Meta And Broadcom’s AI Chip Power Play Could Change Everything

Meta is doubling down on its artificial intelligence (AI) ambitions, expanding its partnership with Broadcom to co-develop custom silicon that could redefine how its platforms deliver real-time AI experiences at global scale.
 

TL;DR

 

• Meta and Broadcom will co-develop multiple generations of MTIA AI chips
• Initial deployment exceeds 1 gigawatt, scaling to multi-gigawatts over time
• Partnership leverages Broadcom’s XPU platform and Ethernet networking
• Focus includes generative AI, recommendations, and real-time AI experiences
• Hock Tan transitions from Meta’s board to an advisory role

 

Meta And Broadcom Deepen AI Silicon Partnership To Power Future Infrastructure

Meta announced an expanded, multi-year collaboration with Broadcom to co-develop multiple generations of its Meta Training and Inference Accelerator (MTIA) chips, signaling a major push toward building a dedicated AI compute backbone.

The partnership focuses on scaling custom silicon capabilities to meet growing demands from generative AI (GenAI), ranking systems, and real-time recommendations across Meta’s ecosystem. This includes platforms such as WhatsApp, Instagram, and Threads, all of which are expected to benefit from faster and more efficient AI processing.

At the core of this effort is Meta’s MTIA portfolio, purpose-built chips designed to optimize inference workloads at scale while maintaining a balance between performance and total cost of ownership.

“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,” said Meta founder and CEO Mark Zuckerberg. “As we roll out more than 1GW of our custom silicon to start and then multiple gigawatts over time, this partnership will give us greater performance and efficiency for everything we’re building.”

“We are pleased to expand our strategic collaboration with Meta as they pioneer the next frontier of artificial intelligence,” said Hock Tan, President and CEO of Broadcom. “This initial MTIA deployment is just the beginning of a sustained, multi-generation roadmap to serve the trajectory of massive growth over the next few years that highlights Broadcom’s unmatched leadership in AI networking and the power of our foundational XPU custom accelerator platform.”

As part of the expanded agreement, and considering the scale of the deal, Hock Tan will step down from Meta’s Board of Directors and transition into an advisory role. In this capacity, he will guide Meta’s custom silicon roadmap and contribute to future infrastructure strategies.

 

Scaling AI Infrastructure To Multi-Gigawatt Levels With Broadcom’s XPU Platform

The agreement begins with an infrastructure commitment exceeding 1 gigawatt, marking the first phase of a broader multi-gigawatt rollout. This reflects the scale at which Meta is preparing to operate its AI systems, aiming to support billions of users globally.

The collaboration is built on Broadcom’s XPU platform, which enables the design of custom AI accelerators tailored to Meta’s evolving infrastructure needs. This approach allows tight integration across compute, memory, and high-speed input and output systems, creating a flexible blueprint for future chip generations.

Meta plans to roll out four new generations of MTIA chips within the next two years, with development extending through 2029. These chips will support both current AI workloads and future applications tied to increasingly complex models.

Broadcom will also contribute advanced packaging and networking technologies, ensuring that Meta’s AI clusters can scale without bottlenecks. Its Ethernet-based solutions are designed to deliver high bandwidth, low latency connectivity across tens of thousands of nodes, a critical requirement for large-scale AI deployments.
   

Networking Innovation Eliminates AI Bottlenecks At Scale

A major component of the partnership lies in networking. Broadcom’s high-radix Ethernet switches, optical connectivity, PCIe systems, and high-speed SerDes technologies will form the backbone of Meta’s AI infrastructure.

These systems are designed to handle scale-up, scale-out, and scale-across requirements, ensuring seamless communication within and between AI clusters. This reduces congestion during intensive workloads and maintains near-zero latency, which is essential for inference-heavy applications.

The infrastructure also supports evolving memory architectures and helps lower total cost of ownership over time, aligning with Meta’s long-term efficiency goals.

The move comes months after Meta signed a $100 billion deal with chipmaker AMD, and under a month after Arm launched its first in-house AI CPU, where Meta acted as a lead partner.