Broadcom Emerges as a Strong Competitor to Nvidia in AI Infrastructure

Broadcom is collaborating with major AI hyperscalers to support the development of their own application-specific integrated circuits (ASICs), a strategy that is becoming a serious competitive factor against Nvidia's GPU dominance.

Compared to Nvidia's general-purpose GPUs, Broadcom's custom ASIC designs offer higher performance and lower power consumption for specific applications. The company's AI chip customers include Alphabet, Meta Platforms, OpenAI, and Anthropic.

Broadcom is working with Alphabet on seventh-generation Ironwood processors and plans to support Anthropic's TPU v8i deployments with 5 gigawatts of capacity in 2027 and 10 gigawatts in 2028. Under a collaboration with Meta for MTIA processors running through 2029, a 3-gigawatt deployment is projected by the end of 2028. Additionally, with OpenAI's first processor, codenamed 'Jalapeño,' the goal is a 1.3-gigawatt XPU deployment in 2027 and a total of 5 gigawatts by the end of 2028.

Broadcom's AI semiconductor revenue for the fiscal third quarter, which ended on August 2, reached $16.7 billion. Custom chips accounted for approximately 73% of this revenue, and total AI semiconductor revenue is expected to rise to $115 billion in fiscal year 2027 and $230 billion in fiscal year 2028. The company also continues to capture a share of AI infrastructure through its Ethernet switching and optical connectivity technologies, even when customers choose Nvidia GPUs.