Cutting corners: OpenAI is turning to Broadcom for help building custom AI chips, a move that underscores how demand for computing power is reshaping the semiconductor industry and forcing developers to diversify beyond Nvidia. With competition for GPUs intensifying, the partnership positions both companies to address shortages that have slowed the rollout of new AI systems.
Broadcom disclosed during its quarterly earnings call on Thursday that a fourth major AI developer had signed on as a customer, placing a one-time order worth $10 billion for AI server racks powered by its chips. People familiar with the matter told The Wall Street Journal that the buyer is OpenAI.
Investors welcomed the news – not only for the size of the order, but also for what it signals about the company's growing role in supplying the infrastructure behind the AI boom.
Broadcom, formerly best known for designing chips used in smartphones, has benefited enormously from the rapid adoption of artificial intelligence. The company has moved aggressively into data center technology, and since early 2023, its shares have risen sixfold.
OpenAI's decision to pursue custom hardware reflects a broader struggle in the AI industry to secure sufficient processing capacity. For years, the company leaned heavily on Nvidia's GPUs. But supply has tightened as more developers compete for Nvidia's hardware.
"If we're talking about hyperscalers and gigantic AI factories, it's very hard to get access to a high number of GPUs," Nikolay Filichkin, co-founder of Compute Labs, said. "It requires months of lead time and planning with the manufacturers."
Sam Altman, OpenAI's chief executive, has repeatedly warned that limited access to GPUs was slowing his company's progress. Earlier this year, he wrote on X that ChatGPT-4.5, which he described as the company's closest attempt at designing a model with human-like reasoning, had faced rollout delays due to hardware shortages.
GPT-4.5 is ready!
– Sam Altman (@sama) February 27, 2025
good news: it is the first model that feels like talking to a thoughtful person to me. I have had several moments where I've sat back in my chair and been astonished at getting actually good advice from an AI.
bad news: it is a giant, expensive model. we...
Altman explained at the time that OpenAI planned to add tens of thousands of GPUs in the near term and eventually hundreds of thousands, noting, "I'm pretty sure y'all will use every one we can rack up."
To address those shortages, OpenAI has been working with Broadcom for over a year on a custom semiconductor designed for training large-scale models. Broadcom refers to its products as XPUs, which are chips designed for specific applications, such as running AI systems.
While the project won't rival Nvidia's dominant GPUs, a person close to OpenAI said the tailored chip is intended to cover gaps in the company's hardware pool.
Broadcom's chief executive, Hock Tan, highlighted the importance of the order during Thursday's earnings call. "The addition of a fourth customer with immediate and fairly substantial demand really changes our thinking of what 2026 would be starting to look like," he said. The $10 billion deal is expected to begin contributing to Broadcom's revenue during the summer quarter of 2025.
Chips are only part of OpenAI's infrastructure push. The company is in the middle of multibillion-dollar commitments to expand its computing backbone. Earlier this year, it signed a contract with Oracle worth more than $30 billion annually for cloud hosting capacity. It also has a smaller agreement with Google to supplement its computing needs and is attempting to develop its own massive data-center project, named Stargate. However, that initiative has been slow to advance.
Broadcom's rise in the AI race reflects how quickly shifts in demand can reorder the global technology landscape. In August, Broadcom's market capitalization surpassed that of Saudi Aramco, placing it among the world's seven largest publicly traded companies. Also last month, it introduced the Jericho networking chip, designed to improve AI efficiency by allowing data centers up to 60 miles apart to function as a single system.
Broadcom to build custom chips for OpenAI in $10 billion deal
