Broadcom Q3 2026 Earnings Preview: AI Chips in Focus

Broadcom closes out the season’s marquee chip earnings when it reports fiscal third-quarter 2026 results after the bell on Tuesday, September 2. Coming a week after Nvidia and Marvell, Broadcom’s print is the last big read on whether the AI hardware boom is still accelerating — and the number that matters most is not total revenue but the single line labeled AI semiconductors, which management has guided to roughly $16 billion for the quarter.

The stock has been one of 2026’s biggest winners, and expectations are stacked accordingly. A beat-and-raise is close to priced in; the market’s reaction will turn on the forward guidance and on whether Broadcom’s custom-silicon backlog is still growing faster than the Street can model.

The headline numbers

Wall Street consensus heading into the report:

• Total revenue of about $29.4 billion, a sharp year-over-year gain driven by AI.
• Adjusted EPS near $2.55.
• Semiconductor revenue of roughly $20.5 billion, up about 124% year over year.
• Infrastructure software of about $8.9 billion, up roughly 31%, the recurring, higher-margin base from Broadcom’s VMware franchise.

But the figure the entire call will orbit is AI semiconductor revenue, guided to about $16 billion — growth of more than 200% year over year, and up from $10.8 billion (itself up 143%) in the prior quarter. That single line has become the company’s center of gravity, and small changes in its trajectory move the stock more than the headline total does.

Profitability is a secondary watch item. Management has signaled Q3 gross margin stepping down to around 74% on product mix — custom accelerators carry lower margins than Broadcom’s legacy chip and software businesses — with operating margin holding near 67%. A steeper-than-expected margin dip would be the kind of detail that complicates an otherwise clean AI story.

Why AI semis are the whole story

Broadcom’s AI revenue comes from two engines. The first is custom accelerators — the application-specific chips it co-designs with a handful of hyperscalers to run their own AI workloads, an alternative to buying Nvidia GPUs off the shelf. The second is AI networking, the switches and interconnect silicon that stitch tens of thousands of those chips into a single training cluster.

Together they have made Broadcom the clearest pure-play beneficiary of the hyperscaler capital-spending boom outside of Nvidia. When Google, Meta, and other cloud giants pour tens of billions into AI infrastructure, a large slice of that flows to Broadcom whether the customer is designing its own chip or simply wiring GPUs together at scale.

The forward guidance is where the real drama sits. Broadcom has maintained a full-year FY26 AI semiconductor target of about $56 billion, and management has repeatedly pointed beyond it, telling investors it expects AI semiconductor revenue to eventually exceed $100 billion and reiterating plans for roughly 10 gigawatts of custom-accelerator shipments in 2027, weighted toward the second half. Any adjustment to those markers — up or down — will matter more to the stock than the quarter just closed.

The custom-silicon race

Broadcom’s ascent has reframed a debate that used to have one answer. For years, “AI chip” meant “Nvidia GPU.” Now the biggest buyers increasingly want their own silicon, and Broadcom is the merchant partner most of them turn to — a dynamic that has turned custom accelerators into the industry’s most-watched battleground.

The competitive picture sharpened all summer. Marvell, Broadcom’s closest rival in custom silicon, struck a multibillion-dollar chip warrant tied to Google that deepened its hyperscaler ties, and its latest results showed data-center revenue dominating the mix. Nvidia, meanwhile, remains the volume leader by a wide margin, and its most recent quarter reset the bar for how fast this market can compound. Broadcom’s job on Tuesday is to show it is still taking share of the custom-accelerator opportunity rather than merely riding the tide.

New entrants complicate the story further. Broadcom has been expanding its roster of custom-silicon customers, and its work on inference-optimized parts — including the OpenAI-tied “Jalapeno” inference chip — points to a pipeline that reaches beyond the two or three anchor accounts that built the franchise. Investors will listen closely for any signal about how many customers are now in volume production versus still in design, because each additional anchor account materially changes the multi-year revenue math.

The financing question

There is a shadow over the AI-hardware trade that Broadcom cannot fully escape: how the buildout is being paid for. The hyperscalers funding these orders are running historic capital-spending deficits, and the AI labs behind much of the demand are financing compute with mountains of debt. Broadcom itself sits close to that leverage — it is tied to a reported $60 billion AI-related debt arrangement involving Anthropic — which links its order book to the same credit conditions rattling the rest of the market.

That matters this week specifically. The report lands during a stretch of rising bond yields and a jittery macro backdrop, with chip stocks already selling off as a global bond repricing lifts discount rates on exactly the long-duration growth names that have led the rally. Broadcom’s guidance will be read not just as a demand signal but as a referendum on whether the debt-financed AI capex cycle can keep expanding at the pace the market has come to expect. For the full macro setup into the print, see our stock market week-ahead.

What it means

Broadcom’s quarter is a high-stakes confirmation vote on the AI hardware thesis. The company has become a proxy for the durability of custom-silicon demand, and after a run this large, the report has to do more than beat — it has to extend the runway.

The bull case is straightforward: AI semiconductor revenue tracking toward and past the $56 billion annual mark, a raised or reaffirmed path toward $100 billion-plus, growing customer count, and networking strength alongside custom accelerators. That combination validates the argument that Broadcom is a second structural winner of the AI era, not a momentum trade.

The bear case is about deceleration and dependence. If AI-semi guidance merely meets rather than exceeds, if margins compress faster than expected, or if management sounds cautious about the pace of 2027 shipments, the market’s reaction could be sharp given how much good news the stock already reflects. And the concentration risk is real: a franchise anchored to a small number of hyperscalers is exposed to any single customer slowing its spend.

What to watch next. Ignore the headline beat and read three things: the AI-semiconductor guidance for next quarter and any change to the full-year and multi-year targets; gross-margin trajectory as custom accelerators grow into a larger share of the mix; and any new-customer disclosure that widens the base beyond today’s anchor accounts. In a market pricing chip stocks for perfection, the gap between “great quarter” and “great quarter, guidance merely in line” is the whole trade.