Broadcom Just Posted Its Best Quarter Ever — Why Is the Stock Falling?

Broadcom just reported the strongest quarter in its history, and the stock went down anyway. In the three months through late July, revenue rose 86% from a year earlier to $29.6 billion, powered by a record $16.7 billion of AI chip sales that grew 221%. When the results crossed the wire on September 2, shares slipped about 5% in extended trading on a guidance disappointment, and kept sliding the next day — leaving the stock roughly 29% below its June high near $495.

That gap between the printed numbers and the market's reaction is the whole story, and it has now happened twice. It is the kind of divergence that usually separates what a business is doing from what investors are willing to believe.

The market is still pricing the old risk

Three months ago the template was set. In early June, Broadcom reported its fiscal second quarter — revenue up 48% to $22.2 billion, above expectations — yet the stock plunged about 12.6% in a single day, erasing roughly $280 billion of market value. The trigger was the AI chip sales outlook, which missed the loftiest estimates. Investors who had bid the shares above $475 into the print wanted more than a beat; they wanted a forecast that cleared the bar of perfection, and they punished anything less.

Since then, skepticism has hardened into a durable frame. The worry is that Broadcom's AI business is a concentrated bet: it designs custom AI accelerators for a small circle of the largest cloud operators, which management describes as six frontier AI labs. If one hyperscaler taps the brakes on AI spending, or chooses to build its own chips in-house instead of buying Broadcom's, the trajectory breaks. Add the wider "AI bubble" fear and the threat of price competition from rivals, and you have the familiar story line the market keeps paying for: a great current quarter is just the setup for a disappointing future one.

That is the old story. The question is whether the underlying trajectory has actually worsened — or whether sentiment simply got uglier while the business kept compounding.

The numbers point the other way

The business has not just held up; it is accelerating. Management's own guidance lays out a three-year ramp that, on its face, is extraordinary: AI semiconductor revenue of roughly $58 billion for fiscal 2026, up 186% from a year earlier, then about $115 billion in fiscal 2027 and $230 billion in fiscal 2028. For the current quarter, Broadcom guided AI chip sales to about $21.7 billion, up 236%, on total revenue near $34.8 billion. Even after an 86% growth quarter, the guide calls for growth to get faster, not slower.

This is exactly the setup the market usually under-prices: expectations have already reset — shares down nearly 30% from the high, a sector that treats every AI print with suspicion — while the reported numbers have not broken. A low multiple is not the argument here, and I would not pretend it is. At roughly $1.7 trillion of market value and about 33 times trailing EBITDA, Broadcom is not cheap, and cheapness should never be mistaken for safety. The case rests on something else: whether this surge converts into cash fast enough to make the valuation look reasonable in a matter of quarters.

Free cash flow is the hard proof

That is where the evidence actually lives. Broadcom is a fabless chip designer plus an infrastructure-software business, which means it sells enormous revenue on a very thin base of physical assets. In the fiscal second quarter it generated $10.3 billion of free cash flow — $10.5 billion from operations against just $231 million of capital spending — a roughly 46% conversion of revenue. The first quarter was similar. This is a cash engine, not an earnings fantasy.

Run the simple version of the bridge, clearly as extrapolation rather than precision: if AI sales roughly double next year as management guides, and cash keeps converting near the 45 cents on the dollar Broadcom has been printing, then fiscal 2027 free cash flow is plausibly on a path that makes today's market cap a far smaller multiple of next-year cash than the trailing numbers suggest. The rerating bridge is not a clever DCF with seven moving parts; it is the difference between the multiple on revenue from last year and the cash that the next year is showing up for. That is the only number that will make "underappreciated" true.

What would prove it wrong

The bear case deserves a straight answer, because it is real and it is the whole bet. This is, at its core, a concentrated position in a handful of hyperscalers' willingness to keep spending on AI infrastructure. The specific condition that would break the operating case is a palpable AI capex pause or a big customer pulling its custom-chip work in-house — a slowdown in orders that shows up at the top of the funnel before any analyst revises a model.

So the tripwire is simple to watch: if quarterly free cash flow keeps stepping up as AI revenue compounds, the market is wrong and the falling stock has been an opportunity all along. If a hyperscaler taps the brakes and the cash path stalls, the falling stock is right. I can be wrong again, and this is a stock that has already humbled people once this year. But the numbers keep resolving to higher cash flow, and until they don't, the gap between what Broadcom is printing and what investors are paying for is worth paying attention to.