Broadcom’s Facebook friend will help train it to accelerate AI workloads

A partnership between Broadcom and Meta could benefit other datacenter operators, as the networking company hones its expertise for supporting AI development. Meta is developing its own in-house AI chips for its exclusive use, and has extended an agreement with Broadcom to support that infrastructure, an agreement that will also provide Broadcom with the know-how to ramp up its expertise in AI networking. The two companies will couple logic, memory, and high-speed I/O for current deployment, and establish a blueprint for future iterations of the Meta Training and Inference Accelerator (MTIA) chips. As part of the agreement, Broadcom is deploying its Ethernet-based backbone to scale up compute within MTIA racks. Its enhanced network capacity is designed to eliminate network congestion, even under the most intensive AI workloads. “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 highlighting Broadcom’s unmatched leadership in AI networking and the power of our foundational XPU custom accelerator platform,” Broadcom CEO Hock Tan said, according to a company news release. Broadcom has been busy polishing its AI credentials. Just last week, it signed agreements with Anthropic to develop a new generation of AI chips, and with Google to boost its infrastructure for AI development.