Semiconductor giant Broadcom (AVGO) saw its stock price rise on Wednesday following the announcement of an agreement with social media giant Meta Platforms (META) to provide artificial intelligence technology.
Broadcom and Meta will join forces to build the infrastructure that will bring advanced generative AI capabilities to Meta’s 3.6 billion users. The two companies will collaboratively design custom AI accelerator chips to support Meta’s massive AI model training and inference workloads. These chips, built on Broadcom’s XPU platform, are designed to strike a balance between performance and total cost of ownership through custom design—offering cost advantages over non-specialized chips. Broadcom’s optical interconnect products and other networking solutions will also provide efficient, low-latency performance for AI workloads.
Meta CEO Mark Zuckerberg stated in a press release: “Meta is partnering with Broadcom on chip design, packaging, and networking to build the massive compute infrastructure we need to deliver personal superintelligence to billions of people.”
By developing custom AI chips in collaboration with Broadcom, Meta is diversifying its supply chain and reducing its reliance on chip manufacturers such as Nvidia and AMD. This will weaken some of these chip suppliers’ pricing power while enabling Meta to optimize chips for its own needs. Broadcom, in turn, is becoming an increasingly important partner for Meta and other cloud computing giants, which will allow Broadcom to capture a larger share of the AI boom in the coming years.
Broadcom and Meta announced a significant expansion of their existing strategic partnership. The two companies will enter into a multi-billion dollar agreement to jointly develop custom AI chips through 2029. Under this multi-year, multi-generational collaboration, Broadcom will assist in designing a family of Meta Training and Inference Accelerator (MTIA) chips, further advancing Meta’s goal of developing AI “superintelligence.” These processors will form the backbone of Meta’s AI computing infrastructure and data centers in the coming years.
Meta initially committed to deploying over one gigawatt of computing power based on Broadcom’s application-specific integrated circuits (ASICs, which Broadcom calls XPUs). Broadcom will also provide a range of data center technologies, including Ethernet switches, optical components, and switches. Together, these products help reduce latency.
Mark Zuckerberg praised the collaboration, stating that Broadcom will help “design chips, packaging, and networking to build the massive compute infrastructure we need.” Meta stated that it will gradually deploy additional computing power measured in “multiple gigawatts.”
This marks Broadcom’s third major deal in just three weeks. Earlier this month, the company signed comprehensive agreements with Alphabet’s Google and with Anthropic to co-develop next-generation custom AI processors. As AI development enters a new phase, market focus has shifted from raw computing power to efficiency—an area where Broadcom excels.
Broadcom’s winning streak continues. During its Q1 earnings call last month, the company stated that it expects AI chip sales to potentially exceed $100 billion by 2027, a figure far surpassing its AI-related sales of $8.4 billion in the previous quarter.
As for Meta, the company has also recently made progress in the AI field. Just last week, Meta launched its latest AI model, Muse Spark, which represents a significant improvement over its Llama AI and which Meta calls its “most powerful model to date.” Meta plans to invest $115 billion to $135 billion in capital expenditures in 2026 to build AI capabilities. Broadcom’s custom chips, with their higher efficiency, will help Meta optimize its spending.
