OpenAI’s AI Processor Is Almost Ready for TSMC Fabrication

OpenAI will partner with Taiwanese chipmaker TSMC to produce its custom-design, AI processor in the next few months, according to a new report . Although details of the processor are still almost non-existent, OpenAI has been working on the design since 2023 and will leverage TSMC's cutting-edge 3nm process technology when it enters production later this year.

As it stands, Nvidia has a near-monopoly on AI processors for training and inference (actually running the models). That's led to a huge surge in its stock price as companies around the world scrambled to join in the AI revolution of the past few years. In turn, companies like Microsoft, Google, and Meta have developed their own AI hardware, and OpenAI will soon join that collective with its first AI chip design.

OpenAI recruited a team of 40 engineers to develop the processor, recruiting former Google chip designer Richard Ho to head the endeavor. It is partnering with semiconductor manufacturer and designer Broadcom. The design is said to include high-bandwidth memory (HBM) and built-in networking capabilities.

The Nvidia H100 Hopper graphics chip is one of the premier AI training chips in the world. They're in such high-demand that Nvidia ships them in armored cars.Credit: Nvidia

The first-generation OpenAI chip design will reportedly be focused on inference roles to start with, though it hasn't ruled out developing a chip for training AI in the future. Although the designs are going to be sent over to TSMC in the next few months, mass production isn't expected to begin until sometime in 2026, and initial rollout will be limited to start with. However, over time, OpenAI is expected to iterate on the design, potentially improving performance and efficiency and expanding its use and capabilities.

Developing this first-generation processor may have cost OpenAI as much as $500 million, or up to double that when companion software and hardware development are factored in, according to the report. However, it's also been suggested that it is as much about giving OpenAI extra leverage in its negotiations with existing chip makers, as it is about easing the company's reliance on Nvidia technology.

This news comes alongside record investment from other companies and countries in AI ventures for the future. Microsoft has announced an $80 billion investment in 2025, while Meta is slated to spend as much as $60 billion on its own AI developments. Similarly, OpenAI announced the $500 billion Stargate infrastructure project, headed by the new US administration, and the UK government recently announced an equivalent $14 billion investment in data centers for AI deployment and training.