OpenAI Partners With Samsung to Secure Next-Gen AI Chip Supply

OpenAI has announced a deepening partnership with Samsung Electronics to design and produce next-generation AI accelerators. The collaboration is intended to fortify OpenAI's semiconductor supply chain by leveraging Samsung's dual capabilities in logic foundry and high-bandwidth memory.

According to Harrison Kim, OpenAI Korea General Manager, joint production and research on next-generation chips represent one of the areas where the company has made the most progress and gained the most recognition with Samsung. This move follows the debut of OpenAI's first-generation 'Jalapeño' inference chips, which reportedly outperformed Nvidia's Blackwell GPUs in several inference workloads, specifically regarding performance-per-watt and tokens per megawatt.

The Shift to Custom Silicon

OpenAI is currently transitioning from a purely fabless model—which relied heavily on third-party hardware from providers like Nvidia—toward the design of its own custom AI Application-Specific Integrated Circuits (ASICs). While the market for leading-edge silicon is currently dominated by TSMC, and high-bandwidth memory (HBM) is controlled by a small trio of players—SK Hynix, Micron, and Samsung—OpenAI is seeking to diversify its sources to ensure stability.

This diversification is critical for the scale of OpenAI's ambitions, including the massive Stargate data center initiative. The memory requirements for such a project are immense; Samsung estimates that OpenAI's demand for DRAM could eventually reach up to 900,000 wafers per month.

Industry Implications

This partnership signals a strategic intent by OpenAI to reduce its systemic dependency on Nvidia for compute and TSMC for fabrication. By integrating custom base dies via Samsung or establishing a second source for logic chips, OpenAI can potentially scale its infrastructure faster and more reliably than competitors who remain bottlenecked by the capacity constraints of a single foundry.

If successful, this shift could alter the balance of power in the AI hardware ecosystem, moving the industry away from a monolithic supply chain toward a more fragmented, specialized model where the largest AI labs control their own silicon destiny.

Future Outlook

While the partnership is confirmed, several technical specifics remain unverified. Reports have suggested the potential use of Samsung's 2nm process technology as an alternative to TSMC to avoid capacity limits, though this has not been explicitly confirmed. The primary focus remains the operationalization of the Samsung supply line to meet the staggering volume of DRAM required for future scaling.