strategic partnership spanning AI data center buildouts and next-generation memory, with SK hynix and Nvidia forming a long-term alliance to co-develop and secure supply of HBM and future memory generations, according to a joint release carried on Nvidia's and SK hynix's newsrooms.
That followed earlier reporting from Digitimes that SK hynix had begun mass-production shipments of 12-layer HBM4 to Nvidia at the end of June 2026, ahead of a broader ramp expected in the following months.
Why it matters
HBM has become the binding constraint on how many AI accelerators Nvidia and its rivals can actually ship, since each GPU package needs a fixed allotment of memory stacks. A formal, long-term supply agreement with the leading HBM producer gives Nvidia more predictable access to a bottlenecked component ahead of its Vera Rubin platform.
For SK hynix, winning a reported two-thirds of Nvidia's next-generation HBM orders, as Korea JoongAng Daily reported in January 2026 citing industry sources, would extend the company's lead over Samsung and Micron in the highest-margin segment of the memory market.
How it works
HBM4 is the sixth generation of high-bandwidth memory, stacking DRAM dies vertically and connecting them with through-silicon vias to a wide interface that sits next to the GPU compute die on the same package. The 12-layer configuration SK hynix is shipping to Nvidia stacks more dies per module than earlier generations, increasing capacity per stack.
SK hynix's newsroom described the Nvidia partnership as covering not just HBM supply but joint memory architecture work, positioning SK hynix closer to a co-design role with Nvidia rather than a purely transactional supplier relationship.
The competing read
Other coverage frames the deal as much broader than a memory contract, tying HBM supply to SK Telecom's plans to build an AI data center of up to 2 gigawatts using Nvidia's DSX systems, which means the memory commitment is bundled with much larger infrastructure spending commitments whose scope and financing are still being worked out.
TheNextWeb's June 2026 coverage described the arrangement as giving SK hynix a structural edge specifically for Nvidia's upcoming Vera Rubin platform, a narrower and more verifiable claim than the headline dollar figure, which spans multiple businesses and years.
What happens next
SK hynix's broader HBM4 ramp, which Digitimes described as following the initial June shipments, is expected to scale through the rest of 2026, and subsequent earnings disclosures should show how much of that output is tied specifically to Nvidia versus other GPU makers.
Samsung and Micron are both pursuing their own HBM4 qualification and supply deals, so the next signal to watch is whether Nvidia diversifies its HBM4 sourcing away from the two-thirds concentration reported earlier in the year.