OpenAI is deepening its strategic alliance with Samsung, moving beyond the supply of high-performance memory to explore a broader collaboration on next-generation AI silicon. Harrison Kim, General Manager of OpenAI Korea, confirmed that significant progress has been made in the joint research and production of future chips, signaling a potential shift in OpenAI's hardware strategy.
While the partnership initially focused on memory—a critical bottleneck for scaling complex computing—the scope is now expanding. Samsung has already been integrated into the Stargate project as a strategic partner, with estimates suggesting OpenAI's memory demand could reach 900,000 DRAM wafers per month.
From Memory to Foundry and Packaging
The current cooperation centers on the Jalapeño inference accelerator, which is manufactured by TSMC but utilizes Samsung's HBM4 memory. However, the latest developments suggest that OpenAI may now be eyeing Samsung's foundry and advanced packaging capabilities for its future roadmap.
OpenAI has already confirmed that a second-generation chip is in advanced development, with a third generation also taking shape. While TSMC remains a primary manufacturer, the possibility of diversifying production to Samsung's nodes could reduce dependency on a single supplier.
A Broader Trend in Custom Silicon
This move mirrors a wider industry shift toward custom AI hardware. Samsung is already attracting other major players; for instance, Anthropic is reportedly exploring Samsung's 2nm process and packaging options.
For Samsung, securing a deeper role in OpenAI's roadmap is a strategic priority to compete with TSMC, especially as the company pushes its HBM5 memory and 2nm GAA processes to maintain a competitive edge in the AI infrastructure market.
No comments yet. Be the first!