Broadcom Fell Again on a Record Quarter. Here's What Got Cheaper — and What Didn't.

The first thing worth noticing about Broadcom this quarter is that it got sold off on a good number, not a bad one.

On September 2, the chipmaker reported third-quarter revenue of $29.59 billion, up 86% from a year earlier, with adjusted profit of $3.32 a share — both ahead of what analysts expected. Its custom AI accelerators, the chips it designs for the biggest cloud companies, were the engine, with AI semiconductor revenue growing 221% year over year and another 54% from the prior quarter. By any number you'd find on a standard report card, this was an elite quarter.

Then the stock dropped about 5% anyway. Not because the quarter was weak, but because the next quarter's forecast — roughly $34.8 billion in revenue — landed a hair below the ~$35 billion the Street had penciled in. Reuters framed it the same way: a forecast that "signaling intense competition could hamper gains" from its custom processors.

That's the whole story, and it's the second time in two quarters Broadcom has been punished for doing exactly what a great AI business is supposed to do.

A report card that's elite, in a market that stopped reading report cards

The business is accelerating, not cooling off. Revenue growth of 86% this quarter is in a different league from the rest of the semiconductor peer set, and the forward picture is steep: management guided fiscal-2026 AI revenue to about $58 billion — up roughly 186% from last year — and outlined targets of $115 billion for fiscal 2027 and $230 billion for 2028.

The profitability and safety underneath are just as clean. Operating margin sits around 43% and free-cash-flow margin near 43%, and the company threw off $40.65 billion in operating cash flow over the trailing twelve months against a debt-to-equity ratio of about 0.6 and a quick ratio near 2.3x. This is not a stretched balance sheet. The balance sheet is the least of your worries.

What's changed is the market's grading scale. This is not the first time. In early June, Broadcom beat on its second-quarter results — $22.19 billion in revenue, a record $10.26 billion in free cash flow — and the stock still fell about 12.6% in a single day, because its AI revenue guide for the following quarter came in below the higher "whisper number" analysts were quietly targeting. Two beats, two sell-offs. The lesson investors are now pricing in is that at this stage of the AI cycle, "great" is not enough; the guide has to be better than the number the Street is already modeling, or the stock gets marked down.

What "not out of the woods" actually means

The reason the market is this sensitive isn't mood — it's that the specific shape of Broadcom's business concentrates the risk.

First, customers. Broadcom's AI growth leans on just six hyperscale customers — names like Google, Meta, OpenAI, and Anthropic — with commitments measured in multi-gigawatt deals. That's an extraordinary order book, but it means a single pause in capex by any one of them creates a hole that's hard to fill. The tension is live and acknowledged: reports have pointed to a rival, MediaTek, circling Google's work, and Broadcom's CEO has conceded the cloud giants want a "diversity of sources." You're not buying a diversified chip portfolio here; you're buying the next several years of spending by a handful of the largest companies on earth.

Second, margins. Those custom accelerators — the Google TPUs in particular — are lower-margin than Broadcom's software business, and as they make up a bigger slice of the top line, management has guided the overall gross margin down toward the mid-70s from the high-70s where it sat a year ago. The growth is real, but it's coming from a part of the business that earns a smaller margin per dollar. That mix shift is slowly eroding the historical high-margin profile that much of the premium multiple was built on.

So "not out of the woods" doesn't mean the business is weakening. It means the premium price depends on those six customers spending for several more years, and on the margin not sliding faster than the growth. Both are live, falsifiable conditions — which is exactly why the stock trades with the kind of volatility it has.

Cheaper is real — but only relative

Now the valuation, which is where "cheaper" earns most of its meaning — and where you have to be careful not to overstate it.

Broadcom is down about 28% from its 52-week high of $495 to the mid-$350s. But is it cheap in absolute terms? No. Put it beside its actual comparison set:

Company	Trailing P/E	EV/EBITDA	Dividend yield
Broadcom	~44x	~34x	~0.7%
Nvidia	~29x	~27x	~0.1%
Marvell	~69x	~65x	~0.1%
AMD	~116x	~77x	—
Texas Instruments	~38x	~25x	~2.2%
Qualcomm	~19x	~16x	~2.7%

On a trailing price-to-earnings basis, Broadcom is richer than Nvidia (29x), Texas Instruments (38x), and Qualcomm (19x) — and it only looks "cheap" next to Marvell (69x) and AMD (116x), the two names growing far more slowly than it is. A low-ish multiple is not the story here. Broadcom commands 44x because its growth is in another league; the premium multiple is justified only if that growth keeps compounding, which loops straight back to the concentration risk above.

Where the "cheaper" is genuinely real is in how the multiple behaves as earnings catch up. The ~44x trailing multiple is calculated before the quarter it just reported is rolled in; once you include the $3.32-a-share quarter, the trailing multiple falls to roughly 37x, and on my estimate of full fiscal-2026 earnings it lands in the low 30s. The stock hasn't been repriced to a bargain — it's been repriced to a more reasonable multiple on earnings that are growing very fast. That's a meaningfully better price than $495, without it being a "cheap" one.

For a cross-check, AInvest's aggregate signal labels Broadcom a Buy (a fundamental score of 3.95 out of 5). That lines up with the growth and profitability read. What an aggregate label like that won't tell you is the one part of the stack that's still red: momentum.

Where it fits in a portfolio

The factor stack is four-for-five on quality. Growth is elite and still accelerating (revenue +86%, AI +221% year over year). Profitability and safety are both at the top of the peer set. Valuation is a neutral-to-fair premium: expensive in isolation, reasonable against the growth. And momentum — the one factor I watch for timing — is unambiguously red: the stock is down about 15% over the past month, trading below both its 50-day (around $384) and 200-day (around $370) moving averages, with a negative MACD and a relative-strength index near 37. In plain terms, the price has not found a floor yet. The de-rating is still in motion.

So what does the stack say to do? Broadcom is a quality-growth name — the kind of proven winner you don't mechanically cut just because the price got ugly — and the de-rating has made it more attractive than it was at the top. That supports holding it as a core AI-growth position, at a better price than a month ago. But with momentum still pointing down, I'd size it like a "give a proven winner room" hold rather than an all-in dip-buy, and I'd let the numbers, not the narrative, set the next move.

If you're nervous about the AI concentration — and this is the single-name, six-customer version of the whole sector's risk — that's exactly where a barbell earns its place: pair Broadcom with a durable-cash-flow or dividend name on the other end so a stumble in AI spending doesn't move your entire book.

Two things would change the call in either direction. A next-quarter guide that lands at or above consensus, plus the stock reclaiming that ~$384 50-day line, would turn this into an active add. A second straight below-whisper guide, or a named customer visibly pulling back on AI capex, would turn it into a reduce. Until one of those shows up, the honest read is the one the price is telling you: a great business, a falling stock, and a floor that hasn't been found yet.