Equity Markets

Broadcom's AI Chip Revenue Tripled in a Year. Its Guidance Still Wasn't Big Enough for Wall Street.

A quarter that beat every major estimate gave way to a stock decline once investors measured Broadcom's fourth quarter revenue guide against the most bullish numbers on the Street. Broadcom reported fiscal third quarter revenue of $29.6 bil…

Broadcom's AI Chip Revenue Tripled in a Year. Its Guidance Still Wasn't Big Enough for Wall Street.
Broadcom's AI Chip Revenue Tripled in a Year. Its Guidance Still Wasn't Big Enough for Wall Street.

A quarter that beat every major estimate gave way to a stock decline once investors measured Broadcom's fourth quarter revenue guide against the most bullish numbers on the Street.

Broadcom reported fiscal third quarter revenue of $29.6 billion, up 86 percent from a year earlier, and adjusted earnings of $3.32 a share, up 96 percent and above the roughly $3.24 analysts had estimated on average. Revenue tied to AI semiconductors reached $16.7 billion, up 221 percent from a year earlier and up 54 percent from the prior quarter alone. Chief financial officer Amie Thuener, delivering her first quarterly report since succeeding Kirsten Spears in June, told investors the company had achieved “record revenue, operating profit and free cash flow” for the period.

On a GAAP basis, net income more than tripled to $13.1 billion, or $2.68 a share. Adjusted operating income rose 92 percent to $20.1 billion. Broadcom's semiconductor solutions segment, which includes its custom AI chip business, grew 127 percent to $20.8 billion, while its infrastructure software segment, built around the VMware acquisition, grew 29 percent to $8.8 billion. Adjusted gross margin was 74.9 percent of revenue. Free cash flow reached $13.7 billion, or 46 percent of revenue, up 95 percent from a year earlier, on capital expenditures of just $532 million, a reminder that Broadcom's chip design model requires far less capital spending than the data center buildouts its hyperscaler customers are funding with that revenue. The company declared a quarterly dividend of 65 cents a share. Over the first three quarters of its fiscal year, Broadcom has generated $71.1 billion in revenue and $31.9 billion in free cash flow.

Shares fell about 5 percent in after hours trading, giving back some of the stock's run into the print, after closing the regular session at $367.24.

The decline followed fourth quarter guidance of approximately $34.8 billion in revenue, itself a 93 percent increase from a year earlier, with AI semiconductor revenue guided to accelerate further, to $21.7 billion. That outlook fell short of the highest estimates some analysts had circulated heading into the print, even though it represents a substantial raise over Broadcom's already elevated growth rate. Chief executive Hock Tan told analysts on the call that demand for Broadcom's custom AI accelerators remains “significantly more” than the company can currently supply, and that Broadcom has already secured enough manufacturing capacity to double its AI revenue again in fiscal 2027, to approximately $115 billion.

Tan named the companies behind that demand directly. Google is deploying Broadcom's custom processors under a multi year agreement Tan described as worth “multi tens of billions” of dollars annually, with its newest generations already in production. Anthropic is deploying a gigawatt of capacity this year under a project internally called Ironwood, expanding to 5 gigawatts in 2027 and a further 10 gigawatts in 2028. OpenAI has a separate 1.3 gigawatt deployment planned for 2027, growing to more than 5 gigawatts in 2028, and Meta is using Broadcom's custom accelerators for its own AI chips, targeting 3 gigawatts of capacity through 2028.

Analysts on the call pressed Tan on how durable that demand is against Broadcom's ability to meet it. Bernstein's Stacy Rasgon asked about the economics of custom chips relative to merchant graphics processors, and Tan said Broadcom's custom accelerators cost less than half as much per unit, generating $20 billion to $30 billion in revenue for every gigawatt of deployed capacity. Goldman Sachs' Jim Schneider asked what could limit Broadcom's growth from here, and Tan pointed to land, power, and building shell availability as the primary constraints, with memory components a secondary concern.

Broadcom's reliance on a small number of very large customers for the bulk of its AI growth cuts both ways for investors: it magnifies the upside if those hyperscalers keep expanding capacity, and the risk if any one of them slows spending. Management's own answer is that the supply commitments needed to hit the 2027 target are already secured, regardless of how conservative this quarter's guidance looked next to consensus.

Whether Wednesday's stock decline reflects a genuine slowdown in the pace of AI infrastructure spending, or a market that had simply priced in numbers even Broadcom's own outlook could not clear, is the question the next quarter's results, and Broadcom's ability to convert its stated backlog into delivered revenue, should help answer.

More articles from FinancialMarkets.com