OpenAI is working with Samsung on next-generation AI chips in deepening partnership

OpenAI plans to collaborate with South Korean semiconductor giant Samsung Electronics to bolster its semiconductor supply chain. This partnership, announced at a press conference by OpenAI Korea general manager Harrison Kim, comes shortly after the company unveiled its first-generation "Jalapeño" AI accelerators, which outperformed Nvidia's Blackwell GPUs in various inference tests.

Kim highlighted the progress made in joint chip production and research with Samsung Electronics, stating that OpenAI still faces the challenge of being a fabless chip designer in a market dominated by TSMC. Securing memory capacity is crucial, as Samsung Electronics and Intel Foundry are among the limited suppliers of high-bandwidth memory (HBM).

The alliance could allow Samsung to supply OpenAI with HBM for its next-generation "Habanero" chips, which use HBM4 like AMD and Nvidia's latest GPUs. This move could provide OpenAI with early access to the scarce HBM supply. Additionally, the partnership may enable OpenAI to explore custom base dies, potentially using Samsung's advanced 2 nm process technology as an alternative to TSMC.

Samsung's expertise in producing accelerators for other chip designers suggests they could accommodate OpenAI's needs, potentially offering larger capacity allocations than those available from TSMC.