OpenAI to Tape Out First AI Chip This Year with TSMC

According to reports, OpenAI is actively developing its proprietary AI chip to reduce reliance on major industry players such as NVIDIA and other AI chip manufacturers. The latest updates indicate that the design phase of OpenAI's custom AI chip is expected to be completed in a few months. If all goes well, trial production could be handed over to TSMC (Taiwan Semiconductor Manufacturing Company) as early as the first half of this year.

Despite potential challenges, OpenAI appears determined to achieve this goal. According to CNBC, the trial production process, known as tape-out, typically takes around six months and costs several million dollars. However, expedited production may be possible if OpenAI agrees to pay a premium to TSMC. Even so, success is not guaranteed on the first attempt. Any failures would require repeating the tape-out process to address technical issues.

It was previously reported that OpenAI planned to use TSMC's A16 process for its Sora video generator chip. However, it remains unconfirmed whether this is the same chip currently under development.

This ambitious project is led by Thomas Norrie and Richard Ho, seasoned engineers who previously worked on Google's Tensor Processing Units (TPUs). The team size has reportedly grown from 20 to 40 members. OpenAI has also partnered with Broadcom for support, though the specifics of Broadcom's contributions are still unclear.

The name of OpenAI's custom AI chip has not yet been disclosed. Its primary functions will center on training and operating AI models, with initial applications expected to be limited. Over time, its role may expand based on the number of deployment units planned by OpenAI.

If the development proceeds as scheduled, mass production of OpenAI's first proprietary AI chip is expected to begin in 2026. TSMC will leverage its 3nm process technology, integrating systolic array architecture and high-bandwidth memory (HBM) for advanced performance capabilities.