What I’m watching when Broadcom reports fiscal third-quarter results on September 2, 2026
Broadcom is approaching the kind of earnings report that makes reasonable investors behave like unpaid detectives.
The company will report fiscal third-quarter 2026 results after the market closes on September 2. Between now and then, I expect analysts to inspect cloud capital-expenditure plans, custom accelerator shipments, networking demand, VMware contract trends, margin assumptions, and possibly the tone of Hock Tan’s breathing. This is what happens when a company becomes one of the market’s most important artificial-intelligence suppliers: excellent execution stops being impressive and becomes the minimum cover charge.
I understand the excitement. Broadcom’s latest results were extraordinary. Fiscal second-quarter revenue reached $22.2 billion, up 48% from a year earlier. AI semiconductor revenue hit $10.8 billion, an increase of 143%. Adjusted EBITDA was $15.2 billion, or 69% of revenue. Free cash flow came to $10.3 billion, equal to 46% of revenue. These are not merely good numbers. These are numbers that arrive wearing cuff links.
Then management raised the stakes.
For the fiscal third quarter, Broadcom guided to approximately $29.4 billion in total revenue, representing 84% year-over-year growth. It expects AI semiconductor revenue of $16 billion, more than triple the year-ago level. It also expects a non-GAAP operating margin of approximately 67% and adjusted EBITDA of approximately 68% of revenue. The company has effectively walked into the quarterly earnings casino, placed a mountain of chips on the table, and told investors not to worry because it understands probability.
That may be true. Broadcom has earned considerable credibility. But credibility does not make the stock immune to expectations.
As I look ahead to the report, I am concentrating on three engines: AI revenue, VMware, and margin strength. I am also watching cash flow, customer concentration, inventory, debt, and guidance, because the numbers companies place below the headline often explain what happens after the confetti lands.
My central view is straightforward: Broadcom does not need to prove that it participates in AI. That case has already been made. It needs to prove that its acceleration is durable, that VMware can support the economic model without alienating the customer base beyond repair, and that the company can preserve remarkable profitability while semiconductor revenue becomes a larger share of the business.
That is a harder test than merely beating an earnings estimate by a few cents.
The Numbers Broadcom Has Already Put on the Table
Before I decide what would count as a strong quarter, I need to begin with management’s own targets.
Broadcom’s June 3 earnings release provided the following fiscal third-quarter outlook:
Revenue of approximately $29.4 billion.
Non-GAAP operating income of approximately 67% of revenue.
Adjusted EBITDA of approximately 68% of revenue.
AI semiconductor revenue of approximately $16 billion, growing more than 200% year over year.
The quarter ended August 2, so the operational work is already finished. Investors are now arguing over what happened inside a period they cannot observe, which is one of finance’s more sophisticated forms of waiting outside a locked restaurant and reviewing the meal.
The $29.4 billion revenue target is the obvious headline. It implies sequential growth of more than $7 billion from the second quarter. That is an enormous step, even for a company benefiting from the largest infrastructure buildout in modern technology.
The $16 billion AI figure is even more important. If Broadcom reaches it, AI semiconductor revenue alone would represent more than half of total company sales. Broadcom would no longer be a diversified semiconductor and infrastructure-software company with an impressive AI business attached. It would be an AI infrastructure powerhouse supported by a highly profitable software portfolio and several mature semiconductor franchises.
That difference matters for valuation, volatility, customer concentration, and how investors interpret every sentence about cloud spending.
AI Revenue: The Headline Has Become the Business Model
When investors talk about Broadcom’s AI opportunity, they sometimes reduce it to custom accelerators. That is understandable because custom XPUs receive the glamorous lighting. Yet Broadcom’s value is broader. The company supplies custom compute silicon, high-speed switching, routing, optical components, digital signal processors, physical-layer technology, PCIe connectivity, and the networking architecture required to connect enormous clusters of processors.
In plain English, Broadcom helps build both the brains and the nervous system of large AI data centers.
That position is attractive because training and serving increasingly large models require more than a powerful chip. Thousands of accelerators must communicate quickly and efficiently. If networking becomes the bottleneck, a data center full of expensive processors begins operating like a committee with terrible meeting software.
