According to DealSite, SK hynix has already begun shipping HBM4 to NVIDIA in small volumes. This gives it the inside track as NVIDIA gears up to launch its next-gen Rubin AI graphics hardware around the fourth quarter of 2025.

HBM4 is the first standard to properly fuse memory and logic elements in a single package. That combo is tipped to be a game-changer for AI compute, and SK hynix looks set to cash in.

Micron came second for consideration becuase it had weaker yield rates and headaches retooling production lines. Meanwhile, Samsung’s HBM4 yields are a mystery, though the company did get a recent boost after AMD picked up its HBM3E modules. That win could improve Samsung’s standing and help claw its way into NVIDIA’s pool of HBM4 suppliers, but nothing’s confirmed.

SK hynix's foothold with HBM4 means its market share is about to grow. The firm is now considered NVIDIA’s top-tier memory partner, a status likely to bring in the lion’s share of orders.

Still, NVIDIA won’t be putting all its memory eggs in one basket. The company is expected to stick to a multi-sourcing strategy, so Micron and possibly even Samsung could land some volume if they can meet expectations.

Rubin GPUs will be paired with TSMC’s N3 process, eight stacks of HBM4, and a 3 nm Grace CPU, with a sharp focus on power efficiency. A customer qualification run is planned for September, keeping NVIDIA on track for its six-month cadence.

With SK hynix supplying solid volumes and HBM4’s integration edge, NVIDIA looks well-positioned to meet AI demand, assuming the rest of the supply chain doesn’t throw a spanner in the works.