OpenAI, Broadcom cooperate to develop AI chip focused on inference

OpenAI is partnering with Broadcom Inc. to develop a new artificial intelligence chip, with a particular focus on running AI models after they are trained, according to two people familiar with the matter.

The AI ​​chipmaker and startup are also advising Taiwan Semiconductor Manufacturing Co., the world’s largest chip contract manufacturer, said the people, who requested anonymity because the discussions are private to the company. know. OpenAI has been planning a custom chip and researching such applications for the technology for about a year, but discussions are still in the early stages.

OpenAI declined to comment. Representatives of Broadcom and TSMC have not yet responded. Reuters reported on OpenAI’s ongoing talks with Broadcom and TSMC on Tuesday. The Information reported in June that Broadcom had discussed creating an AI chip for OpenAI.

The process of bringing a chip from design to manufacturing is time-consuming and expensive. OpenAI is less focused on graphics processing units, the chips used to train and build general AI models — a market that has been cornered by Nvidia Corp. Instead, OpenAI is looking for a specialized chip that will run the software and respond to user requests, a process called inference. Investors and analysts expect demand for chips that support inference to only increase as more tech companies use AI models to perform more complex tasks.

OpenAI may continue to research establishing its own network of foundries or chip factories, one person said, but the startup has realized that working with partners on custom chips is the key. The road is faster and more feasible than today. Reuters previously reported that OpenAI is backing away from efforts to establish its own chip manufacturing capacity.

Broadcom shares closed up 4.2% to $179.24 in New York trading Tuesday. Even before the latest increase, they were up 54% this year. TSMC’s US-traded shares closed 1% higher.

Broadcom is the largest designer of application-specific integrated circuits – chips designed to fit a single purpose specified by the customer. The company’s largest customer in this field is Alphabet Inc. Google. Broadcom also partnered with Meta Platforms Inc. and TikTok owner ByteDance Ltd.

Last month, when asked if he had new customers for the business, given the huge demand for AI training, Broadcom CEO Hock Tan said he would only add to his short list of customers when the The project achieved a large number.

“This is not an easy product to deploy to any customer, and so we don’t consider proof of concepts to be production volumes,” he said on the earnings conference call.

OpenAI services require massive computing power to develop and run – much of which comes from Nvidia chips. To meet demand, the industry is working to find alternatives to Nvidia. That includes using processors from Advanced Micro Devices Inc. and develop internal versions.

OpenAI is also actively planning investments and partnerships in data centers, the ultimate home of such AI chips. The startup’s leadership has convinced the US government of the need for larger data centers, and CEO Sam Altman has called on global investors, including some in the Middle East, to fund this effort.

“It’s definitely been a long road,” OpenAI Chief Financial Officer Sarah Friar told Bloomberg Television on Monday. “Stretched from a capital perspective but also my own learning. Honestly, we are all learning in this space: Infrastructure is destiny.”

One more thing! We are now on WhatsApp Channel! Follow us there so you never miss any updates from the world of technology. ‎To follow HT Tech channel on WhatsApp, click This to join now!