In Broadcom’s earnings call, analysts kept pressing Hock Tan on gross margin, and he told them outright not to focus on that metric, but on operating margin instead. When a CEO actively asks the market to use a different yardstick to measure the company, it usually means the old one is starting to look bad. The recent move in $AVGOB has been following that old yardstick.
The reading on the old yardstick is this: for the quarter ended August 2, gross margin was 75%, next quarter guidance is 73%, and it was still 78% in the same period last year. The CFO said the reason is that custom XPU products now contain more and more memory, and that memory has to be bought from outside and packaged in. That part of the product goes through Broadcom’s books as revenue, but it dilutes margins. The more AI chips Broadcom sells, the thinner the margin gets.
Tan’s statement holds up in accounting terms. This quarter’s operating margin was actually higher than last year’s, because expenses did not increase at all, and R&D spending was even lower than a year ago. When revenue nearly doubles and expenses stay flat, leverage naturally appears. But that kind of leverage can only happen once. Next quarter’s operating margin guidance has already been cut to 66%, while AI revenue is still supposed to head toward $115 billion in 2027 and $230 billion in 2028. At that scale, R&D and capacity investment cannot stay suppressed forever, while gross margin is still trending lower.
The earnings report itself was actually solid. Revenue came in at $29.59 billion, about twice the level of a year earlier, and both revenue and EPS beat expectations. AI semiconductor revenue was $16.7 billion, up 221% year over year, with next-quarter guidance at $21.7 billion. The company also raised its 2027 AI target from the figure it gave last quarter. Even so, the stock fell more than 6% after hours. Most reports blamed that on total revenue guidance of $34.8 billion coming in slightly below sell-side estimates, but a gap of just a few billion dollars hardly explains such a large reaction.
What was re-priced was the structure of growth. The company now really has only one leg left, #AI ; non-AI semiconductors are up just 5% year over year and flat sequentially, and wireless is still a drag. Infrastructure software guidance for next quarter also ticked down slightly, as the growth from VMware’s shift to subscriptions has reached a plateau. Almost all of the incremental revenue next quarter will come from AI. Traditional businesses provide no cushion; if the AI curve slows, the entire income statement slows with it.
Tan said that in 2027 Broadcom’s biggest XPU customer will switch to Anthropic, with OpenAI second, and that the long-term Google agreement also amounts to shipments in the tens of billions of dollars per year. He added another point: next year’s supply is already locked in, demand exceeds the company’s guidance, and the bottleneck is now on the customer side, where it is still unclear when data centers will actually get powered on. He said the company gives guidance conservatively because once chips are shipped, they may not be installed into racks on time.
That shifts the risk elsewhere. Broadcom is no longer worried about orders or capacity; it is worried about other people’s construction projects. Whether AI revenue arrives on schedule now depends on the buildout progress and financing pace of a few customers, and those customers are highly concentrated.
That is where the disagreement lies. On September 3, Bernstein raised its target price to $575, arguing that the multi-year guidance itself matters more than a single quarter’s gross margin and that confirmed demand is more important than one gross margin print. On the same day, RBC kept a Neutral rating and a $400 target, saying that component supply and data-center readiness are not in Broadcom’s hands, and that on 2027 earnings the stock is about 30% more expensive than Nvidia.
The strongest argument on the other side is cash. This quarter free cash flow was $13.67 billion, nearly half of revenue, so even if customers’ construction projects are delayed by a year, Broadcom itself can still handle it.
I side more with RBC’s ranking of the risks. Confirmed demand is certainly a good thing, but the timing of delivery has been handed over to customers, and among those customers, two are still going through round after round of fundraising, with money coming from capital markets rather than operating cash flow. For $230 billion to be real, their capital expenditure plans have to be executed without a single year of cuts. Betting on Broadcom’s execution is one thing; betting on someone else’s cash flow is another. This time, that is what is really being sold.
What could most easily overturn my view is still gross margin. If next quarter’s actual gross margin comes in above the 73% guidance, that would mean memory cost pressure is not as rigid as feared, and Tan’s request to change the yardstick would make more sense. $AVGOB is currently at $359.87, up 1.64% over 24 hours, still below its pre-earnings level. In the two days after earnings, trading volume jumped to several times normal, then fell back again. Where gross margin lands next quarter is worth watching.
