Broadcom Beat, Raised, and Still Sold Off: Why ‘Good’ AI Earnings Aren’t Good Enough Anymore

Broadcom’s fiscal Q4 print had all the ingredients of a “good” AI quarter: a clean beat on revenue and EPS, better-than-feared margins, and guidance that cleared the bar. And yet, the stock still slid hard after the call. The message from the tape is the same one investors got earlier in the week from Oracle: AI is no longer an “everyone wins” trade, and the market is now grading on a curve where “beat-and-raise” is just the entry fee. For AVGO, the culprit wasn’t demand. It was the mix, the margin trajectory, and a valuation that doesn’t tolerate any hint of gravity.

Start with the numbers. Broadcom reported Q4 revenue of $18.015B versus roughly $17.46–$17.50B expected, and adjusted EPS of $1.95 versus $1.87 expected. The beat was supported by strong profitability: adjusted EBITDA was $12.218B (above expectations) with EBITDA margin around 68% (about 100 bps ahead of consensus). Free cash flow was also substantial at $7.5B, or about 41% of revenue, reinforcing that this is still an extremely cash-generative franchise even as it leans harder into AI and continues digesting VMware. In a vacuum, this is exactly the kind of quarter that usually gets rewarded.

Guidance, on the surface, was also the opposite of a miss. For fiscal Q1, Broadcom guided revenue to about $19.1B versus Street expectations around $18.3–$18.4B. Management also guided adjusted EBITDA margin to roughly 67% of revenue. That’s a very strong profitability level in absolute terms, and it implies continued top-line momentum into the January quarter. The company sweetened the shareholder-return story with a 10% dividend increase to $0.65 per share for fiscal 2026. Again: not the profile of a company waving a caution flag.

So why the selloff? Because the call was a reminder that the fastest-growing part of Broadcom’s business is also lower margin, and it’s about to become a bigger slice of the pie. Management explicitly framed it: AI revenue carries lower gross margins than the rest of the business, and margins will “start to deteriorate” as the AI mix rises and as system sales ramp, particularly when Broadcom is effectively passing through more third-party components that dilute percentage margins. They tried to offset the concern with the more important concept—gross profit dollars and operating profit dollars should still grow, and operating leverage can still work even if margin percentage compresses. The market’s response was basically: understood, but at forty-plus times earnings, we’re not paying premium multiple for “margin down, dollars up” unless the upside is unmistakable.

The AI commentary itself was undeniably strong. Management said AI semiconductor revenue is expected to double year-over-year in Q1 to $8.2B, driven by custom AI accelerators (ASICs/XPUs) and Ethernet AI switches. They also reiterated the scale of demand visibility: AI backlog in excess of $73B, expected to be delivered over the next 18 months—nearly half of the company’s total backlog. On customers, the call provided both reassurance and a new set of questions. Broadcom identified the previously referenced $10B “new qualified customer” as Anthropic and said it placed an additional $11B order for delivery in late 2026. Management also cited a fifth AI customer with a $1B order for late 2026. That expands the roster beyond the usual names (Meta, ByteDance, Google), which matters because “AI concentration risk” is one of the first things investors attack when a stock is priced like a royalty stream.

But the market wanted either (1) more customer breadth, sooner, or (2) cleaner margin optics, now. Instead, it got a nuanced answer: there is a multi-year OpenAI partnership, but management doesn’t expect much revenue from OpenAI in 2026, with a ramp beginning in 2027 and extending through 2029. In other words, some of the “dream scenario” demand is real, but it’s back-half weighted. That’s fine for the business, but it complicates the stock when investors are already jumpy about the AI spending cycle, and when the nearest-term margin narrative is “down 100 bps sequentially in Q1 due to mix.”

The backdrop matters here. After a year where AI winners could stumble and still levitate, the market is tightening the screws. The combination of an AI-heavy tape, high index valuations, and a growing belief that “the easy part is over” means there’s less patience for anything that looks like peak margins, peak excitement, or peak multiple. Oracle’s post-earnings drop earlier in the week contributed to that tone: even if the specific issues differ (Oracle’s centered on capex and cash burn versus Broadcom’s margin mix), the broader read-through is that investors are shifting from “AI demand exists” to “AI demand exists, but who captures the economics?”

That brings us to valuation. Broadcom is being valued like a premier AI compounder, not a diversified semi + software platform with moving parts. Around “forty-plus times earnings” (as one market participant put it), the market is implicitly assuming sustained AI growth, continued success in networking (including the Tomahawk 6 switch ramp and strong AI switch backlog), and steady execution in Infrastructure Software (VMware Cloud Foundation adoption and bookings strength). It’s also why the “valuation gap” conversation shows up: if NVDA is the AI standard-bearer, AVGO is often treated like the highest-quality “second derivative” with a custom silicon angle. That’s flattering—until the market decides it would like the multiple to come with fewer caveats.

What should investors watch from here?

First, the margin path versus the growth path. Management is telling you percentage margins can compress while dollar profits still rise. The stock will trade on whether that’s credible quarter after quarter, especially as system sales increase and pass-through content grows. If gross margin percentage slides faster than expected, the multiple will compress even if revenue is strong.

Second, conversion of backlog into reported revenue and the durability of the customer pipeline. The $73B backlog number is big enough to anchor the bull case, but investors will want evidence that it broadens beyond a handful of counterparties and that delivery timing doesn’t slip.

Third, the OpenAI timeline. “Not much in 2026, ramps in 2027” is honest guidance, but it means the market will constantly debate whether the stock is pricing 2027 in 2025. Any incremental clarity on scope, volume, or timing can matter disproportionately.

Fourth, non-AI stabilization. Management flagged non-AI semis as modestly growing and software as steady, but bears will keep probing whether “everything that isn’t AI” is stalling. Broadcom doesn’t need non-AI to be a rocket ship, but it can’t become dead weight.

Fifth, supply chain and advanced packaging. Management highlighted investments (including Singapore advanced packaging capacity) to support AI demand. Investors will watch whether Broadcom can scale delivery without adding cost friction that worsens the very margin narrative the market is punishing.

Bottom line: Broadcom delivered a beat and raised the near-term top line, but it also confirmed that the next phase of the AI buildout is less about whether demand exists and more about who captures the economics. If you want the stock to levitate from here, you probably need either re-acceleration in AI upside (more customers, more orders, more near-term visibility) or proof that margins won’t decay as quickly as the market fears. In the current tape, “phenomenal demand but slightly less phenomenal margins” is still good business—just not always good enough for a stock priced for perfection.