OpenAI has confirmed it will work with Samsung Electronics to help design its next generation of custom AI accelerator chips, easing some of the supply chain pressures that come with being a fabless chip designer. The partnership, announced by OpenAI Korea general manager Harrison Kim, follows weeks after OpenAI unveiled its first-generation "Jalapeño" inference chips, which reportedly outperformed Nvidia's Blackwell GPUs on several inference workloads. The tie-up matters because most leading-edge chip manufacturing is dominated by a single firm, TSMC, forcing OpenAI to compete with AMD, Nvidia and other hyperscalers for capacity.

Samsung could help OpenAI in two main ways: supplying high-bandwidth memory (HBM) for its next chip, likely dubbed "Habanero", and potentially acting as an alternative foundry to TSMC using Samsung's 2-nanometre process. Only three firms make HBM at scale — SK Hynix, Micron and Samsung — and given an ongoing memory shortage, securing early supply deals is seen as important, particularly as future chip designs may use custom base dies requiring close coordination with memory suppliers. Samsung, the world's second-largest leading-edge foundry, previously made chips for Nvidia and Apple but lost ground to TSMC due to advanced packaging trends and past yield issues on its 3nm process, though its prospects are reportedly improving in 2026.