The earnings report had the numbers investors once paid anything for. Broadcom said revenue in its fiscal second quarter reached $22.2 billion, up 48 percent from a year earlier, and that AI-related semiconductor revenue hit $10.8 billion, up 143 percent. The company guided to $29.4 billion of revenue in the current quarter, an 84 percent increase, with AI chip sales expected to double again to $16 billion. The stock fell about 13 percent the next day.
The drop was the market’s verdict on what Broadcom did not say. Chief Executive Hock Tan, who has built Broadcom into the largest seller of custom AI accelerators, reiterated the company’s full-year target of more than $100 billion in AI chip revenue, but he did not raise it. For a stock that had priced in continuous upgrades, the absence of a surprise was the surprise, and the shares paid for it.
The details explain the market’s mood. Broadcom’s total revenue came in slightly below analysts’ estimates, and its infrastructure-software business, which includes VMware, grew just 9 percent and missed expectations. The software miss mattered more than the numbers suggested, because it broke the narrative that Broadcom could grow everywhere at once. Investors who had bought the stock for AI growth had to accept that the rest of the company was growing at a more ordinary pace.
This is the second time this pattern has played out. AI companies report record numbers, the guidance is strong, and the stock falls because the numbers were not strong enough. The pattern reflects a shift in investor psychology that has been building for months: tolerance for AI stories without incremental evidence is eroding, and each quarter raises the bar for what counts as good news.
Broadcom’s position in the AI supply chain is genuinely strong. The company designs custom chips for six core customers, including Google, Meta, OpenAI and Anthropic, and its networking components are in nearly every large AI cluster. Its AI revenue has grown faster than almost any other part of the semiconductor industry, and Tan has said the momentum continues into next year. The question is not whether Broadcom is benefiting from AI; it is whether the benefit is growing fast enough to justify the valuation.
The software business adds a second question. Broadcom bought VMware in 2023 for about $69 billion, a bet that enterprise software would balance the cyclicality of chips. The integration has gone better than critics expected, but software now accounts for a third of revenue and grows slowly, and the company’s attempt to sell AI infrastructure software alongside chips has not yet produced the growth investors were promised.
The market’s reaction also reflects the broader anxiety about AI returns. Every cloud company and chip maker is spending at unprecedented rates, and investors are starting to ask who collects the profit. Broadcom’s earnings are one of the cleanest windows into that question, because its custom-chip business ties directly to the biggest AI spenders. When its numbers disappoint on any margin, the entire trade feels it.
Tan’s guidance for the current quarter, with AI revenue doubling again, is a statement that the demand is real and accelerating. The market’s response, selling the stock anyway, is a statement that the demand was already expected. For the shares to regain their footing, analysts say, Broadcom will need to show not just growth but growth above the trajectory investors have already priced in.
The software business has been a persistent question. Broadcom paid about $69 billion for VMware in 2023, and the acquisition transformed the company’s revenue mix and its margins. VMware’s renewal cycle has generated cash and profit, but its growth has been modest, and the company’s effort to sell AI infrastructure software through the VMware franchise has not yet shown up in the numbers. Investors who hoped software would provide ballast during chip cycles have instead watched it lag.
The custom-chip market is Broadcom’s moat and its exposure. The company designs accelerators for customers who want performance tailored to their workloads, and the six core customers that drive its AI revenue have all signaled they will keep spending. The concentration is the risk: a single customer’s decision to bring design in-house, or to slow its build-out, would show up directly in Broadcom’s numbers, and the market’s sensitivity to that risk has grown.
The bigger question is the cycle itself. Broadcom’s guidance assumes that the hyperscale companies keep spending at unprecedented rates, and that the spending converts into revenue for the industry. Any sign that the build-out is pausing, in the form of a customer’s lower guidance or a delayed data-center opening, would hit Broadcom harder than almost any other company, because its growth is the most concentrated expression of the AI capital-spending boom.
The lesson of the June quarter is uncomfortable for the whole sector: in an AI market this crowded, a company can do everything right and still fall. Broadcom’s executives described the quarter as a record on nearly every line. The market, having heard that story before, wanted a bigger one. Until the numbers come in above the expectations that have been built into AI stocks, the pattern is likely to repeat across the industry, no matter how good the reports look.
