OpenAI Partners with Samsung Foundry for Next-Generation AI Chips, Diversifying from TSMC

OpenAI is set to outsource production of at least some of its processors to Samsung Foundry, Harrison Kim, General Manager of OpenAI Korea, revealed at a press conference in Seoul. The announcement, reported by Reuters, signals a deepening collaboration between the two companies that already includes Samsung's memory supply for OpenAI's Stargate data center initiative.

"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," Kim said.

OpenAI already operates its own AI ASIC program, relying on Broadcom's design services and TSMC's wafer processing and advanced packaging. Earlier this year, the company introduced its first inference AI accelerator, dubbed Jalapeño, which was defined by OpenAI engineers, co-designed with Broadcom, and manufactured by TSMC in under 18 months.

Kim did not specify whether Samsung's involvement concerns a second production source for Jalapeño, an entirely new processor, a jointly developed chip, or another aspect of chip production. The distinction matters because Samsung and SK hynix already supply memory for OpenAI's Stargate project, whereas foundry services involve fundamentally different manufacturing processes.

If confirmed, the dual sourcing arrangement would make OpenAI one of the few AI companies to work with both TSMC and Samsung Foundry simultaneously, underscoring the massive scale of its hardware ambitions.

Why This Matters

This move could reduce OpenAI's reliance on TSMC, which dominates advanced chip manufacturing but faces capacity constraints and geopolitical risks (primarily Taiwan's vulnerability).
Samsung Foundry's involvement suggests OpenAI anticipates enormous chip volumes for training and inference, potentially accelerating the deployment of its Stargate supercomputing network.
A dual-sourcing strategy may pressure TSMC to prioritize OpenAI's orders or improve pricing, reshaping the competitive dynamics among AI chip designers.

Background

OpenAI, the company behind ChatGPT, has been investing heavily in custom silicon to reduce dependence on Nvidia's GPUs, which are costly and in short supply. Its first ASIC, Jalapeño, was designed for inference workloads and taped out at TSMC. Samsung Foundry has long struggled to win major AI chip contracts, lagging behind TSMC in advanced nodes (3nm and below). Expanding from memory supply to foundry services would represent a significant strategic win for Samsung's semiconductor division.

Key Perspectives

OpenAI: The company gains supply-chain redundancy and potentially better pricing by splitting production between two foundries, enabling it to scale its AI infrastructure more aggressively. Samsung Foundry: Securing OpenAI as a customer would validate its manufacturing capabilities and provide a major revenue stream, helping it compete with TSMC's dominant position. Critics: Samsung's yield rates on advanced nodes have historically trailed TSMC's, and shifting between foundries introduces design complexity and verification challenges. The lack of specificity in the announcement raises questions about the scope and timeline of the partnership.

What to Watch

Official confirmation from OpenAI or Samsung on which specific chips are involved (Jalapeño or a new design).
Announcements regarding Samsung's foundry node and timeline (e.g., production on SF3 or SF2).
Capacity allocation details at TSMC and how this affects other AI chip customers.

ZOTPAPER