Broadcom Inc. has secured a multi-year agreement to continue supplying Google with custom Tensor Processing Units (TPUs) and high-bandwidth networking hardware, extending their partnership to 2031 and solidifying Broadcom’s role as a key player in the artificial intelligence infrastructure buildout.
"This extended agreement with Google is a testament to our strong engineering collaboration and leadership in core technologies for AI infrastructure," Hock Tan, President and CEO of Broadcom, might say in a statement reflecting the deal's importance.
The deal ensures Broadcom remains the primary designer and supplier of Google's custom AI accelerators, which are crucial for training and running the search giant's large language models and other AI services. While specific financial terms were not disclosed, the long-term nature of the contract provides significant revenue visibility for Broadcom's Application-Specific Integrated Circuit (ASIC) division. The previous generation of this partnership was estimated to be worth over $1 billion annually to Broadcom.
This agreement is a major strategic win for Broadcom, locking in a key hyperscale customer and reinforcing its position in the competitive market for AI silicon. It puts pressure on rivals like Marvell Technology, which also designs custom chips, and Nvidia, whose merchant GPUs dominate the AI training market. For Google, it guarantees a stable supply of custom-designed hardware optimized for its specific software and data center needs, a critical factor in the race to advance AI capabilities against competitors like Microsoft and Amazon. The deal also includes high-performance networking components, such as Tomahawk switches, which are essential for connecting large clusters of TPUs.
This article is for informational purposes only and does not constitute investment advice.
