OpenAI is collaborating with Broadcom to develop custom AI chips in a move highlighting the escalating demand for computing power and the industry’s diversification beyond Nvidia. The partnership aims to alleviate shortages that have hindered the rollout of new AI systems, positioning both companies to capitalize on the growing need for processing capabilities.
During Broadcom’s recent quarterly earnings call, the company disclosed a $10 billion order from a fourth major AI developer for AI server racks powered by its chips. Sources familiar with the matter identified OpenAI as the buyer. This announcement was positively received by investors, signaling Broadcom’s increasing significance in providing essential infrastructure for the AI boom.
Broadcom, traditionally known for smartphone chips, has strategically expanded into data center technology, experiencing a sixfold increase in its share value since early 2023 due to the rapid adoption of AI. OpenAI’s decision to invest in custom hardware reflects a broader industry challenge in securing adequate processing capacity, particularly given the increased competition for Nvidia’s GPUs.
Nikolay Filichkin, co-founder of Compute Labs, noted the difficulties faced by hyperscalers in obtaining a sufficient number of GPUs, requiring significant lead times and planning with manufacturers. OpenAI CEO Sam Altman has repeatedly expressed concerns that limited GPU availability is impeding the company’s progress. Earlier this year, Altman mentioned on X (formerly Twitter) that the rollout of ChatGPT-4.5, described as OpenAI’s closest attempt at a model with human-like reasoning, was delayed due to hardware shortages.
Altman stated that OpenAI plans to add tens of thousands of GPUs in the near term, with eventual plans to incorporate hundreds of thousands, anticipating that “y’all will use every one we can rack up.”
To address these shortages, OpenAI has been working with Broadcom for over a year on a custom semiconductor, referred to by Broadcom as XPUs, designed for training large-scale models. These chips are tailored for specific applications, including running AI systems. While this project is not intended to rival Nvidia’s GPUs, a source close to OpenAI indicated that it aims to fill gaps in the company’s hardware resources.
Broadcom CEO Hock Tan emphasized the importance of the $10 billion order during the earnings call, stating that it “really changes our thinking of what 2026 would be starting to look like.” The deal is expected to contribute to Broadcom’s revenue starting in the summer quarter of 2025.
In addition to chip development, OpenAI is making substantial investments to expand its computing infrastructure. The company signed a contract with Oracle earlier this year for over $30 billion annually for cloud hosting capacity. OpenAI also has a smaller agreement with Google to supplement its computing needs and is pursuing the development of its own large-scale data center project, named Stargate, although progress has been slow.
Broadcom’s rise in the AI sector demonstrates how shifts in demand can reshape the technology landscape. In August, Broadcom’s market capitalization surpassed that of Saudi Aramco, making it one of the world’s seven largest publicly traded companies. Furthermore, Broadcom recently introduced the Jericho networking chip, designed to enhance AI efficiency by enabling data centers up to 60 miles apart to operate as a single system.
