Broadcom AI revenue hits $8.2 billion but backlog miss pressures stock

Broadcom's latest earnings report had two stories inside it, and Wall Street chose to focus on the smaller one. AI-related revenue nearly doubled year-over-year to $8.2 billion, driven by demand for the company's custom silicon and high-speed networking products in AI data centers. That is a strong number by any ordinary measure. Yet the stock slid after Broadcom reported a $73 billion AI product order backlog, a figure that came in slightly below what analysts had penciled in.

The reaction is a useful reminder of how semiconductor stocks get priced right now. The baseline expectation for AI infrastructure companies has moved so far ahead of historical norms that beating last year's revenue is no longer enough. The market is pricing in continued acceleration, and anything that reads as a deceleration signal, even a modest one, gets punished quickly.

What is driving Broadcom's AI revenue

Broadcom occupies a specific and defensible position in the AI hardware supply chain. Unlike Nvidia, which sells general-purpose GPUs, Broadcom focuses on custom application-specific integrated circuits, or ASICs, designed for particular customers and workloads. Google's tensor processing units are built with Broadcom's help. Meta has also been a customer for custom AI chips. These are not commodity relationships. Replacing Broadcom in those design cycles takes years and carries significant engineering risk for the customer.

The networking side of the business is the other piece. As AI clusters scale to tens of thousands of chips, the interconnect fabric between those chips becomes a performance bottleneck. Broadcom's Ethernet and InfiniBand networking products are used extensively in hyperscale data centers. Revenue from that segment has grown alongside GPU deployments, since every large cluster needs the switching infrastructure to go with it.

Why the $73 billion backlog number disappointed

Order backlogs are a forward-looking metric. They tell you how much revenue a company has already committed to deliver over the coming quarters. When Broadcom reported $73 billion in AI product backlog, the number sounds large in absolute terms. The issue was context. Analysts tracking Broadcom had built models projecting a higher figure, based on the pace of hyperscaler capital expenditure announcements over the previous two quarters. A miss against those internal models, even a small one, signals that the order intake rate may be slowing.

It is worth separating two different concerns here. One is whether Broadcom's actual business is deteriorating. The revenue numbers say it is not. The other is whether the rate of growth is sustainable at the level the stock price already assumes. That second question is harder to answer, and the backlog figure gave bears something concrete to point to.

Custom silicon demand and the hyperscaler connection

The longer story inside Broadcom's results is the ongoing shift toward custom silicon among the largest cloud providers. Google, Meta, Amazon, and Microsoft have all been investing in proprietary chip designs to reduce dependency on third-party GPU suppliers and lower their cost per AI compute operation. Broadcom is a primary manufacturing and design partner for several of these programs.

This creates a revenue structure for Broadcom that is different from a company selling chips on the open market. Revenue is more predictable because it is tied to specific customer programs with multi-year roadmaps. The tradeoff is concentration risk. If Google, for example, shifted a major program to a different chip architecture or brought more design work in-house, the impact on Broadcom's revenue would be immediate and visible.

How the stock pressure fits into the broader semiconductor picture

Broadcom is not the only semiconductor company dealing with this dynamic. Nvidia faced a similar pattern earlier in 2025, where strong earnings still produced short-term stock pressure because the results did not exceed increasingly stretched analyst estimates. The AI infrastructure buildout is real and ongoing, but the stocks of companies exposed to it have in many cases already priced in several more years of high growth. That leaves little room for even minor disappointments.

Broadcom's next earnings report is expected in the summer of 2026 and will cover the period when several hyperscaler customers are scheduled to ramp new AI cluster deployments. The order backlog figure at that point will be closely watched to see whether the Q1 2026 miss was a timing issue or the beginning of a more sustained slowdown in custom chip orders.