Broadcom’s custom accelerator business gives hyperscale customers an alternative to relying entirely on merchant GPUs. A custom chip can be optimized for a customer’s workload, power budget, software stack, and data-center design. It may offer better economics at sufficient scale, although “sufficient scale” in this market means spending amounts that would cause most finance departments to seek spiritual counsel.
What I want to hear on the call is not merely that demand remains strong. Demand has been strong for several quarters. I want specifics about the durability and breadth of the opportunity.
First, is the $16 billion quarter driven by repeatable volume or a particularly favorable shipment schedule? Semiconductor revenue can be lumpy. A large customer’s deployment timetable, packaging availability, foundry capacity, or product transition can move billions of dollars between periods. A spectacular quarter built partly on timing is still spectacular, but it should not automatically be annualized with the confidence of a person multiplying in a casino.
Second, how much of the growth comes from custom accelerators versus AI networking? Broadcom benefits from both, but their product cycles, margins, competitive risks, and customer dynamics differ. Networking may provide a broader way to participate in AI infrastructure even when the identity of the winning compute architecture shifts.
Third, what is management seeing for fiscal 2027? Earlier this year, Broadcom projected more than $100 billion in AI semiconductor revenue for 2027. That number transformed the scale of the thesis. The September call will be an opportunity to confirm whether customer plans, product schedules, and capacity arrangements still support it.
I am not expecting management to provide a perfectly measured staircase into the future. AI spending does not move in a perfectly measured staircase. It moves like a crowd that just heard somebody else found gold. But I do want evidence that the pipeline is converting into committed programs, not simply expanding as an impressive collection of conversations.
The Customer-Concentration Question Nobody Should Whisper
Broadcom’s AI strength comes with a risk that deserves more attention: a relatively small group of enormous customers accounts for a large portion of revenue.
In its latest Form 10-Q, Broadcom said one semiconductor distributor represented 42% of quarterly revenue. It estimated that its five largest end customers collectively accounted for approximately 45% of total revenue.
Customer concentration is not automatically bad. Selling vast quantities of mission-critical technology to the world’s strongest companies is a lovely problem. These customers have enormous balance sheets, expanding infrastructure budgets, and a need for products that very few suppliers can deliver.
But concentration changes the risk profile. A schedule adjustment at one customer can affect a quarter. A customer’s internal chip program can alter future demand. Pricing negotiations become consequential. Export restrictions, supply constraints, or a decision to slow capital spending can travel through Broadcom’s income statement faster than investors can refresh a dashboard.
I will therefore listen for evidence of diversification across the company’s major AI customers. I want multiple programs, multiple generations, and multiple forms of revenue. I would rather see Broadcom embedded throughout a customer’s compute and networking architecture than dependent on a single chip generation that happens to be having a magnificent year.
The bull case is not simply that hyperscalers are spending heavily. It is that Broadcom becomes structurally difficult to remove from their infrastructure.
Inventory Is Telling Me Broadcom Expects to Ship
One of the most revealing details in the last filing was inventory. Broadcom’s inventory rose to $4.33 billion at May 3 from $2.27 billion at the beginning of the fiscal year. Management said the increase primarily supported higher expected shipments of custom AI accelerators.
That is a substantial build.
In the bullish interpretation, it is physical evidence that Broadcom is preparing for the revenue ramp it has already guided. Inventory must exist before it can become sales. A company does not support a leap from $10.8 billion to $16 billion in quarterly AI revenue by placing a few extra boxes near the loading dock.
In the cautious interpretation, inventory increases working-capital risk. If customer schedules change or demand shifts, inventory can sit longer than expected. Custom products may have limited alternative uses. The more specialized the silicon, the less likely it can be politely redirected to another buyer.
I do not currently view the inventory growth as a red flag because it aligns with explicit revenue guidance and a known ramp. But I will compare inventory, receivables, and cash conversion in the upcoming quarter. If revenue surges while cash flow weakens sharply and working capital absorbs more cash, I will want to understand why.
