Meta Deepens AI Chip Partnership with Broadcom, Locking in Capacity Through 2029

Meta Platforms Inc. has just renewed and expanded its tech relationship with Broadcom Inc., aiming to crank up the horsepower behind its AI initiatives. The fresh agreement stretches their collaboration all the way until 2029 and commits to delivering over one gigawatt of computing capacity, roughly enough juice to power around 750,000 average American homes.

This deal marks a key step as Meta seeks to scale its AI-driven services across its suite of social apps. It's all about having bespoke processors that can handle complex AI workloads without leaning heavily on pricey off-the-shelf chips, namely those from NVIDIA. Broadcom's involvement doesn't stop at chips - its Ethernet networking solutions will also support Meta's swelling AI compute clusters.

As the AI boom prompts tech giants to craft their specialized silicon, Broadcom emerges as a significant player, designing custom processors and infrastructure software tailored for AI tasks. Their partnership reflects the trend among companies like Meta, Google, and Amazon to reduce dependencies on standard GPU suppliers.

Interestingly, as part of this extended engagement, Broadcom's CEO Hock Tan will step down from Meta's board to serve as an advisor on Meta's custom chip strategy, signaling a closer, though more specialized, working relationship between the two firms.

Meta's CEO Mark Zuckerberg framed the development as a way to "build out the massive computing foundation" necessary to bring advanced AI capabilities to billions of users. The initial phase of this expanded deal is described as the kickoff to a sustained rollout that could see multi-gigawatt capacity improvements in upcoming years, anchoring Meta's AI ambitions firmly in the hardware domain.

The recently launched MTIA 300 chip, part of this broader custom silicon line, currently powers Meta's ranking and recommendation systems, with several next-generation processors slated for release through 2027. These subsequent chips will mainly target inference workloads - basically how AI models respond to what users ask them.

This deal comes at a time when AI's appetite for computing resources is driving an arms race in chip innovation. Meta's push for custom, purpose-built processors reflects the industry's pivot toward reducing reliance on third-party suppliers, aiming instead to fine-tune hardware for cutting-edge AI efficiency and scale.

On the equity front, Broadcom's shares ticked up about 3.5% in after-hours trading after the announcement, while Meta's stock saw little movement, suggesting investors are digesting these long-term strategic investments differently.