SAN FRANCISCO — Nvidia has signed a deal worth up to $500 billion to secure high-bandwidth memory from SK Hynix, locking in a steady supply of the critical component for its AI graphics processing units and integrated systems.
HBM is a specialized type of RAM that delivers significantly higher bandwidth than traditional DDR memory. It is essential for handling the massive datasets and complex computations required by large language models and other AI applications. Its integration directly onto the GPU package reduces latency and boosts throughput.
SK Hynix is a leading HBM producer alongside Samsung and Micron. The company has said its HBM production is sold out through the end of 2025. Nvidia's long-term commitment reflects intense demand for these chips against limited manufacturing capacity across the industry.
The supply agreement extends Nvidia's competitive moat in the AI accelerator market. By securing HBM allocation, Nvidia reduces the risk of production bottlenecks that could limit its ability to meet demand for the H100 and upcoming B200 GPUs. Competitors including AMD and Intel face greater difficulty securing their own HBM supply.
Nvidia's stock traded at $206.84, down 0.9 percent on the day, as the broader Nasdaq fell 0.6 percent. The deal signals continued aggressive capital deployment in AI infrastructure, raising input costs across the industry.
