Meta extends Broadcom deal to develop AI chips

• Meta extends Broadcom AI chip deal to 2029.
• May reduce reliance on Nvidia, increase control over AI infrastructure.

Large tech firms are designing their own hardware as spending on AI infrastructure increases. Hyperscaler spending on AI infrastructure is expected to increase in 2026, with some estimates placing total investment between $635 billion and $665 billion. The recent extension of the partnership between Meta Platforms and Broadcom is one example of this approach.

Meta has agreed to continue working with Broadcom on custom AI chips through 2029, according to Reuters. The deal covers future chip generations and forms part of Meta’s effort to build its own AI infrastructure. Meta has also indicated that capital expenditure could reach between $115 billion and $135 billion in 2026, reflecting continued investment in AI systems.

The partnership centres on Meta’s in-house chip programme, MTIA (Meta Training and Inference Accelerator). Meta introduced its MTIA chips in 2023 and outlined additional versions earlier this year as part of its silicon roadmap. The chips are designed primarily for inference workloads, including ranking systems and generative AI features.

MTIA is part of a broader mix of accelerators in Meta’s infrastructure, where different chips are deployed for different workloads. Meta uses these chips for internal operations not offering them as part of a public cloud platform.

Meta has begun deploying versions of these chips, with future generations planned under the extended agreement. The first chip, known as MTIA 300, is used for ranking and recommendation systems, with additional versions expected through 2027. Broadcom contributes to chip design and networking components that connect large clusters of servers.

To support these systems, Broadcom is also providing Ethernet-based networking technology designed to connect large clusters of AI servers. This infrastructure lets compute capacity to scale in large numbers of nodes while maintaining low latency. According to Reutersthe initial rollout tied the partnership involves more than one gigawatt of compute capacity. The first phase exceeds one gigawatt and forms part of a broader multi-gigawatt deployment plan.

These networking systems are designed to support large-scale clusters, with infrastructure capable of connecting tens of thousands of nodes. Compute workloads can be distributed in systems while maintaining low latency.

“The initial MTIA deployment is just the beginning of a sustained, multi-generation roadmap,” said Hock Tan, President and CEO of Broadcom. “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,” said Meta founder and CEO Mark Zuckerberg.

The collaboration also includes integration in compute and networking systems. This includes joint design in chip architecture and connectivity to support large-scale deployments. Meta said the approach is intended to improve efficiency in its AI infrastructure. The focus on inference workloads means the infrastructure is designed for low-latency processing and continuous use in deployed systems.

Tan will step down from Meta’s board and take on an advisory role focused on chip strategy, according to Reuters.

Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London. This comprehensive event is part of TechEx and is co-located with other leading technology events, click here for more information.

AI News is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here.