MW Why Broadcom's stock is falling, even as earnings showed strong AI demand
By Britney Nguyen
AI semiconductor revenue at Broadcom could double this quarter after seeing 74% growth in the latest period
Broadcom reported fiscal fourth-quarter earnings results on Thursday.
An upbeat quarter doesn't seem like enough to send Broadcom's stock higher after a strong run so far this year.
While Broadcom shares $(AVGO)$ initially rose in Thursday's extended session following a beat-and-raise earnings report and signs of accelerating artificial-intelligence momentum, they had turned about 4% lower by the end of the company's earnings call.
Among the high points in the latest report: AI chip revenue rose 74% in the October quarter relative to a year earlier, and the company predicts a doubling of AI semiconductor revenue in the ongoing quarter.
But dynamics elsewhere in the tech sector have shown that "raw AI spending isn't enough to keep traders excited," Jake Behan, Direxion's head of capital markets, said in emailed comments.
Read: Looking for the next Nvidia? This analyst recommends quantum stocks - but patience is required
Despite the momentum for Broadcom's AI chips looking "to be intact," he predicted that Broadcom's stock gains have "come so far and so fast this year that even a solid report could end up seeing profit-taking." Shares have increased 75% so far in 2025.
At this point, "traders don't need much of an excuse to lock in gains," Behan continued.
Broadcom's stock has benefitted from recent enthusiasm around Google's latest AI model. Alphabet $(GOOG)$ $(GOOGL)$ unveiled Gemini 3 in mid-November, which it said outperformed OpenAI's GPT-5.1 and Anthropic's Claude Sonnet 4.5 on certain tasks. Google trains and runs its Gemini models on its Tensor processing units developed through a longstanding partnership with Broadcom, whose shares are up more than 15% in the past month.
Don't miss: OpenAI strikes back in the chatbot race against Google with new ChatGPT model
Google's advances have stoked the growing debate around custom chips known as application-specific integrated circuits, or ASICs, made by companies like Broadcom. ASICs compete against graphics processing units from Nvidia (NVDA), which have powered much of the current AI boom. Investors increasingly are wondering whether Broadcom is poised to eat into more of the market opportunity in AI chips.
On the company's earnings call, Tan said Google's TPUs have not only been used for Gemini, but for AI cloud computing by Apple and AI startups like Cohere and Safe Superintelligence, "and the scale at which we see this happening could be significant."
While Tan created some mystery three months ago by noting that Broadcom had an unnamed fourth customer that had signed on for at least $10 billion worth of custom chips, he revealed Thursday that Anthropic was that customer.
Broadcom sold the latest Ironwood system to the AI startup in the last quarter, he added, and the company has received an additional $11 billion order from Anthropic for delivery in late 2026.
But that doesn't mean all of its major customers are using TPUs, Tan said. There are also types of ASICs that appeal to customers, beyond the ones jointly designed with Google.
Some clients "prefer to control their own destiny by continuing to drive their multiyear journey to create their own custom AI accelerators, or XPUs, as we call them," Tan said. He added that Broadcom acquired a fifth custom-chip customer in the quarter with a $1 billion order that is expected to be delivered late next year.
Tan said on the call that Broadcom's AI networking segment is also seeing strong demand as customers build out data centers to deploy AI chips. The company's current order backlog for AI switches is more than $10 billion, he said, and Broadcom has "also secured record orders" for other AI data-center components.
More from MarketWatch: Investors fear Time just jinxed the AI stock-market rally with its 2025 Person of the Year
Altogether, Broadcom's total orders on hand exceed $73 billion, which is almost half of its consolidated backlog of $162 billion. However, Tan said he expects more bookings to come over the next 18 months as that backlog ships out.
Tan also addressed the recent concerns about "customer tooling," in response to an analyst's question. Customer tooling refers to the idea that customers might take on chip design themselves.
The AI buildout will be "a multiyear journey" he said, "and not too many players doing [large language models] want to do their own custom accelerator for very good reason."
And because chip technology continues to change and improve, Tan said LLM makers have to decide where to put resources to compete in the chip space, "especially when you have to compete at the end of the day against a merchant GPU who are not slowing down in the rate of evolution." There he seemed to be alluding to Nvidia and Advanced Micro Devices $(AMD)$, which make GPUs and have fast-moving roadmaps.
Tan called customer tooling "an overblown hypothesis."
The latest quarter saw overall revenue of $18 billion, representing growth of 28% over the previous year. Analysts had been expecting $17.5 billion, according to the consensus on FactSet. Broadcom reported adjusted earnings of $1.95 per share, which was also above the $1.87 analysts were eyeing.
For the fiscal first quarter, Broadcom is guiding for total revenue of $19.1 billion, which tops the $18.4 billion analysts tracked by FactSet were expecting.
See more: Broadcom's stock is a 'top pick' on Wall Street ahead of its earnings report. Here's why.
-Britney Nguyen
This content was created by MarketWatch, which is operated by Dow Jones & Co. MarketWatch is published independently from Dow Jones Newswires and The Wall Street Journal.
(END) Dow Jones Newswires
December 11, 2025 19:14 ET (00:14 GMT)
Copyright (c) 2025 Dow Jones & Company, Inc.
Comments