오픈에이아이(OpenAI)는 차세대 AI 프로세서의 공급을 삼성과 TSMC 간 2배로 확대할 계획임을 확인하며, 자체 실리콘에 대한 막대한 수량 요구를 시사했다.

OpenAI is expanding its relationship with Samsung beyond memory supply and enterprise software, planning to outsource the production of at least some of its processors to Samsung Foundry. Harrison Kim, General Manager of OpenAI Korea, revealed the initiative this week. If accurate, sourcing AI accelerators from both TSMC and Samsung Foundry indicates massive volume requirements.

“One of the areas where we have made the most progress and gained the most recognition with Samsung Electronics is our joint production and research on the next-generation chips we are developing,” said Harrison Kim, General Manager of OpenAI Korea, at a press conference in Seoul, Reuters reports.

OpenAI currently operates its own AI ASIC program leveraging Broadcom’s design services alongside TSMC’s wafer processing and advanced packaging capabilities. The company recently introduced its first inference AI accelerator, Jalapeño, which was defined by its engineers, co-designed with Broadcom, and manufactured by TSMC in less than 18 months.

It remains unclear whether Samsung’s involvement represents a second production source for Jalapeño, a different processor currently under development, or a chip jointly developed by Samsung and OpenAI. While Samsung and SK hynix already supply memory for OpenAI’s Stargate data center initiative, this new focus on joint chip development and production is largely unrelated to DRAM supply agreements.

The first-generation Jalapeño is unlikely to be double-sourced, as it is already in mass production at TSMC and OpenAI specifically referenced “next-generation chips.” Development of Jalapeño’s successor is well underway and approaching tapeout, meaning mass production is imminent. Consequently, it is entirely possible that OpenAI will manufacture its second-generation inference ASIC at Samsung Foundry.

This potential shift aligns with broader industry partnerships. In late July, Samsung Electronics and Broadcom announced a strategic collaboration valued at over $200 billion through 2030, focusing on advanced foundry, memory, and packaging technologies for AI infrastructure. Since OpenAI holds an agreement with Broadcom to procure 10GW of custom AI accelerators, it would be positioned to fabricate these chips across both Samsung and TSMC facilities, provided it covers the costs for Broadcom to port the designs to Samsung’s manufacturing processes.

OpenAI may be following Tesla’s precedent by double-sourcing Jalapeño’s successor to secure higher volumes. However, Tesla’s scale differs significantly. The automaker requires extraordinary volumes of its AI5 processor to support three distinct high-volume applications: AI data centers, vehicles, and Optimus robots. Tesla could need millions of AI5 chips for vehicles alone, in addition to deployments for robots and data centers. Securing physical implementations from both TSMC and Samsung provides Tesla with essential supply-chain resilience and the aggregate capacity required to serve multiple product lines.

Data center accelerators, however, are vastly more silicon-intensive per unit than ASICs designed for vehicles or robots. If OpenAI and Broadcom’s next ASIC is a large, leading-edge processor featuring multiple dies, and OpenAI aims to deploy gigawatts of computing power, wafer demand could quickly outstrip TSMC’s available capacity, which is already heavily booked by competitors like AMD and Nvidia. In such a scenario, securing a secondary foundry would become necessary. While this remains speculative, the strategic logic points toward dual-sourcing to meet unprecedented scale.