A new report from Reuters ( ) (spotted by Thurrott,) suggests that OpenAI could finalize its first 3nm AI chips in the coming months. The goal is to start mass production at TSMC by 2026.
A team of 40 OpenAI staffers is developing the chip in collaboration with Broadcom. Richard Ho, OpenAI’s new head for hardware, previously worked with Google on infrastructure and cloud services.
Reuters reports that OpenAI’s chip can both train and run AI model, but it will initially be used primarily for inference (running AI model) and to a small extent within the company infrastructure.
The demand for Nvidia’s AI chip remains high, with companies such as OpenAI, Microsoft Meta, and Google investing billions of dollars in AI data centers. But developing AI chips to compete with Nvidia will be a difficult task.
Reuters reports that OpenAI will need to invest $500,000,000 to develop its first version of the 3nm chip. The cost could double as software and peripherals become available. This article was originally published in our sister publication PC for Alla (19459030) and has been translated and localized.
