Broadcom’s Custom-Chip Bet Challenges Nvidia’s AI Crown

Broadcom’s AI semiconductor revenue reached $16.7 billion in fiscal third-quarter 2026, up 221% from a year earlier. The figure gives fresh weight to an argument made recently by Earn Your Leisure hosts Troy Millings and Rashaad Bilal: the next major AI winner may not be one of the familiar Magnificent Seven.

Millings favored Broadcom’s custom-accelerator business, while Bilal argued that investors should stay with the established leaders, including Nvidia, Taiwan Semiconductor, Micron, Apple, Alphabet and Microsoft. The disagreement captures a widening split in the AI trade. Nvidia still sells the dominant general-purpose platform, but Broadcom is positioned where hyperscalers are trying to cut costs and tailor hardware to their own workloads.

Broadcom expects AI semiconductor revenue of $21.7 billion in its fiscal fourth quarter, a 236% year-over-year increase. Chief Executive Hock Tan has described a path toward roughly $115 billion in AI revenue in fiscal 2027 and $230 billion in fiscal 2028. Those targets depend on large design wins and sustained capital spending, but they show why investors increasingly view Broadcom as more than a networking supplier.

The mechanism is custom silicon. Application-specific chips are less flexible than Nvidia GPUs, making them a poor fit for rapidly changing training workloads. Once an AI model enters production, however, a chip designed around a stable inference workload can improve power efficiency and lower the cost of each query.

OpenAI’s Jalapeño chip illustrates the shift. OpenAI designed the processor with assistance from Broadcom and began testing it for inference, with commercial deployment planned across data centers. OpenAI still relies heavily on Nvidia for training, so this is not a clean break. It is a second architecture growing alongside the first.

Nvidia’s scale remains formidable. Its latest quarter produced $96.22 billion in revenue, including $89 billion from data centers, and management guided the following quarter to about $108 billion. The company also expects major demand from cloud providers, enterprises and industrial customers beyond the largest hyperscalers.

That leaves Broadcom with the more concentrated bet. Its upside is tied to custom-chip adoption and a handful of powerful customers, while Nvidia offers broader exposure and a mature software ecosystem. Broadcom does not need to replace Nvidia outright. It only needs a larger share of the infrastructure built around Nvidia’s chips.