Broadcom reported its fiscal third-quarter 2026 earnings on September 2, 2026, and the numbers were extraordinary by almost any measure. Total revenue grew 86% year-over-year to $29.6 billion, adjusted earnings per share came in at $3.32, up from $1.69 a year earlier, and free cash flow hit $13.66 billion — all records, and all ahead of Wall Street’s expectations. CEO Hock Tan called it “an exceptional quarter with revenue, operating income and free cash flow all exceeding prior records.”

Despite that, Broadcom’s stock fell on the news. The culprit was a single number: fourth-quarter revenue guidance of $34.8 billion, which came in just below the $35.03 billion analysts had modeled — a gap of roughly 0.7%, small enough to be considered a rounding error in most contexts, but enough to spook a stock that had been trading at a price-to-earnings ratio above 60, a valuation that leaves little room for anything short of perfection.

What’s Actually Driving the Growth

The headline figure behind Broadcom’s results is its AI semiconductor business, which generated $16.7 billion in revenue for the quarter, up 221% year-over-year and 54% from the prior quarter alone — enough to push Broadcom’s Semiconductor Solutions division to 70% of total company revenue, with AI chips alone now accounting for 56% of the company’s overall business. Broadcom’s own guidance suggests the growth isn’t slowing: management projects $21.7 billion in AI semiconductor revenue for the fourth quarter, which would represent 236% year-over-year growth, and full-year fiscal 2026 AI revenue near $58 billion.

Looking further out, Broadcom projected AI revenue of roughly $115 billion for fiscal 2027 and approximately $230 billion for fiscal 2028 — forecasts that describe a business roughly quadrupling in size over two years, built primarily on custom silicon designed for individual large-scale customers rather than general-purpose chips sold off a catalog.

The Chip Strategy Behind the Numbers

Broadcom’s approach centers on ASICs, or application-specific integrated circuits — chips designed and tuned for one company’s specific AI workload, rather than the flexible, general-purpose GPUs Nvidia is best known for. That specialization comes with a tradeoff: ASICs are less adaptable than a GPU, but cheaper and more efficient for the specific task they’re built to handle, which has made them particularly attractive for inference work, the process of actually running a trained AI model to generate responses for end users.

Broadcom already has a long-term agreement to supply Alphabet’s Google with custom AI processors through 2031, and the company partnered with OpenAI to co-develop Jalapeño, OpenAI’s first custom inference chip, which OpenAI presented benchmark results for in August. OpenAI framed the collaboration as a way to “serve more intelligence with greater efficiency,” with the two companies targeting a combined 10 gigawatts of AI computing capacity.

How Broadcom’s Numbers Compare to the Rest of the Industry

Broadcom’s results land amid a broader pattern of explosive AI infrastructure spending across the semiconductor industry. Notably, none of Broadcom’s major customers appear to be abandoning Nvidia in favor of custom chips entirely — OpenAI CEO Sam Altman has publicly said Nvidia makes “the best AI chips in the world,” and Anthropic runs its Claude models across AWS Trainium, Google TPUs, and Nvidia GPUs specifically to avoid depending on any single supplier. Google, meanwhile, struck a separate custom-chip deal with Marvell Technology last month, a reminder that even Broadcom’s largest customers are actively working to avoid over-relying on one chip partner.

What This Means for the AI Infrastructure Race

Broadcom’s results offer some of the clearest evidence yet that demand for AI computing infrastructure hasn’t slowed, even as questions persist across the industry about whether AI spending is outpacing genuine near-term returns. For everyday consumers, this kind of infrastructure buildout mostly shows up indirectly — in the pace at which AI features get added to everyday software, and in the ongoing memory and component shortages that have been pushing up prices on laptops and other devices, since many of the same underlying manufacturing resources are being pulled toward AI data center hardware. Whether Broadcom’s ambitious 2027 and 2028 projections hold up will depend heavily on whether the current pace of AI infrastructure investment across the industry proves durable rather than a temporary spending surge.