OpenAI is planning to start mass-producing its own custom AI chip next year.

The chips, which have been designed in partnership with Broadcom, will be used internally by OpenAI, rather than being made available to external customers, the FT reported.

The report came after Broadcom’s CEO, Hock Tan, told analysts on Thursday, September 4, that the company had secured a fourth customer for its AI chip business that had committed to $10 billion in orders. Tan did not name the customer, but the FT reported that people with knowledge of the situation confirmed it was OpenAI.

It has long been rumored that OpenAI has been working with Broadcom to develop its own custom AI training and inference semiconductor, with the company expanding its semiconductor team in recent years to support the effort.

In August, DCD was the first to report that the generative AI giant hired former Google TPU (Tensor Processing Unit) software engineer Safeen Huda to join its chip team, joining Richard Ho – a previous leader of Google’s Cloud TPU project. The company currently has approximately 40 people working in its custom silicon team.

In addition to working with OpenAI on its AI chip, Broadcom was also heavily involved in the development of Google's TPU AI chip.

Currently, OpenAI is one of the largest users of Nvidia GPUs for both training and inference workloads, but has opted to make its own custom chips in order to reduce its reliance on Nvidia hardware.

However, the company has previously sought to diversify its compute resources by deploying Google TPUs and AMD GPUs, and said it would provide Microsoft with feedback on its own custom AI chip project, dubbed Maia.