AI enthusiasm is wonderful. Cash collection remains the traditional method by which enthusiasm becomes a business.
VMware: The Less Glamorous Engine With the Better Economics
AI will receive most of the headlines, but VMware may be the more revealing measure of Broadcom’s discipline.
Broadcom acquired VMware in November 2023 for approximately $69 billion. Since then, it has simplified the product portfolio, emphasized VMware Cloud Foundation, shifted more business toward subscription licensing, narrowed its customer focus, and reworked the partner ecosystem. Customers and resellers have complained about pricing, licensing changes, product bundling, and reduced flexibility. Broadcom has responded, in effect, by insisting that the destination will justify the turbulence.
The financial results so far support part of that argument.
Infrastructure-software revenue reached $7.18 billion in the second quarter, up 9% year over year. Segment operating income rose 13% to $5.65 billion. Broadcom said the improvement was driven primarily by strong demand for VMware Cloud Foundation. That implies a segment operating margin approaching 79%, which is the sort of profitability usually found in dreams, toll roads, and software sold to enterprises that cannot casually uninstall it on Friday afternoon.
VMware serves two purposes in the Broadcom story.
First, it supplies recurring, high-margin revenue that balances the greater cyclicality and capital intensity of semiconductors. Second, it expands Broadcom’s control over essential enterprise infrastructure. VMware Cloud Foundation combines compute virtualization, storage, networking, management, and private-cloud capabilities into an integrated platform.
The thesis is that large enterprises want a standardized private-cloud stack for critical workloads, including AI applications that cannot or should not run entirely in the public cloud. If that thesis works, VMware is more than an acquired cash-flow machine. It becomes part of Broadcom’s broader AI infrastructure position.
What I need to see is evidence that the financial optimization remains sustainable.
There is a difference between improving revenue quality and extracting more money from customers who feel trapped. Both can look excellent in a quarterly margin calculation. Only one builds durable goodwill.
I will listen for VCF bookings, annualized contract value, renewal activity, customer adoption, and the balance between large strategic accounts and the wider customer base. Broadcom may decide that serving fewer, larger customers is economically rational. That is management’s choice. Investors still need to understand whether the resulting growth comes from expanding platform value or from concentrating revenue among customers with limited short-term alternatives.
High switching costs are an asset until customers begin treating the switch as a liberation project.
Margin Strength: Where the Two Businesses Collide
Broadcom’s margin performance is the part of the story I find most impressive.
In the second quarter, GAAP gross margin was 69%, up from 68% a year earlier. Adjusted EBITDA reached 69% of revenue. Free-cash-flow margin was 46%. These figures reflect strong revenue growth, disciplined spending, the high profitability of software, and an operating model that seems capable of turning scale into cash with very little ceremony.
Yet the product mix creates tension.
Infrastructure software carries higher gross margins than semiconductor solutions. As AI chips become a larger percentage of revenue, Broadcom’s consolidated gross margin can face mix pressure even while operating profit dollars surge. The company acknowledged this in its filing: the benefit from revenue growth was partially offset by a larger semiconductor mix.
This is not necessarily bad. I would happily accept modest gross-margin compression in exchange for massive, profitable AI revenue growth. Percentages do not pay dividends; dollars do. The danger arises if investors fixate on one margin line without understanding the mix underneath it.
Management expects the third-quarter non-GAAP operating margin to remain around 67% and adjusted EBITDA margin around 68%. Holding those levels while revenue increases 84% would demonstrate remarkable operating leverage. It would show that the AI ramp is not merely adding low-quality volume and that VMware continues to provide an economic counterweight.
My threshold is not perfection. If the company lands near guidance and explains any variation through product mix or ramp costs, the fundamental story remains intact. If margins weaken materially without a corresponding increase in future opportunity, I will become more cautious.
I also keep one eye on the gap between GAAP and non-GAAP results. Broadcom’s non-GAAP calculations exclude stock-based compensation, acquisition-related amortization, restructuring charges, and other items. In the second quarter, stock-based compensation alone was approximately $2.09 billion, up from $1.77 billion a year earlier. The company had $20.1 billion of unrecognized compensation cost associated with unvested awards expected to vest.
