Sponsored by edgeful

Broadcom $AVGO ( ▼ 12.59% ) dropped about 13 on a quarter where it beat on both revenue and earnings, which sounds like a contradiction and sort of is. The print itself was fine. The selloff was about the gap between what got reported and everything investors had quietly talked themselves into expecting on top of it.

Here is what actually came out, because it gets lost in the reaction. April quarter revenue was $22.2B, up 48% from a year ago and a touch ahead of the Street. Earnings were $2.44 a share against $2.39 expected. AI chips did $10.8B, up 143%, a little better than people had modeled, and even the non-AI side of the business, the part everyone forgets is there, picked up. Then they guided the July quarter to roughly $29.4B, up 84% year over year and about a billion above what Wall Street was looking for. On any normal reading that is a strong quarter and a strong guide.

So you have to look at the call to understand why it sold off, because that is where the disappointment lived. A few things landed badly. The first was the AI number for July. They guided it to $16B. That is up more than 200% from a year ago, but the published Street estimate was $17.2B and the buyside had quietly pushed its own number up toward $18.5B. So a figure growing triple digits got treated as a miss, entirely because of where expectations had drifted.

The bigger one, and the one that really moved the stock, was the 2027 AI target. Broadcom left it at "more than $100B" instead of raising it. Going into the print the average buyside expectation for that number had stretched to something like $150B, with some people closer to $200B. When a company reiterates and the room is leaning hard for a raise, even a reiteration of a giant number reads as a letdown. Hock did say plainly that holding the number at $100B does not mean it has not gone up internally, only that he is choosing to stay conservative, but the market wanted the printed figure to move and it did not.

Then gross margin. They guided it down about 300 basis points to roughly 74%, and seeing margin head the wrong way unsettled people. And finally, on the call, management said out loud for the first time what the industry had been whispering for a while, which is that Google, its single biggest custom-chip customer, is going to start spreading some of that chip work over to MediaTek. Hearing it confirmed by the company, rather than as a rumor, gave the share-loss worry a real voice. Add the tone on top of all that. The call was a little less crisp than what people are used to from Hock, and a couple of the longtime bulls said as much afterward. When the delivery is slightly off on a quarter where the bar was sky high, it does not take much to tip sentiment.

Now, having sat with all of it, almost none of it actually worries me on the business, and a couple of pieces quietly make the story better. Start with the orders, because this is the number that tells you what is really happening and almost nobody led with. Broadcom booked more than $30B of new AI orders in the quarter while recognizing $10.8B of AI revenue in that same stretch. That is a book-to-bill near 2.8, meaning new orders came in at close to three times what they shipped. You do not get that in a business where demand is fading. The backlog jumped, and for the first time management is talking with confidence about 2028, where a quarter ago they would only stick their neck out to 2027. Demand is plainly not the issue.

So if the orders are that strong, why did 2027 not go up? This is the part worth slowing down on, because it is the whole misunderstanding in one place. That 2027 AI number is tied to deploying roughly 10 gigawatts of data center capacity, at something like $13B to $15B per gigawatt. What gates that figure is not whether Broadcom can make the chips, and it is not whether anyone wants them. It is how fast the customers can physically stand up gigawatts of compute, the power, the cooling, the racks, the whole build wrapped around the silicon. The bottleneck is the building, not the chip. That is why the number is not moving up yet even though they clearly think it ends higher, and it is why the 2028 confidence got pulled forward. The demand is already booked and stacking up behind a deployment schedule that can only go so fast.

The margin drop is just mix. Custom chips carry lower gross margins than the networking silicon does, and custom chips are becoming a bigger slice of the revenue as they ramp. That is margin moving because the mix is shifting, not because of pricing or competition, and the number that matters more, operating margin, is still guided around 67%. Lower gross margin with operating margin holding is what a healthy ramp looks like, not a business under pressure.

The MediaTek piece is real but bounded. In practice MediaTek takes the lower-end, lower-margin chip variants while Broadcom keeps the premium, high-end silicon, and the read I would agree with is that Broadcom hangs onto something like 80% of the custom-chip market it actually serves. There is one real bear here, Macquarie, which cut the stock specifically on the idea that Google insourcing plus the MediaTek shift eats into share in 2027 and 2028. That is the one view worth taking seriously, and it is a 2027-2028 question, not a this-quarter one. Nearly everyone else on the Street raised their price target.

And the customer list is getting deeper, not thinner, which is the actual answer to the share-loss fear. Beyond Google, Anthropic is layering in another 5 to 6 gigawatts of next-gen chips from 2027, on top of the gigawatt-plus it already runs. OpenAI has first silicon delivered and 1.3 gigawatts set for 2027 inside a much bigger 2029 commitment. Meta is ramping its own program, about 3 gigawatts through 2028. And there are two more customers behind $6B of orders, widely thought to be ByteDance and Fujitsu. On the money side, Google has committed something like $80B and there is a new vehicle with Apollo and Blackstone to fund 20 gigawatts through 2028. The financing to build all of this is already in place.

So the way I would leave it: this was a subpar showing against the standard Broadcom set for itself, and after a roughly 50% run since the last report, with the broad semi index up almost 96% in five months, the stock was always going to get measured against the optimistic version in people's heads rather than the printed estimates. It cleared the real bar and missed the imaginary one. But nothing in the quarter changes the AI semiconductor story, and the bookings and the 2028 visibility quietly improve it. The stuff that did the damage was idiosyncratic, timing on the AI number, mix on the margin, a forecast paced by how fast capacity can be built, a competition headline that was already understood, and a CEO having an off night on the mic. None of that is the business cracking. If anything, the distance between the order book and the revenue says the next couple of years are more locked in than they were three months ago. It just needs one clean beat to get going again.