OpenAI is collaborating with Broadcom (AVGO, Financial) to develop a new AI chip for model inference, while also engaging in talks with TSMC (TSM), the world's largest chip contract manufacturer. Although OpenAI is planning to customize chips for this purpose, discussions are still in the early stages. Previously, OpenAI had considered building its own chip manufacturing network but has since decided to focus on internal chip design efforts due to cost and time constraints.
This strategic move highlights OpenAI's approach of leveraging industry partnerships to ensure chip supply and manage costs, a method also employed by major competitors like Amazon (AMZN), Meta (META), Google (GOOGL), and Microsoft (MSFT). As one of the largest chip buyers, OpenAI's decision to procure from various manufacturers while developing custom chips could have wider implications for the tech industry.
OpenAI aims to create a specialized chip for software operation and responding to user requests, known as inference, rather than focusing on GPUs used for training AI models. Demand for inference-supporting chips is expected to grow as more tech companies adopt AI models for complex tasks. Broadcom's shares rose by 4.2%, closing at $179.24, while TSMC shares also saw a rise.
OpenAI has been working with Broadcom for months to develop its first inference-focused AI chip. Although there is currently a greater demand for training chips, analysts predict that demand for inference chips will soon surpass it as more AI applications are deployed.
The company's chip team, led by top engineers formerly from Google, is planning to produce its first custom-designed chip by 2026 in collaboration with Broadcom and TSMC, although timelines may change. OpenAI's services require significant computational power, primarily from NVIDIA chips, which currently hold more than 80% of the market share. Due to rising costs and shortages, major clients like Microsoft, Meta, and OpenAI are exploring alternative solutions.
Additionally, OpenAI is planning investments and partnerships in data centers, advocating for more large-scale data centers to the U.S. government, and seeking funding from global investors, including those in the Middle East.