Non-GAAP measures are useful for understanding operating trends. They do not possess the authority to make compensation economically imaginary.
Free Cash Flow Is the Reality Check
I can forgive many accounting complexities if cash keeps arriving.
Broadcom generated $10.49 billion in operating cash flow during the second quarter and spent only $231 million on capital expenditures, producing $10.26 billion in free cash flow. The 46% free-cash-flow margin was extraordinary.
Broadcom’s low capital expenditure partly reflects its fabless semiconductor model. The company designs complex chips but relies on manufacturing partners rather than building its own leading-edge fabrication plants. That allows it to participate in enormous silicon demand without personally funding every clean room, though it also creates supply-chain dependence.
For the upcoming quarter, I want free cash flow to remain strong after working-capital needs. Receivables rose to $10.83 billion at the end of the second quarter, and inventory nearly doubled during the first half. Those increases were understandable in the context of growth, but growth consumes cash before it produces cash.
If Broadcom can convert the third-quarter revenue surge into proportional operating cash, the company will have immense flexibility. It can reduce debt, pay dividends, repurchase shares, fund research, and pursue acquisitions. During the first half of fiscal 2026, Broadcom generated $18.75 billion in operating cash flow, repurchased $8.45 billion of shares, and paid $6.18 billion in dividends.
That is not a company surviving on a narrative. That is a company converting a narrative into deposits.
Debt Still Matters, Even When Cash Flow Is Excellent
The VMware acquisition left Broadcom with substantial debt. As of May 3, the company reported $66.72 billion in outstanding indebtedness, including $2.25 billion due within twelve months. Quarterly interest expense was $776 million.
This does not frighten me at the current rate of cash generation. Broadcom had $19.63 billion in cash and access to a $7.5 billion revolving facility, and its free cash flow provides significant coverage. The company has also been refinancing and managing maturities.
Still, debt is not decorative. It claims cash before shareholders do. It can limit flexibility during a downturn, and it makes the durability of VMware cash flow and AI revenue more important.
I will watch the net-debt trajectory and management’s capital-allocation priorities. Broadcom has rewarded shareholders through dividends and buybacks, but debt reduction remains valuable when rates are elevated. Repurchasing an expensive stock while carrying tens of billions in debt can be rational, but it requires more than enthusiasm and a confident spreadsheet.
The Valuation Has No Interest in Modesty
Broadcom shares traded around $421 on August 5. The exact valuation depends heavily on whether I use GAAP earnings, adjusted earnings, or aggressive future estimates. That spread is itself instructive.
On trailing GAAP earnings, the stock looks extremely expensive. On forward adjusted earnings that incorporate the enormous AI ramp, it looks considerably more reasonable. Investors are effectively paying today for earnings Broadcom is expected to produce tomorrow, which is standard practice until tomorrow develops scheduling conflicts.
I do not think a premium valuation is automatically unjustified. Broadcom combines high growth, market-leading semiconductor assets, recurring software revenue, exceptional margins, and prodigious free cash flow. Few companies offer that combination.
But the higher the expectations, the less meaning a conventional “beat” carries. The market will focus on the size of the beat, the quality of the revenue, and the direction of guidance. Broadcom can report fantastic results and still fall if investors were positioned for something closer to a religious experience. That happened after the previous report, when the stock initially dropped despite spectacular growth and aggressive guidance.
Price reaction and business performance are not the same thing. I try to remember that before allowing an after-hours chart to explain the company to me.
My Earnings Scorecard
I am judging the report across six areas.
1. Total revenue
I want revenue near or above the $29.4 billion target. A modest variation caused by timing would not destroy the thesis, but a material miss would raise questions because the guidance was issued only one quarter earlier and the customer base is concentrated.
2. AI semiconductor revenue
The $16 billion target is the center of the event. I want to see the number met, plus evidence that fiscal 2027 programs remain on schedule. Commentary about custom accelerators and networking should support a multi-year opportunity rather than a single-quarter shipment surge.
