 |
| Photo: OpenAI CEO Sam Altman |
These high-stakes discussions on collaborative chip design and server hardware are fundamental to OpenAI's monumental ambitions, including massive initiatives like "Stargate," which aims to construct data centers and "AI factories" valued at hundreds of billions of dollars over the next several years.
While the meetings were not publicly announced by any of the involved parties, Altman's presence in the region is undeniable. Just a day later, on Wednesday, September 30, he was officially confirmed to be in South Korea for a meeting with President Lee Jae Myung. The Korea Times reported that Altman secured non-binding agreements to build a 20-megawatt data center in Pohang in partnership with Samsung, and a second facility in South Jeolla province with SK Hynix.
However, the alleged meetings in Taiwan are considered to be of even greater long-term significance. While OpenAI is striking deals worldwide to build out its AI infrastructure, these facilities will require hundreds of thousands of advanced chips to function and TSMC and Foxconn are the linchpins of that supply chain. TSMC is the world's largest and most advanced contract chip designer and manufacturer, while Foxconn is the top supplier for Oracle, the cloud computing firm with which OpenAI has a staggering $300 billion compute capacity deal.
Further cementing these ties, Foxconn is also an indispensable partner to Japanese investment firm Softbank, a major investor in both OpenAI and its various data center projects. Foxconn is slated to manufacture hardware for Softbank at the Ohio facility it sold to the company earlier this year.
Altman's discussions with TSMC and Foxconn were likely aimed at solidifying their collaboration for OpenAI's global expansion and gaining deeper insight into the cutting-edge process nodes planned by TSMC for future chip designs.
A pivotal topic that was almost certainly on the agenda is OpenAI's own ambition to design bespoke chips. Currently, the company's models run on hundreds of thousands of expensive and power-intensive GPUs from Nvidia. In a strategic move to break this dependency, OpenAI, like many other tech giants, is developing its own custom AI-specific integrated circuits (ASICs) optimized for AI inference.
OpenAI reportedly established an in-house ASIC design team in 2024 and is believed to be working with Broadcom to develop a custom AI chip on TSMC's state-of-the-art 3nm process. By integrating high-bandwidth memory and advanced packaging technologies, OpenAI's custom hardware is now slated for mass production in the third quarter of 2026, following a recent delay. Securing a direct line to the world's top foundry is a critical step in making that a reality.