Nvidia + Broadcom Earnings: Same Substitution, Different Room

Published September 3, 2026

Nvidia and Broadcom are the two biggest sellers of the compute the AI boom runs on. Nvidia builds the GPUs. Broadcom builds much of the custom silicon and networking gear that hyperscalers use to wire those systems together at data center scale, and Wall Street treats it as effectively the number two seat in that market, just behind Nvidia itself. So, when both companies report earnings three trading days apart, in a week that also includes a market moving speech from Fed Chair Kevin Warsh, there is a real question worth asking:

Does the market actually judge these two companies on their own merits, or does reading one report change how it reads the next?

The first report: Nvidia

Nvidia's fiscal second quarter, reported August 26, beat on every major line. Revenue hit $96.2 billion, up 106% from a year earlier and 18% from the prior quarter, against a consensus around $92 billion. Data center revenue, the number that actually drives the stock, reached $89.0 billion, up 117% year over year. Gross margin held at 75.0%. Non-GAAP earnings per share came in at $2.22, ahead of the $2.09 consensus. The company guided next quarter's revenue to $108 billion and pointed to roughly 70% revenue growth for fiscal 2028, well above what analysts had modeled. CEO Jensen Huang described a supply constrained outlook and said he expects supply to remain a bottleneck through the end of fiscal 2028.

The stock did not treat this simply. Nvidia rose 8.7% the next trading day, its best day since April 2025, adding $442 billion in market value. The session after that, it gave back 4.6%. Two days later, on the day Warsh delivered his first Jackson Hole speech and struck a hawkish tone on inflation, Nvidia fell again alongside other chip names. That drop had nothing to do with Nvidia's own results. It was the market repricing risk for every growth stock at once, a layer of information even broader than the sector Nvidia and Broadcom compete in. By the following week the stock had settled into a narrower range, still up from where it started but well off its post earnings peak.

Two companies, one market

The reason Broadcom's report even gets read against Nvidia's is not just proximity on the calendar. It is that the two companies genuinely compete for the same dollars. Nvidia's data center business did $89.0 billion in the quarter. Broadcom's AI semiconductor business, smaller but growing faster, did $16.7 billion, up 221% year over year and 54% from the prior quarter. Cody Acree, an equity analyst at StoneX, put the relationship plainly: across the data center ecosystem, Broadcom is really second only to Nvidia.

The overlap runs deeper than market share tables. Both companies count the same hyperscalers and AI labs as customers. On its earnings call, Broadcom disclosed that Anthropic is on track to become its largest custom chip customer in 2027 and stay there into 2028, the same kind of AI lab whose GPU purchases show up directly in Nvidia's own data center number. Days before Broadcom reported, Nvidia disclosed a new memory technology called NVHBM that pushes directly into the custom AI chip territory Broadcom and Marvell currently lead, a real competitive threat, not a mood.

None of that makes the two stocks the same investment. But it does mean that when Nvidia reports first, its results become the best available evidence about a question that matters to both companies: is demand for AI infrastructure still accelerating. Broadcom hadn't answered that question yet. Nvidia had, and the market used what it had.

Borrowing Nvidia's answer

The day after Nvidia's report, Broadcom shares rose 4.5% with no news of its own. Intel gained 4%, SK Hynix 2%. The next morning, chip stocks opened higher across the board, Nvidia itself up nearly 6% in premarket trading, Marvell up 5.8%, Micron up 4.5%, the two major semiconductor ETFs both up more than 3%. It was the market using Nvidia's numbers to answer a question about the whole sector, before any of the other companies involved had said a word of their own.

The pattern continued into Broadcom's own report date. On September 1, Nvidia jumped roughly 4% on a strong quarter from Dell, whose servers mostly run Nvidia chips. On the morning of September 2, coverage noted Nvidia inching higher specifically because investors were positioning ahead of Broadcom's results that evening, treating Broadcom's number as a read on AI demand that would feed back into Nvidia. JoAnne Feeney, a portfolio manager at Advisors Capital Management and a Broadcom shareholder, described investors going in as still carrying the memory of Nvidia's own uneven post earnings reactions in recent quarters, and its clean positive response on August 26 specifically, as part of what they brought to Broadcom's number. Jay Woods at Freedom Capital Markets framed it as pressure transferring directly from one report to the next: Nvidia had already shown the AI trade was intact, so Broadcom now had to prove its own share of that story was still growing. In both directions, investors were using one earnings report to answer a question about a company that hadn't reported yet.

