• OpenAI partners with Broadcom on custom AI chip development to reduce Nvidia dependency
  • The $10 billion deal aims to create specialized chips for inference tasks, with production starting in 2026
  • Broadcom shares surged 15% following the announcement, boosting its valuation to approximately $1.7 trillion

OpenAI has entered into a strategic partnership with semiconductor giant Broadcom to develop custom artificial intelligence chips in a deal valued at approximately $10 billion, according to people familiar with the matter. The arrangement represents one of the largest custom chip development agreements in recent years and signals OpenAI's aggressive push to secure specialized hardware for its growing AI compute needs.

The collaboration, which will see Taiwan Semiconductor Manufacturing Company (TSMC) handle manufacturing, is targeting production to begin in 2026. The custom chips are specifically designed for inference tasks—the process of running trained AI models—and are intended for OpenAI's internal use rather than commercial release. This move follows similar initiatives by other tech giants including Google, Amazon, and Microsoft, who have all developed proprietary chips to optimize AI workloads and reduce reliance on dominant supplier Nvidia.

"This partnership represents a significant step in our infrastructure strategy," said a person close to the negotiations, who spoke on condition of anonymity because the details haven't been publicly disclosed. "The goal is to achieve better performance and cost efficiency for our specific AI models while ensuring long-term supply chain stability."

Broadcom's stock responded positively to the news, climbing 15% in recent trading and lifting the company's market valuation to around $1.7 trillion. The deal significantly bolsters Broadcom's position in the rapidly expanding AI chip market, where demand has surged following the widespread adoption of large language models like OpenAI's ChatGPT.

OpenAI's chip development initiative comes amid the company's rapid expansion, including a recent secondary stock offering that valued the AI firm at approximately $500 billion. The company raised $40 billion in a Series F funding round in March, though its recent financial activities have focused more on employee stock offerings rather than raising new capital.

While the custom chip development represents a substantial investment, industry analysts note that OpenAI will likely continue relying on Nvidia for certain critical tasks, particularly training complex AI models. "Nvidia's hardware remains essential for training state-of-the-art AI systems," said one semiconductor analyst who requested anonymity to discuss client relationships. "What we're seeing is a diversification strategy rather than a complete replacement."

OpenAI's broader infrastructure efforts include involvement in the ambitious "Stargate" project, a large-scale AI infrastructure initiative that may involve collaboration with government entities. The company declined to comment on the Broadcom partnership when reached Tuesday, while Broadcom representatives didn't immediately respond to requests for comment.

Correction: An earlier version of this article misstated the timeline for production. The custom chips are expected to begin production in 2026, not 2025.