In September 2026, OpenAI said it was working with Samsung on next-generation chips. The development adds a new connection to the competition between Samsung and SK Hynix over high-bandwidth memory, or HBM, used in AI systems.
OpenAI’s work with Samsung adds a new AI-chip connection
The companies’ reported work concerns next-generation chips. Samsung and SK Hynix were also reported to have joined OpenAI’s Stargate project in October 2025.
A separate Korean initiative plans to deploy 260,000 Nvidia GPUs across Samsung, SK Hynix, Hyundai, Naver, and the South Korean government. Samsung’s reported allocation is around 50,000 GPUs.
Why HBM matters—and where Samsung and SK Hynix stand
HBM is specialized memory that lets AI processors access large amounts of data quickly. It is important to systems that train and run AI models, making memory supply a significant part of the AI hardware race.
SK Hynix has held a long-standing lead in HBM. A reported figure puts Samsung and SK Hynix together at 83% of the global memory-chip market; that figure covers memory chips overall. It is a broader measure than HBM alone.
Nvidia’s allocations and the supply-chain squeeze
Nvidia’s decisions about HBM allocations can influence suppliers’ market shares. Capacity also matters: the wider memory industry faces production constraints, and suppliers are using customer deals and long-term agreements to tailor HBM products.
The supply chain also faces export-control changes. In 2025, the United States ended the Validated End User Program and replaced it with annual licensing for shipments of export-controlled tools to semiconductor plants in China. Samsung and SK Hynix both have legacy plants in China and continue testing new chipmaking tools there.
