Broadcom Reports $29.6 Billion in Quarterly Revenue, Up 86%; Shares Close Down 3.4% After $34.8 Billion Guidance

A Record Quarter That Cost Broadcom 3.4% of Its Stock

Broadcom sold $16.7 billion of AI chips in three months, up 221% from a year earlier^[1][3]. Total revenue hit $29.591 billion, up 86%, beating the $29.519 billion Wall Street had penciled in^[2][5]. And the stock fell anyway, closing down 3.43% at $357.16 the next day^[6][7].

Both of those things are true at once, and they are the whole story. Broadcom did not miss on anything it reported for the quarter that ended Aug. 2, 2026^[1][2]. What moved the stock was the forecast for the next one: about $34.8 billion, versus the roughly $35.03 billion analysts expected^[1][5]. That gap is about $200 million, under 1% of the number^[5].

To understand why a rounding error can erase billions in market value, you have to understand how growth stocks get priced. Investors are not really paying for the quarter that already happened. They are paying for the quarters still to come. A company like Broadcom is valued on the assumption that AI revenue keeps compounding for years, so a forecast that comes in even slightly soft can matter more than a blowout quarter that everyone already expected^[1][5]. That is a mechanical feature of how growth is priced, not a verdict on the business.

The Debt Question Sitting Inside the Chip Business

There is a second, less visible fight running underneath the guidance gap, and it goes back to June. Broadcom, along with Apollo Global Management and Blackstone, set up a $35 billion financing platform called AI XPV, meant to fund more than 20 gigawatts of chip deployments for AI labs including Anthropic and OpenAI^[9][10]. That is roughly the electricity draw of a mid-sized U.S. state.

Here is the mechanism that matters. AI labs need enormous amounts of computing power but often don't have the cash to buy it outright. The AI XPV platform lets them lease chips instead, financed through debt — a $6 billion note and a $24 billion note. Broadcom agreed to back both with what's called a residual value support agreement^[11].

In plain terms: if an AI lab keeps paying its lease, the arrangement costs Broadcom nothing, and Broadcom gets to book the chip sale today instead of waiting years^[11]. If a lab stops paying, and the used chips sell for less than what's owed, Broadcom has to make up the difference^[11]. Bank of America has modeled that exposure growing to roughly $370 billion by mid-2029 as the platform scales — though the bank's own estimate of the maximum realistic loss is far smaller, about $42 billion if every single lease defaulted, and $10.5 billion at a 25% default rate^[11].

Nobody disputes that this agreement exists or what it says. Broadcom disclosed it. What's disputed is what it means: is it a smart way to help cash-poor AI labs afford chips, or is Broadcom quietly financing its own growth?

Supply-Constrained, Not Demand-Constrained

CEO Hock Tan's answer is that the debt question is beside the point, because demand is no longer the thing limiting Broadcom. Supply is^[5]. He told analysts a fourth major AI lab had signed on with a $10 billion order, that Google had committed to buying "multi-tens of billions" of dollars of custom chips a year, and that Anthropic is projected to become Broadcom's largest custom-chip customer by 2027^[3][12].

Custom chips, known as XPUs, are different from the general-purpose GPUs Nvidia sells. An XPU is designed around one customer's specific workload, which takes years of engineering and large upfront costs. Once a lab commits to that design, switching to a different chipmaker is expensive and slow. That lock-in is Broadcom's business model, and it's also why Tan raised his fiscal 2027 AI revenue target to about $115 billion, up from the $100 billion he had held to as recently as June, plus $58 billion for this fiscal year and $230 billion by 2028^[5][8].

Broadcom's argument about the financing structure follows the same logic. New AI labs have huge compute needs and thin balance sheets. By lending its own creditworthiness to the senior debt tranches, Broadcom lowers the borrowing cost, letting a customer deploy chips now rather than paying over a decade — while Broadcom books the sale immediately and commits relatively little of its own cash^[11].

Bulls, Skeptics, and a Stock That Already Fell Once

Wall Street's bulls mostly took Tan's side. The day after earnings, BMO Capital raised its price target to $575 from $455, Macquarie upgraded the stock, and Morgan Stanley nudged its target to $505^[13]. Across 49 analysts tracked by S&P Global, the consensus rating stayed Strong Buy, with an average target of $530.52 — well above where the stock actually closed^[7]. Their view is that a sub-1% guidance shortfall is noise against 221% growth, and that Broadcom's forecast is conservative because it can only ship what its chip foundries and memory suppliers can build.

Skeptics, including analysts at Bank of America, aren't arguing the sales are fake. Their point is that revenue quality matters as much as revenue growth, and that a business partly funded by debt it also guarantees is a different kind of growth than one built entirely on outside cash^[11]. BofA cut its price target to $460 from $530 the day after earnings; Evercore ISI trimmed its target too^[13]. They also point to customer concentration: AI now makes up about 56% of Broadcom's revenue, and the buyer list is short — Google, Anthropic, OpenAI and a handful of hyperscalers^[3][4]. At that level of concentration, one buyer pulling back on spending is not a rounding error. Some traders are betting on exactly that: longer-dated 2027 options showed more than 2.5 times as many downside bets as upside ones^[4].

The stock's reaction suggests the skeptical read is already partly priced in. Shares closed Sept. 3 about 28% below their June peak of $495, even though the business kept growing^[6][7]. Both sides are working from the same disclosures. They just weigh the same numbers differently.

What the Coverage Left Out

Financial outlets covering the earnings mostly picked one of two frames: growth story or sentiment story. CNBC ran with "Broadcom delivers strong earnings view," borrowing the company's own language for a guide that knocked billions off the stock, while its investing club conceded in the same breath that investors "wanted more"^[5]. The Motley Fool led with AI chip sales tripling and demoted the guidance shortfall to a number "investors need to watch"^[3].

24/7 Wall St. ran two headlines within a day, one framing the quarter as barely beating estimates and another describing Wall Street as having "suddenly" developed doubts about the AI business — as if the concern were a mood shift rather than a response to a disclosed financial structure^[4]. Independent trackers rate both 24/7 Wall St. and The Motley Fool as center to left-center outlets, not part of any identifiable U.S. right-leaning coverage; no outlet with a clear right-leaning political orientation turned up covering this earnings report at all. Overseas, South Korea's Seoul Economic Daily ran the most stripped-down version: "Broadcom Revenue Jumps 86% but Shares Slide on Q4 Outlook," stating both facts with no verdict attached^[6].

What was largely missing from earnings-day coverage, across outlets, was sustained attention to the off-balance-sheet financing question — the residual value guarantee that is the strongest specific evidence the skeptics have, and it comes from Broadcom's own disclosures^[11]. The company's preferred numbers, the $115 billion and $230 billion multiyear targets, were released the same day as a quarterly guide the market didn't like, and largely dominated the story that got told.

What Nobody Actually Knows Yet

Strip away the framing, and the physical facts aren't in dispute. Broadcom generated $13.7 billion in free cash flow this quarter — cash left over after running the business and buying equipment, the money actually available for dividends, buybacks, or paying down debt^[1]. Chips are being built. Racks are being installed. The AI XPV platform is targeting more than 20 gigawatts of real deployment through 2028^[9][10].

What's genuinely unknown is whether AI labs will generate enough revenue over the coming years to keep paying multiyear leases on chips that lose value fast, and what a three-year-old custom accelerator will actually be worth if it needs to be resold. Nobody has that data yet, because the leasing structure and the chips themselves are too new^[11]. Broadcom's bulls and its skeptics are both making forecasts about that same unknown. The quarter that just closed doesn't settle it either way.