Openai reportedly plans to manufacture its first customized artificial intelligence (AI) chipset this year. According to the report, the San Francisco-based AI company has begun the design process internally and is expected to complete the processor design in the next few months. It is said that the main reason for the company to manufacture custom AI chipsets is to reduce reliance on NVIDIA and strengthen negotiations with other chip suppliers. It is worth noting that OpenAI’s recent trademark filings show that the company plans to manufacture a variety of hardware, including chipsets.
Openai’s chipset
According to a Reuters report, OpenAI is currently finalizing the design of the internal chipset and is expected to complete the process in the next few months. The publication cited sources familiar with the matter that AI companies reportedly would tape (the process of sending the first design through a chip factory) from Taiwan Semiconductor Manufacturing Corporation (TSMC).
TSMC is reportedly going to handle the manufacturing process of OpenAI. The chipset is said to be a 3nm process technology with a shrink array architecture with high bandwidth memory (HBM) and extensive network capabilities. It is worth noting that HBM-based designs are also used in NVIDIA chipsets.
Openai reportedly believes that building its own chipset will give other chipset suppliers a strategic advantage in negotiating deals. This is said to have also reduced the company’s reliance on NVIDIA, which has widely used its chips. The publication notes that AI companies plan to use future iterations of chipsets to develop “increasingly advanced processors with wider capabilities.”
Reuters quoted sources asserting that the chipset was designed by the AI company’s internal team, Richard Ho, who heads the hardware at OpenAI. Interestingly, Ho has worked at LightMatter and Google before and specializes in semiconductor engineering. The team under HO has reportedly doubled in the past few months and now includes 40 employees.
It is worth noting that the report claims that OpenAI’s first chipset will initially be used on a limited scale, with the main function being to run some of the company’s AI models. It is said that there is a limited role in the company’s infrastructure at present, but it may increase in the future. Ultimately, AI companies intend to use chips for reasoning and training AI models.
