Reuters quoted sources as saying that OpenAI is working with Broadcom and TSMC to build the first in-house chip designed to support its AI systems, with AMD chips added alongside Nvidia chips to meet growing infrastructure demands.
OpenAI, the fast-growing company behind ChatGPT , has been exploring a range of options to diversify chip supplies and lower costs.
OpenAI considered building everything in-house and raising capital for an expensive plan to build a network of factories known as “foundries” to manufacture chips.
The company has dropped its ambitious foundry plans for now because of the costs and time needed to build the network, and instead plans to focus on internal chip design efforts, according to the people, who asked not to be identified because they were not authorized to discuss private matters.
The company's strategy, detailed here for the first time, highlights how the Silicon Valley startup is leveraging industry partnerships and a mix of internal and external approaches to secure chip supplies and manage costs like larger rivals Amazon, Meta, Google and Microsoft.
As one of the largest chip buyers, OpenAI's decision to tap a variety of chipmakers while developing its custom chip could have broader implications for the tech sector.
OpenAI, which helped commercialize generative AI that produces human-like responses to queries, relies on massive computing power to train and operate its systems.
As one of the largest buyers of Nvidia's GPUs, OpenAI uses AI chips to train models where AI learns from data and to make inferences, applying AI to make predictions or decisions based on new information.
Reuters previously reported on OpenAI’s chip design efforts. The information said talks were underway with Broadcom and others.
OpenAI has been working with Broadcom for months to build the first AI chip focused on inference, according to sources.
Demand is now greatest for training chips, but analysts predict that the need for inference chips could outpace them as more AI applications are deployed.
Broadcom helps companies, including Alphabet's Google unit, improve chip designs for manufacturing, and also provides design parts that help move information in and out of chips quickly, which is important in artificial intelligence systems where tens of thousands of chips are linked together to work side by side.
OpenAI is still determining whether to develop or acquire other elements of its own chip design, and may bring on additional partners, two of the sources said.
The company has assembled a chip team of about 20 people, led by senior engineers who previously built Tensor Processing Units (TPUs) at Google, including Thomas Nouri and Richard Hu.
Through Broadcom, OpenAI has secured manufacturing capacity with Taiwan Semiconductor Manufacturing Co. to make its first custom-designed chip in 2026, the sources said. The timeline could change, they said.
Currently, Nvidia GPUs have over 80% market share, but shortages and rising costs have prompted major customers like Microsoft, Meta, and now OpenAI to explore in-house or external alternatives.
OpenAI's planned use of AMD chips through Microsoft's Azure, first reported here, shows how AMD's new MI300X chips are trying to grab a slice of the market dominated by Nvidia.
AMD expects AI chip sales to reach $4.5 billion in 2024, after the chip launches in the fourth quarter of 2023.
Training AI models and operating services like ChatGPT is expensive, and OpenAI has projected a $5 billion loss this year on $3.7 billion in revenue, according to sources.
Computing costs, or the hardware, electricity, and cloud services needed to process large data sets and develop models, are a company's largest expense, driving efforts to improve utilization and diversify suppliers.
OpenAI has been cautious about stealing talent from Nvidia because it wants to maintain a good relationship with the chipmaker, which it remains committed to working with, especially to access its new generation of Blackwell chips, the sources added.