The number investment committees were staring at Thursday morning was not $16.7 billion. It was 73%.
Broadcom’s Q3 AI semiconductor revenue of $16.7 billion grew 221% year-over-year and 54% quarter-over-quarter. Total revenue came in at $29.6 billion, up 86% year-over-year, and adjusted earnings per share of $3.32 topped estimates of $3.24. Every top-line metric beat. The stock fell more than 6% by mid-morning Thursday despite beating on earnings the night before.
The reason is the margin trajectory, and it is not subtle. Broadcom’s gross margin fell 210 basis points sequentially, and Q4 consolidated gross margin is guided to 73%, down from 78% a year earlier. That step-down is a mechanical result of custom accelerators with higher memory content growing as a share of revenue, so Broadcom is converting spectacular top-line growth into thinner incremental profit.
Why Professionals Are Focused Here
Institutional investors have spent two years paying a premium for AI revenue growth. The implicit assumption was that more AI revenue meant more profit dollars at roughly stable margins. Broadcom’s Q4 guide breaks that assumption in a visible way. Gross margin declined sequentially to 75% in Q3 and is expected to fall further to 73% in Q4 due to the increasing mix of XPUs with higher memory content, which dilutes margins. The bull case for custom silicon was always that lock-in with hyperscalers created pricing power. The margin compression suggests the cost structure of serving those customers is moving faster than pricing can absorb it.
Broadcom said the financing structures it has in place are mainly for two customers, including Anthropic and OpenAI, while other customers are funding deployments themselves. That is an unusual arrangement: a chip supplier effectively helping finance a portion of its customers’ capacity build. It confirms the strategic importance of those relationships while raising a quiet question about who holds leverage in the long run.
Management’s Counter
CEO Hock Tan told analysts they should focus on operating margin, which benefits from significant operating leverage. Q3 operating income was a record $20.1 billion, up 92% from a year ago, and even with the gross margin decline, operating margin increased 240 basis points year-over-year to 67.9%. The case is coherent: scale absorbs operating costs faster than gross margin erodes. But portfolio managers who underwrite a stock at 30-plus times earnings are sensitive to the direction of gross margin, not just its level. A multi-quarter trend toward 73% changes the terminal value math.
Broadcom forecasts fiscal 2026 AI revenue of $58 billion, with long-term outlooks of approximately $115 billion in fiscal 2027 and $230 billion in fiscal 2028. If those numbers land, the gross margin question becomes secondary. If execution slips, or if one of the six hyperscaler customers shifts spending, the margin cushion that once protected investors will not be there.
What Investors Are Missing
The read-across to Nvidia is being misread. Nvidia gained about 1% Thursday as Broadcom’s longer-term AI outlook was read as a demand positive, while AMD slipped about 2% amid worries about margin read-across. That divergence is the signal. Nvidia’s gross margin has stayed above 70% in recent quarters and has proven durable through multiple product cycles. Broadcom’s compression is specific to the custom XPU model, where the customer co-designs the chip and therefore captures more of the value. The market is beginning to price that difference. Custom silicon is not a better business than merchant silicon at scale. It may be a larger one. Those are not the same thing.
Stocks to Watch
- Broadcom (AVGO): The core question. The bear case is clear: quarterly guidance missed a high bar, valuation remains demanding, and much of the growth depends on a few massive customers. The bull case requires trusting that operating leverage holds as memory content rises.
- Nvidia (NVDA): Thursday’s gain is the market’s verdict that merchant GPU margins are structurally superior to custom XPU margins. Worth watching whether that relative preference firms up over Q4.
- Alphabet (GOOGL): Broadcom cited commitments from hyperscaler customers alongside AI model builders such as Anthropic and OpenAI. Google’s TPU roadmap and its willingness to keep co-designing with Broadcom is central to whether the fiscal 2027 target is real.
- Marvell Technology (MRVL): The other pure-play custom silicon supplier. If the market is re-rating Broadcom’s margin profile downward, Marvell faces the same scrutiny on its AI ASIC pipeline before it has even reported comparable scale.
