Meta has announced an expanded partnership with Broadcom to jointly develop several generations of its next-generation MTIA (Meta Training and Inference Accelerator) chips. These custom silicon chips are crucial for powering AI across all of Meta’s applications and services.
At Meta, a portfolio approach is used for AI silicon, ensuring that the appropriate accelerator is matched to each workload to optimize performance and total cost of ownership. MTIA has been specifically designed for large-scale inference and recommendation tasks, and this collaboration with Broadcom aims to deliver the chips at a scale that matches Meta’s ambitions. Recently, Meta revealed plans to develop and deploy four new generations of MTIA chips over the next two years, which will support ranking, recommendations, and generative AI workloads. This partnership is expected to accelerate the development of these custom silicon chips.
Broadcom will collaborate with Meta across various aspects such as chip design, advanced packaging, and networking to build the extensive computing infrastructure needed to deliver real-time AI experiences to billions of users. The partnership leverages Broadcom’s XPU platform, a technology crafted for creating custom AI accelerators (XPUs), which will help optimize Meta’s AI infrastructure across multiple generations of silicon. Additionally, Broadcom’s advanced Ethernet technologies will facilitate seamless, high-bandwidth networking across Meta’s expanding AI compute clusters.
The agreement includes a commitment that exceeds 1GW, marking the first phase of a sustained, multi-gigawatt rollout. This reinforces a shared roadmap to co-design and scale the necessary hardware to bring personal superintelligence to billions of people worldwide.
Hock Tan, President and CEO of Broadcom, stated, “We are pleased to expand our strategic collaboration with Meta as they pioneer the next frontier of artificial intelligence.” He emphasized that this initial MTIA deployment is just the beginning of a sustained, multi-generational roadmap aimed at supporting the massive growth anticipated in the coming years. This highlights Broadcom’s leadership in AI networking and the capabilities of their XPU custom accelerator platform.
Meta’s founder and CEO, Mark Zuckerberg, noted, “Meta is partnering with Broadcom across chip design, packaging, and networking to build out the massive computing foundation we need to deliver personal superintelligence to billions of people.” As Meta begins the rollout of over 1GW of custom silicon, followed by multiple gigawatts over time, this partnership is expected to enhance performance and efficiency in all of Meta’s projects.
This expanded collaboration with Broadcom and the ongoing development of MTIA are vital to Meta’s broader AI infrastructure strategy. It ensures that Meta has the custom compute foundation necessary to achieve its long-term AI goals on a global scale.
Due to the scale of this partnership, Hock Tan will transition from Meta’s Board of Directors to an advisor role, where he will provide guidance on Meta’s custom silicon roadmap and help shape future infrastructure investments. Tan has been a valuable member of Meta’s board for the past two years, offering his expertise in silicon and systems architecture.
The announcement includes forward-looking statements regarding Meta’s business, which should not be considered as predictions of future events. More information about potential risks and uncertainties is available in Meta’s most recent Form 10-K filed with the Securities and Exchange Commission. Meta is not obligated to update these statements based on new information or future events.
