Meta will collaborate with chip designer Broadcom to create multiple generations of custom artificial intelligence processors under an extended agreement as the social media leader strives to enhance the computing capacity required to support AI functionalities throughout its applications.

Tuesday's announcement prolongs the partnership until 2029 and features an initial commitment of more than one gigawatt of computing capacity, sufficient to power around 750,000 U.S. households on average.

According to a joint statement from the companies, Broadcom CEO Hock Tan will step down from Meta's board and transition to an advisory position focusing on its custom chip strategy.

As AI fuels a rise in computing needs, major tech firms like Meta, Google, and Amazon are creating their own chips to lessen dependence on Nvidia's expensive processors.

The surge in demand for custom chips has positioned Broadcom as one of the top beneficiaries of generative AI. The firm collaborates with clients to create tailored processors and provides infrastructure software.

Its stocks rose 3.5percent in after-hours trading, while Meta's remained relatively stable.

Also Read: NVIDIA's Rubin Moment at CES: AI Supercomputing Breakthrough

The partnership aims to "develop the immense computing infrastructure required to provide personal superintelligence to billions of individuals," stated Meta CEO Mark Zuckerberg.

Also Read: AI Autonomy Race: How Advanced are Top Countries' AI Strategies?

The firm, which revealed a plan for four new chips last month, stated that the initial capacity agreement with Broadcom was "the first stage of a continual, multi-gigawatt expansion."

Broadcom’s Ethernet networking technology will likewise be utilized to link Meta’s swiftly expanding clusters of AI computers.

Also Read: From Silos to Synergy: Building the Future With Agentic AI

The initial chip from the Meta Training and Inference Accelerator (MTIA) initiative, named the MTIA 300, currently fuels Meta's ranking and recommendation systems, with three additional chips expected by 2027. The subsequent generations are built for inference, the method through which AI models answer user inquiries.

Independently, Meta announced on Tuesday that Tracey Travis, a board member since 2020, would not seek re-election at the upcoming annual shareholder meeting.