What Broadcom was walking into

That pressure sat on top of a rough stretch. Broadcom's stock was down about 23% from its June 2 record close, after its prior quarterly report on June 3 triggered a 13% one day drop, the worst post earnings reaction for the stock in at least fifteen years. That report had guided AI semiconductor revenue for the following quarter to $16 billion, short of the more than $17 billion analysts wanted, and repeated rather than raised the company's existing $100 billion AI semiconductor target for fiscal 2027. Coverage going into September framed the September 2 report bluntly, as the one Broadcom needed to get right to stop the slide.

The second report: Broadcom

Broadcom beat, cleanly. Revenue reached $29.6 billion, up 86% year over year, ahead of the roughly $29.5 billion analysts had modeled. Non-GAAP earnings per share came in at $3.32, ahead of a consensus near $3.25. AI semiconductor revenue hit $16.7 billion, up 221% year over year, exceeding what analysts had expected. Free cash flow reached $13.7 billion, 46% of revenue, a genuinely strong number by any standard. CEO Hock Tan called demand for the company's custom AI accelerators very strong and guided AI semiconductor revenue to accelerate further, to $21.7 billion next quarter, up 236% year over year.

The guide for total revenue fell short. Broadcom projected fourth quarter revenue of $34.8 billion, up 93% year over year in absolute terms, but below the roughly $35 billion analysts expected. The stock fell as much as 6% in after hours trading before narrowing to a decline closer to 5% as the session went on. Acree's read afterward was direct: for a company this exposed to the AI trade, and priced accordingly, the beat wasn't big enough.

What this means

Here is the distinction worth sitting with. Before September 2, nobody had direct evidence about Broadcom's own quarter. What they had was Nvidia's answer to a related but different question: is demand for AI infrastructure still accelerating? That isn't the same as asking whether Broadcom specifically is capturing enough of that demand, at good enough economics, to justify what its own stock already cost. Investors used Nvidia's report anyway, because it was the best evidence available at the time, and a close competitor's demand signal really does tell you something. That's not confusion. That's what people do with any proxy when the real answer hasn't arrived yet.

Then the real answer arrived. Broadcom grew AI semiconductor revenue 221% year over year, beat on both revenue and earnings, and generated $13.7 billion in free cash flow. By almost any plain reading, that's an extraordinary quarter. The stock fell anyway, because the guide for the next quarter landed a few hundred million dollars under what was already priced in. Good results and a falling stock aren't a contradiction once you separate what the business actually did from what the valuation had already assumed it would do. Nvidia's report never had a way to speak to that second number. Only Broadcom's own report did, and once it arrived, that's what the market traded on. Nothing about the reaction after September 2 required reaching back to Nvidia, or to Warsh's speech, or to whatever mood the previous week had left behind.

Same substitution, different room

This isn't unique to chip stocks. A sales team hears a competitor land a big contract and assumes their own pipeline must be healthy too. A board sees an industry-wide growth report and expects every company in it to be capturing a share of that growth, including the ones quietly losing ground. A manager reads a strong economic forecast and assumes a specific product line will benefit, when the forecast never said anything about that product line at all. In each case, someone borrowed a category-level answer to a question that only direct, specific evidence can actually settle. The borrowing isn't the mistake. Not noticing you did it is.

What to watch

The next real test already has a rough date. Broadcom has forecasted reporting its fiscal fourth quarter around December 9. If Nvidia has reported its own third quarter by then, the same sequence is worth watching for: does Broadcom's stock move on Nvidia's numbers again before Broadcom has said anything about itself, and does that move hold or dissolve once Broadcom's own report supplies the direct evidence? If the pattern repeats, that says something durable about how these two companies get priced relative to each other, not just about one unusual week in late summer. If it doesn't, that's worth knowing too. It would suggest that whatever made Nvidia such a powerful proxy this time wasn't permanent, and would leave us asking what changed.

See you next quarter.