Broadcom Looks Undervalued. Its Biggest Customer Just Hired the Competition.

Sell-side note titles like "Broadcom's valuation is absurdly inadequate" are doing something specific: they're pointing at a stock that is down roughly 28% from its $495 high while still posting some of the fastest revenue growth of any large company in the market. Broadcom's AI chip business grew 143% last quarter to $10.8 billion, and its CEO guides that business to "in excess of $100 billion" next fiscal year — a number bigger than the company's entire current revenue. On a forward basis, the stock trades at a multiple that makes the "too expensive" debate feel pointless.

If I were you, I'd grant all of that. The growth is real, the cash is real, and the forward number does look cheap. That is exactly why the discount is worth taking seriously rather than shopping. "Undervalued" and "the market is making a mistake" are not the same sentence. The first describes a price. The second requires the price to be wrong — and a price is only wrong if you believe the future the price is refusing to pay for is guaranteed. Broadcom's discount is the market quietly refusing to pay for that guarantee.

The Number Looks Cheap, Because the Earnings Are a Promise

Start with what the bulls are actually measuring, because there's a trap hiding inside the cheap number.

When people say Broadcom trades at "about 19 to 22 times earnings," they are dividing today's price by the profit management promises for fiscal 2027 — a year at the top of a steep, still-unfinished ramp. That is a forward multiple, and it is doing a lot of the heavy lifting in the "inadequate" framing.

But the same stock, priced against the twelve months of earnings already in the books (trailing GAAP, which is dented by acquisition-related amortization), runs well into the 40s. The stock is not cheap and expensive. It is cheap on the year management is promising and expensive on the year it has already delivered. The entire bull case rides on the gap between those two years actually closing — on the $100 billion showing up.

That matters for the simple reason that the $100 billion is not a base rate you can lean on. It is a projection built on a handful of very large, very specific commitments from a handful of very large customers. Which brings us to where the "cheap" story runs out of road.

A Franchise Built on Six Phone Numbers

The celebrated number is the AI revenue. The neglected number is who buys it.

Broadcom does not sell a product to the world. It designs custom silicon — chips built from scratch for a specific customer's own AI workloads — and it sells them to a small club. Management frames it as six core customers, named or not: Google (Alphabet), Meta, OpenAI, Anthropic, and two it does not name. Spread the $100 billion target across those six and each one has to buy more than $16 billion a year. One of them, Anthropic, has committed a $10 billion order. Two of them have placed only about $6 billion so far. There is no long tail of small buyers standing behind the number to soak up a stumble. If one of the six slows its spending — and these are businesses whose budgets flex with the state of the AI trade — there is no cushion underneath.

And the most important chair at that table is Google. Broadcom designs Google's in-house TPU chips, and in April it signed a long-term agreement to build Google's future TPU generations and supply the chips for its AI racks through as late as 2031. That relationship is a big reason the growth ran at 143% instead of some more ordinary rate. The 143% is not the base rate of a durable market position. It is the price of exclusivity — Broadcom being, for a while, the chosen partner.

That is the hidden premise the "undervalued" framing rests on: that the exclusivity is permanent, and that the earnings the forward multiple prices in are therefore almost certain.

The Customer Just Picked a Second Vendor

Here is the fact that the cheap-stock story tends to flatten.

On August 19, Marvell Technology announced an expanded custom-chip agreement with Google — a warrant tied to as much as $120 billion of future purchases. Read that again. Broadcom's most important AI customer is adding Marvell as a second silicon partner. The stock fell about 5% that morning, while Marvell jumped roughly 8% and Nvidia barely moved. The market was not fleeing the sector. It was repricing who gets Google's next dollar.

A week and a half later, the message repeated itself, this time from Broadcom's own mouth. On September 2, Broadcom guided its next-quarter revenue to about $34.8 billion — below the Street's $35.03 billion estimate — and flagged "intense competition" as a reason it could cap its custom-chip gains. (A month earlier, its June quarter AI chip guide of $16 billion had already landed under the $17.2 billion consensus.) Two signals, same direction: the exclusivity that produced the 143% is beginning to loosen.

This is not a random correction. There is a base rate for exactly this pattern, and it is not flattering. When Apple — about 20% of Broadcom's revenue at the time — was first reported to be "designing away" its Broadcom Wi-Fi and Bluetooth chip in 2023, Broadcom's bulls called it noise. By September 2025, Apple was shipping its own N1 wireless chip. Big customers can decide the supplier's specialty is a commodity, and then shop it out. Google is not dropping Broadcom; it is doing something slower and more durable: it is making Broadcom replaceable.

What Would Prove the Crowd Wrong

I owe you the strongest counter, because it has legs.

Google is adding Marvell, not firing Broadcom — the 2031 TPU deal still stands, and a second vendor does not automatically mean zero work for the first. Hock Tan has repeatedly reaffirmed the $100 billion. And Broadcom is not only defending: in June it unveiled "Jalapeño," an inference chip co-developed with OpenAI, a fresh demand source aimed at serving end-user AI rather than training models. If the $100 billion holds, if no second anchor customer follows Google in dual-sourcing, and if the custom-silicon niche keeps expanding faster than the new entrants can fill it, then the discount is a genuine bargain and the stock overcorrected on a scare.

That is a live, testable case. But notice what it requires: it requires the exclusivity to survive, quarter by quarter, customer by customer. It is not "the technology is real" — the technology is real. It is the harder bet that the scarcity which made Broadcom the pick stays scarce.

The Discount Is a Question, Not a Discount

So here is the reframe the "absurdly inadequate" title skips.

The stock is not cheap because the market is dumb. It is discounted because the market is pricing a risk the forward multiple pretends isn't there: that Broadcom's explosive growth is concentrated in six relationships, and the most important of those relationships just hired a competitor and is, per Broadcom's own guidance, facing "intense competition." The multiple you're being told is "inadequate" is not a free option on a guaranteed $100 billion. It is a price that already reflects a question — who else will these six customers hire, and how much of the work will they take away?

The bulls are right about the past: this is a superb business growing at an exceptional rate. The question the discount is asking is about the future, and it's a different question: whether custom-silicon exclusivity is a moat, or a menu. The same company can keep winning the headline — pushing toward $100 billion — while a shareholder who paid for a monopoly quietly pays for a market that is becoming a bidding war. Before you call the price inadequate, ask what the price already knows.