OpenAI CEO Sam Altman reportedly took a quiet trip to Taiwan this week to hold meetings with representatives of TSMC and Foxconn and discuss collaborating on chip design and manufacturing, as well as AI server infrastructure development, according to Digitimes. These kinds of partnerships will be crucial if OpenAI is to meet its commitments as part of major initiatives like Stargate, to build hundreds of billions of dollars worth of data centers and "AI factories," over the next few years.
Although not announced by any of the involved companies or Altman himself, there's no denying he's in the region, as it was confirmed just one day later, Wednesday, September 30, that Altman had been in South Korea meeting with President Lee Jae Myung. Korea Times reported that he secured non-binding deals to build a 20-megawatt data center in Phang in partnership with Samsung, and another in the South Jeolla province with SK Hynix.
The alleged meetings in Taiwan were arguably more important, however. Although Altman and OpenAI have been doing deals around the world to build out AI infrastructure, it needs many hundreds of thousands of chips to do it with, and TSMC and Foxconn are set to be major suppliers. TSMC is the world's largest and most advanced chip design and manufacturing company, while Foxconn is the largest supplier for Oracle, the cloud computing company that OpenAI has struck a deal worth $300 billion for compute power.
Foxconn is an integral partner with Japanese investment vehicle Softbank, which is also heavily invested in OpenAI and a range of its data center projects. Foxconn will manufacture hardware for Softbank at the Ohio facility it sold to the company earlier this year.
Altman's meetings with TSMC and Foxconn, if real and fruitful, likely further entrenched their cooperation on scaling up OpenAI's global AI infrastructure, as well as gaining deeper insight into TSMC's planned advanced process nodes for future chip designs.
Another area likely discussed, and potentially more impactful for the future, is OpenAI's own chip design ambitions. As it stands, it relies on hundreds of thousands of expensive and power-hungry Nvidia GPUs. Alongside China and other countries and companies, OpenAI wants to untether itself from being so reliant on Nvidia moving forward, so it is developing its own AI application-specific integrated circuit (ASIC) chips for AI inference.
OpenAI reportedly formed an ASIC design team in 2024 and is alleged to be working with Broadcom to develop a custom AI chip on TSMC's advanced 3nm process. Combining high bandwidth memory and advanced packaging technologies, OpenAI's bespoke hardware is expected to enter mass production in Q3 2026, following a recent delay.
Follow Tom's Hardware on Google News, or add us as a preferred source, to get our up-to-date news, analysis, and reviews in your feeds. Make sure to click the Follow button!
