OpenAI Partners with Broadcom and TSMC to Develop Custom AI Chips
OpenAI has temporarily shelved its ambitious plan to build a semiconductor foundry, opting instead to partner with Broadcom and TSMC to develop custom AI chips. According to reports, this strategic shift aims to support the company's large language models while reducing reliance on external suppliers like NVIDIA and AMD. OpenAI has been collaborating with Broadcom for several months to design its first AI inference chip, with manufacturing capacity secured through TSMC. The first custom-designed chip is expected to be produced in 2026. This move is driven by the need to manage soaring compute costs, as OpenAI projects a $5 billion loss this year against $3.7 billion in revenue. By diversifying its chip supply and focusing on inference capabilities, which analysts predict will soon surpass training demand, OpenAI seeks to ensure scalability and cost-efficiency. The initiative is led by a specialized team of around 20 engineers, including former Google TPU developers Thomas Norrie and Richard Ho. Additionally, OpenAI continues to procure chips from NVIDIA and AMD via Microsoft’s Azure to meet immediate infrastructure demands amidst rapid growth in the competitive AI landscape.
Wire timeline
OpenAI Partners with Broadcom and TSMC to Develop Custom AI Chips
OpenAI has temporarily shelved its ambitious plan to build a semiconductor foundry, opting instead to partner with Broadcom and TSMC to develop custom AI chips. According to reports, this strategic shift aims to support the company's large language models while reducing reliance on external suppliers like NVIDIA and AMD. OpenAI has been collaborating with Broadcom for several months to design its first AI inference chip, with manufacturing capacity secured through TSMC. The first custom-designed chip is expected to be produced in 2026. This move is driven by the need to manage soaring compute costs, as OpenAI projects a $5 billion loss this year against $3.7 billion in revenue. By diversifying its chip supply and focusing on inference capabilities, which analysts predict will soon surpass training demand, OpenAI seeks to ensure scalability and cost-efficiency. The initiative is led by a specialized team of around 20 engineers, including former Google TPU developers Thomas Norrie and Richard Ho. Additionally, OpenAI continues to procure chips from NVIDIA and AMD via Microsoft’s Azure to meet immediate infrastructure demands amidst rapid growth in the competitive AI landscape.
TechNode