There is a specific kind of trap a company falls into when it is winning too visibly. Broadcom is walking into it Wednesday.
The semiconductor and infrastructure software company reports fiscal third-quarter results on September 2, and the setup is unusual: management's own long-term target has been publicly overtaken by the Street's estimate of it. Broadcom has guided to more than $100 billion of AI semiconductor revenue in fiscal 2027. Consensus is sitting near $120 billion.
That gap is the entire trade. Beating your own guidance is not the bar anymore. Beating the number analysts invented on top of your guidance is.
The numbers going into Wednesday
Q3 consensus EPS: $3.21, which would be roughly 90% year-over-year growth.
Q3 consensus revenue: about $29.25 billion, up roughly 83% year over year.
Management guided Q3 revenue to approximately $29.4 billion, implying about 84% growth.
Management guided Q3 AI semiconductor revenue to roughly $16 billion — more than double the prior-year figure.
Fiscal Q2 was a record: $22.2 billion in revenue, up 48%, with AI semiconductor revenue up 143% to $10.8 billion.
Full-year fiscal 2026 AI semiconductor revenue is guided to approximately $56 billion, around 180% above fiscal 2025.
Those are extraordinary growth rates by any historical standard for a company this size. They have also not been enough. Broadcom stock has declined about 13% over the past three months. In the prior quarter, revenue came in slightly below the Street estimate despite the surge in AI sales, and the $16 billion Q3 AI forecast landed fractionally below what analysts wanted. The shares fell sharply.
A company can grow AI revenue 143% and still disappoint. That is what a crowded expectation looks like.
The competitive question got louder this month
Two things happened in August that changed the conversation around Broadcom's custom-silicon franchise.
First, Marvell Technology disclosed a custom-chip relationship with Google involving warrants worth up to $12.2 billion in stock — a deal analysts sized at up to $120 billion in potential value. Broadcom shares fell 4.6% on the news. Google's TPU program has been one of the most important pillars of Broadcom's custom accelerator business, and any suggestion that the work is being second-sourced hits directly at the durability of that revenue.
Second, Marvell reported Thursday night and gave the market a live demonstration of how these prints get judged. Net revenue was a record $2.739 billion, up 37% year over year and $39 million above the guidance midpoint. Data center revenue growth accelerated to 46%. Non-GAAP EPS was $0.94 against $0.93 expected. The company raised its fiscal 2028 outlook to roughly $18 billion in revenue, about 50% growth, up from a prior $16.5 billion view.
Marvell fell 10.28% Friday, closing at $216.62.
Goldman Sachs analysts wrote that "investor expectations were elevated heading into the quarter based on robust spending at key customers, as well as the previously disclosed Google relationship," and called the results an "incremental positive" while remaining neutral on valuation. Marvell is still up roughly 184% this year.
That is the template Broadcom inherits Wednesday: a beat, a raise, and a selloff — because the raise was not large enough to justify the multiple the market had already assigned. It is the same skepticism now attached to backlog-driven AI stories, including Oracle's $638 billion backlog and the lenders backing away from it.
What the bulls are pointing to
Bank of America's Vivek Arya reiterated a Buy on Broadcom with a $530 price target, arguing the company benefits from strong near-term demand for Google's TPUs and expecting management to reiterate its custom-silicon trajectory. Arya has pointed to new ramps from Meta Platforms and OpenAI — each roughly one gigawatt of capacity representing more than $10 billion — beginning in calendar 2027, on top of the TPU business.
The bull case is straightforward: customer concentration is a risk right up until the customer list gets longer, and Broadcom's is getting longer at the exact moment investors are worried about it.
The bear case is equally straightforward: at 83% revenue growth and 90% EPS growth, there is no room in the price for a single quarter of merely good news.
The macro is not helping
Wednesday's print lands in the middle of the heaviest data week since the summer began. Dallas Fed manufacturing on Monday, August 31. ISM manufacturing and JOLTS job openings on Tuesday, September 1. ADP payrolls and the Federal Reserve's Beige Book on Wednesday, September 2 — the same day Broadcom reports. ISM services Thursday, September 3. And the August employment report Friday, September 4, the last major labor reading before the September 16 FOMC.
Forecasts for that jobs number cluster tightly: Wolfe Research looks for 65,000 payrolls against a 55,000 consensus, with unemployment rising to about 4.16% from 4.09% and average hourly earnings up 0.35% month over month versus 0.2% expected. Continuum Economics models 75,000 payrolls and a 4.2% unemployment rate.
Overlay that with Fed Chair Kevin Warsh's first Jackson Hole keynote on Friday, in which he reiterated that the 2% target is not negotiable, noted the 12-month PCE index at 3.7% and the six-month change at 4.1%, and said the Fed has "more work to do" absent clear progress. Traders added to September rate-hike bets. Stocks closed lower. The S&P 500 finished at 7,711.76, down 0.25%. The Dow Jones Industrial Average ended essentially flat at 53,559.99, off 0.02%. The Nasdaq Composite fell 0.52% to 26,402.42. The VIX slipped 0.55% to 14.43, while the 10-year Treasury yield rose 5.8 basis points to 4.73% — the bond market moved more than stocks did. Broadcom itself closed Friday at $368.79, down 0.74%. Higher discount rates are not what a 90%-growth semiconductor multiple wants to hear — a lesson Salesforce learned while growing its AI business 200% and still finishing last in the Dow.
What lands next
Three questions decide the reaction Wednesday afternoon.
Does Broadcom raise the fiscal 2027 AI revenue target off "more than $100 billion" toward the $120 billion the Street already models?
Does management quantify the Meta and OpenAI ramps, or leave them qualitative for another quarter?
Does anything in the commentary address whether Google's custom silicon work is being split?
A beat on the September quarter answers none of those. That is the difficulty with a stock that has already been told it is winning.
This article is for informational and educational purposes only and does not constitute investment advice or a recommendation to buy or sell any security.
Found this helpful? Share it with others.