Broadcom, not AMD or Intel, is Nvidia's biggest threat

Broadcom's rise as an AI chip contender rests on a different model from Nvidia's. Nvidia sells general-purpose GPUs that handle a wide variety of accelerated computing tasks. Broadcom helps hyperscalers design custom ASICs (application-specific integrated circuits) tuned to specific workloads, potentially offering better performance per dollar and lower energy use for certain jobs. It also supplies the networking technology that connects clusters of chips inside data centers.

The customer roster is what gives this argument weight. Alphabet works with Broadcom on its Tensor Processing Units, including the seventh-generation Ironwood processor, and Broadcom expects Anthropic to deploy 5 gigawatts of eighth-generation TPU v8i capacity in 2027, followed by an additional 10 gigawatts in 2028. Meta is collaborating with Broadcom on its MTIA processors, a partnership extending through 2029 with visibility into 3 gigawatts of deployments through 2028. Last October, Broadcom and OpenAI announced plans to deploy 5 gigawatts of XPUs by the end of 2028, with OpenAI's first processor, code-named Jalapeño, expected to support 1.3 gigawatts of deployments in 2027.

The financial trajectory reinforces the case. After AI semiconductor revenue more than tripled to $16.7 billion in the fiscal third quarter, management guided to $21.7 billion in the fourth quarter, and Broadcom expects roughly $115 billion in fiscal 2027 and potentially $230 billion in fiscal 2028. The Motley Fool notes that as AI adoption shifts toward inference—the process of using trained models to generate responses—specialized processors could become more attractive for repetitive workloads. Broadcom's networking business means it can benefit from expanding AI infrastructure even if customers keep buying Nvidia GPUs, so Nvidia need not lose its dominance for Broadcom to win.