The reading on the old yardstick is this: for the quarter ended August 2, gross margin was 75%, next quarter guidance is 73%, and it was still 78% in the same period last year. The CFO said the reason is that custom XPU products now contain more and more memory, and that memory has to be bought from outside and packaged in. That part of the product goes through Broadcom’s books as revenue, but it dilutes margins. The more AI chips Broadcom sells, the thinner the margin gets.
Tan’s statement holds up in accounting terms. This quarter’s operating margin was actually higher than last year’s, because expenses did not increase at all, and R&D spending was even lower than a year ago. When revenue nearly doubles and expenses stay flat, leverage naturally appears. But that kind of leverage can only happen once. Next quarter’s operating margin guidance has already been cut to 66%, while AI revenue is still supposed to head toward $115 billion in 2027 and $230 billion in 2028. At that scale, R&D and capacity investment cannot stay suppressed forever, while gross margin is still trending lower.
The earnings report itself was actually solid. Revenue came in at $29.59 billion, about twice the level of a year earlier, and both revenue and EPS beat expectations. AI semiconductor revenue was $16.7 billion, up 221% year over year, with next-quarter guidance at $21.7 billion. The company also raised its 2027 AI target from the figure it gave last quarter. Even so, the stock fell more than 6% after hours. Most reports blamed that on total revenue guidance of $34.8 billion coming in slightly below sell-side estimates, but a gap of just a few billion dollars hardly explains such a large reaction.
What was re-priced was the structure of growth. The company now really has only one leg left, #AI ; non-AI semiconductors are up just 5% year over year and flat sequentially, and wireless is still a drag. Infrastructure software guidance for next quarter also ticked down slightly, as the growth from VMware’s shift to subscriptions has reached a plateau. Almost all of the incremental revenue next quarter will come from AI. Traditional businesses provide no cushion; if the AI curve slows, the entire income statement slows with it.
Tan said that in 2027 Broadcom’s biggest XPU customer will switch to Anthropic, with OpenAI second, and that the long-term Google agreement also amounts to shipments in the tens of billions of dollars per year. He added another point: next year’s supply is already locked in, demand exceeds the company’s guidance, and the bottleneck is now on the customer side, where it is still unclear when data centers will actually get powered on. He said the company gives guidance conservatively because once chips are shipped, they may not be installed into racks on time.
That shifts the risk elsewhere. Broadcom is no longer worried about orders or capacity; it is worried about other people’s construction projects. Whether AI revenue arrives on schedule now depends on the buildout progress and financing pace of a few customers, and those customers are highly concentrated.
That is where the disagreement lies. On September 3, Bernstein raised its target price to $575, arguing that the multi-year guidance itself matters more than a single quarter’s gross margin and that confirmed demand is more important than one gross margin print. On the same day, RBC kept a Neutral rating and a $400 target, saying that component supply and data-center readiness are not in Broadcom’s hands, and that on 2027 earnings the stock is about 30% more expensive than Nvidia.
The strongest argument on the other side is cash. This quarter free cash flow was $13.67 billion, nearly half of revenue, so even if customers’ construction projects are delayed by a year, Broadcom itself can still handle it.
I side more with RBC’s ranking of the risks. Confirmed demand is certainly a good thing, but the timing of delivery has been handed over to customers, and among those customers, two are still going through round after round of fundraising, with money coming from capital markets rather than operating cash flow. For $230 billion to be real, their capital expenditure plans have to be executed without a single year of cuts. Betting on Broadcom’s execution is one thing; betting on someone else’s cash flow is another. This time, that is what is really being sold.
What could most easily overturn my view is still gross margin. If next quarter’s actual gross margin comes in above the 73% guidance, that would mean memory cost pressure is not as rigid as feared, and Tan’s request to change the yardstick would make more sense. $AVGOB is currently at $359.87, up 1.64% over 24 hours, still below its pre-earnings level. In the two days after earnings, trading volume jumped to several times normal, then fell back again. Where gross margin lands next quarter is worth watching.