Broadcom's AI Chip Sales Triple as OpenAI Comes Aboard

Broadcom reported fiscal third-quarter results on September 2 that leave little doubt about where its business now lives. AI semiconductor revenue reached $16.7 billion, up 221% from a year earlier, and for the first time it made up more than half of the company's total sales. Overall revenue came in at $29.6 billion, an 86% rise, which chief executive Hock Tan summed up plainly on the call: "We delivered an exceptional quarter with revenue, operating income and free cash flow all exceeding prior records."

The guidance is where it gets interesting. Broadcom expects AI chip revenue to reach $21.7 billion next quarter, a 236% jump, with total revenue of about $34.8 billion. Looking further out, the company sketched roughly $58 billion in AI revenue for the current fiscal year, then around $115 billion in fiscal 2027 and $230 billion in fiscal 2028. Those are the kind of figures that only hold if the current buildout keeps compounding.

OpenAI joins the custom-chip club

The most concrete new detail was a customer. Broadcom confirmed it shipped Jalapeño, OpenAI's first-generation custom accelerator, during the quarter, with Tan noting the part outperforms some rivals on specific inference workloads. We covered Jalapeño when its first benchmarks appeared last month; Broadcom's results confirm the chip is now moving in volume. OpenAI joins Google, Meta and Anthropic on the roster of large customers designing their own silicon with Broadcom rather than buying every accelerator from Nvidia.

That is the quieter story inside the numbers. The hyperscalers want alternatives to a single supplier, and Broadcom has positioned itself as the partner that helps them build one. It is a different bet from Nvidia's, which cleared $96 billion in its most recent quarter selling finished GPUs to nearly everyone at once.

The limits are physical now

Tan was candid that demand is no longer the constraint. Deployment is. He pointed to data-center readiness, power, land, advanced wafers, substrates and high-bandwidth memory as the things standing between orders and installed capacity. Investors seemed to hear the caution as much as the beat: despite tripling AI profits, the stock barely moved. When a company guides to 236% growth and the market shrugs, the interesting question is what it would take to genuinely surprise anyone anymore.