OpenAI is collaborating with Broadcom Inc. to create a new AI chip that is specifically designed to run AI models after they have been trained.

According to insiders who asked not to be named because the talks are confidential, the AI company and chipmaker is also consulting with Taiwan Semiconductor Manufacturing Co., the biggest chip contract manufacturer in the world. According to the persons, OpenAI has been developing such applications for the technology and designing a bespoke chip for almost a year, but the talks are still in their early phases.

OpenAI chose not to respond. A TSMC spokeswoman stated that the company doesn't comment on rumors and speculation, while a Broadcom representative did not reply to a request for comment. According to reports, OpenAI's ongoing discussions with TSMC and Broadcom on Tuesday. In June, The Information revealed that Broadcom had talked about developing an AI processor for OpenAI.

A chip must go through a drawn-out and costly manufacture process. Nvidia Corp. has dominated the market for graphics processing units, which are chips required to train and create generative AI models. OpenAI is less interested in this area. Rather, it is searching for a specific chip that will execute the software and react to user commands—a process known as inference. Analysts and investors anticipate that as more tech businesses employ AI models to handle increasingly complicated jobs, the need for chips that support inference will only increase.

OpenAI may still look into establishing its own network of foundries, or chip factories, but for the time being, collaborating with partners on custom chips is a more practical and expedient option.

Tuesday's closing price of Broadcom's stock in New York trading was $179.24, up 4.2%. They were up 54% this year even before the most recent increase. US-traded shares of TSMC ended the day more than 1% higher.

The biggest manufacturer of application-specific integrated circuits, or chips made to fulfill a single customer-specified purpose, is Broadcom. Alphabet Inc. Google is the company's largest client in this field.

Additionally, Broadcom collaborates with ByteDance Ltd., the owner of TikTok, and Meta Platforms Inc.

Given the high demand for AI training, Broadcom CEO Hock Tan was asked last month if he has any new clients for the company. He said that he will only add clients to his short list when projects reach volume shipments.

During an earnings conference call, he stated, "We do not consider proof of concepts as production volume because it is not an easy product to deploy for any customer."

The development and operation of OpenAI's services demand enormous amounts of processing power, much of which is provided by Nvidia chips. The industry has been rushing to discover Nvidia substitutes in order to meet the demand. This includes adopting Advanced Micro Devices Inc. chips and creating proprietary variants.

Additionally, OpenAI is actively preparing collaborations and investments in data centers, which will eventually house these AI chips. In order to raise money, the startup's leadership has approached the US government about the need for larger data centers, and CEO Sam Altman has reached out to international investors, including some in the Middle East.

Sarah Friar, the chief financial officer of OpenAI, told Bloomberg Television on Monday that it's "definitely a stretch." Stretch both from a financial standpoint and in light of my own education. To be honest, we are all learning in this area that infrastructure determines fate.