OpenAI May Tap Samsung Foundry for AI Chips, Adding Second Source Alongside TSMC
OpenAI's Korea general manager indicated that the company is expanding its partnership with Samsung to include chip production, potentially using Samsung Foundry for some of its AI accelerators. This would mean OpenAI sources from both TSMC and Samsung, suggesting very high demand for its custom silicon. The details remain unclear, but it could involve the existing Jalapeño inference chip or future designs.
OpenAI's existing Jalapeño inference accelerator is currently manufactured by TSMC, with design work handled by Broadcom. The company's Korea chief stated that joint production and research with Samsung targets "next-generation" processors, which are approaching the tapeout stage, rather than the current chip.
This potential dual-sourcing follows a $200 billion Samsung-Broadcom alliance announced in July, covering foundry, memory, and packaging. Given OpenAI's 10GW procurement agreement with Broadcom, utilizing both major foundries would help meet the massive silicon volume required for such large-scale AI infrastructure.
Diversifying production across TSMC and Samsung could strengthen supply chain resilience for AI hardware, potentially reducing delays in data center expansion. This may lower costs and speed up AI service deployment, affecting businesses and consumers who depend on generative AI. It could also heighten geopolitical competition over advanced manufacturing, as nations seek control over critical technology. Ultimately, the move may signal that AI demand is now substantial enough to sustain multiple foundry partners.