What happened
Meta extended its custom AI processor deal with Broadcom until 2029, committing to over one gigawatt of computing capacity for its AI infrastructure. This expanded tie-up, announced in a joint statement, includes Broadcom's Ethernet networking technology for Meta's AI computer clusters. Broadcom CEO Hock Tan will transition to an advisory role on Meta's custom chip strategy. Meta's MTIA programme, which already uses the MTIA 300 chip for ranking and recommendation systems, plans three more inference-focused chips through 2027.
Why it matters
Access to custom AI silicon will reduce Meta's reliance on external GPU vendors, directly impacting procurement costs and supply chain stability for its AI initiatives. The commitment of over one gigawatt of computing capacity, coupled with Broadcom's networking technology, provides Meta's platform engineers with a dedicated, scalable hardware foundation for its "personal superintelligence" goals. This strategic shift, also seen with Broadcom's custom chip deals with Google, aims to control the cost and availability of compute for large-scale AI model training and inference.
Subscribe for Weekly Updates
Stay ahead with our weekly AI and tech briefings, delivered every Tuesday.
