OpenAI is working with Broadcom to develop an AI inference chip

OpenAI is working with Broadcom Inc. is developing a new artificial intelligence chip specifically designed to run AI models after they’ve been trained, according to two people familiar with the matter.

The artificial intelligence startup and chipmaker is also consulting with Taiwan Semiconductor Manufacturing Co., the world’s largest contract chip maker, said the people, who asked not to be identified because the discussions are private. OpenAI plans to build its own chip and has been working on such uses of the technology for about a year now, but discussions are still in the early stages, sources said.

Read also: “Literally cried”: Blinkit user orders a 1 gram gold coin, receives 0.5 grams, complaint window closes

OpenAI declined to comment. A Broadcom representative did not respond to a request for comment, and a TSMC spokesperson said the company does not comment on rumors or speculation. On Tuesday, Reuters reported OpenAI’s ongoing negotiations with Broadcom and TSMC. In June, The Information reported that Broadcom was discussing the possibility of creating an AI chip for OpenAI.

The process of moving a chip from design to production is long and expensive. OpenAI is less focused on GPUs and chips used to train and build generative AI models, a market that Nvidia has cornered. Instead, the company is looking for a specialized chip that will run software and respond to user requests. a process called inference. Investors and analysts expect the need for inference chips to only grow as more tech companies use artificial intelligence models to solve more complex problems.

OpenAI may continue to explore building its own network of foundries or chip factories, one of the people said, but the startup has realized that working with partners to create custom chips is a faster and more achievable route for now. Earlier, Reuters reported that OpenAI was abandoning the creation of its own chip production facilities.

Broadcom shares rose 4.2% to $179.24 in New York trading on Tuesday. Even before the latest increase, they were up 54% this year. US-traded shares of TSMC closed up more than 1%.

Read also: Gautam Singhania vs Lamborghini: In a world of high demand for supercars in India

Broadcom is the largest developer of application-specific integrated circuits—chips designed to perform a single purpose specified by a customer. The company’s largest client in this area is Alphabet Inc.’s Google. Broadcom also partners with Meta Platforms Inc. and TikTok owner ByteDance Ltd.

When Broadcom CEO Hock Tan was asked last month if he had any new clients for the business given the huge demand for artificial intelligence training, he said he would only add to his short list of clients when projects start shipping en masse.

“This product is not easy to implement for any customer, and so we are not looking at proof of concept as a production volume,” he said during the earnings conference call.

Developing and operating OpenAI services requires enormous computing power, much of which comes from Nvidia chips. To meet demand, the industry is scrambling to find alternatives to Nvidia. This includes the use of Advanced Micro Devices Inc. processors. and developing your own versions.

OpenAI is also actively planning investments and partnerships in the data centers that will eventually become home to such artificial intelligence chips. The startup’s leadership convinced the U.S. government to build larger data centers, and CEO Sam Altman approached global investors, including those in the Middle East, to fund the effort.

“It’s definitely a stretch,” OpenAI Chief Financial Officer Sarah Friar told Bloomberg Television on Monday. “Expand in terms of capital as well as my own training. Honestly, we are all learning in this area: infrastructure is destiny.”

Read also: The Longest and Shortest Nonstop Flight Routes in India This Winter Every Passenger Should Know About