3. VMware and infrastructure software
I want continued VCF adoption, stable or improving software growth, and evidence that bookings support future revenue. I also want some reassurance that customer discontent is not becoming a structural threat hidden behind near-term contract economics.
4. Margins
I want non-GAAP operating margin near 67% and adjusted EBITDA margin near 68%. I will tolerate mix-driven gross-margin movement if operating profit and cash flow remain strong.
5. Cash conversion
I want operating cash flow and free cash flow to rise with revenue. Inventory and receivables should make sense relative to shipment growth. Cash is where promotional language goes to receive a background check.
6. Guidance
The next-quarter outlook may matter more than the quarter being reported. I will focus on AI revenue, customer program timing, networking demand, software growth, and margin expectations. Any update to the 2027 AI opportunity will move the stock because investors have already built a large portion of their valuation case around it.
Bull, Base, and Bear Cases
My bull case is that Broadcom exceeds the $29.4 billion revenue guide, AI sales surpass $16 billion, VMware growth remains healthy, adjusted EBITDA margin holds near 68%, and management raises or strengthens its fiscal 2027 AI outlook. In that scenario, the company proves that custom accelerators and networking are scaling faster than expected while software protects profitability. The stock could justify a premium, though the immediate reaction would still depend on how much optimism is already priced in.
My base case is that revenue and AI sales land close to guidance, margins remain strong, VMware progresses without a dramatic acceleration, and management reiterates its long-term outlook. That would be an excellent operational result, even if the stock reacts with all the warmth of a restaurant critic informed that the meal was merely outstanding.
My bear case is a material AI shipment delay, weaker guidance, margin pressure beyond product mix, or evidence that customer concentration is creating volatility. A slowdown in VMware bookings or signs of customer migration would add concern. The most damaging outcome would be a combination of softer AI growth and weakening software economics, because the two businesses are supposed to balance each other.
What I Would Do as an Investor
I would not buy Broadcom solely because I expect an earnings beat. That is speculation about the market’s reaction to a result that thousands of other people are also attempting to predict.
I would buy it only if I wanted to own a leading supplier of custom AI compute, data-center networking, and mission-critical infrastructure software for several years. I would also size the position with the understanding that a highly valued AI stock can fall sharply after good news. Expectations turn small disappointments into trapdoors.
For existing shareholders, the quarter should be used to test the thesis, not celebrate or mourn one candle on a chart. Is the AI opportunity expanding? Is Broadcom winning additional generations of customer programs? Is VMware becoming a stronger platform? Are margins and free cash flow holding? Is debt becoming more manageable?
If the answers remain yes, temporary volatility may create opportunity. If the answers begin turning no, no amount of AI vocabulary should substitute for reassessment.
The Bottom Line
Broadcom enters its fiscal third-quarter report with enormous momentum and almost equally enormous expectations.
The AI semiconductor business is scaling at a rate rarely seen in a company of this size. VMware supplies recurring revenue and exceptional operating profitability. The combined model is producing margins and cash flow that would make most technology executives quietly close the comparison slide.
But the report is not risk-free. Customer concentration is high. Inventory has increased rapidly. VMware’s financial optimization has generated customer frustration. Debt remains substantial. The semiconductor mix can pressure gross margin. And the stock price assumes that Broadcom will continue converting the AI boom into real, durable earnings.
I remain constructive because the company has demonstrated something more valuable than a fashionable story: execution. Broadcom is not merely describing AI demand. It is shipping products, growing segment profit, building inventory for committed ramps, and generating cash.
What I need on September 2 is confirmation that the engine is durable.
I want AI revenue at or above $16 billion. I want VMware to show that VCF is becoming a sustainable platform rather than an exceptionally profitable customer endurance test. I want margins near guidance and cash flow strong enough to make the accounting debate less important. Most of all, I want management’s forward commentary to show that 2027 demand remains supported by actual customer programs.
Broadcom has already proven it can produce a great quarter.
Now it has to prove that great has become ordinary.
That is a wonderful problem for a company to have—and an exhausting one for its shareholders.
Market data and company information reflect what was available on August 5, 2026. This article is commentary for informational purposes and is not personalized investment advice.
Comments
Post a Comment