Broadcom Posted Its Best Quarter Ever — The Stock Is Still 29% Off Its High. Here's the Math Behind the Gap.

Broadcom's fiscal Q3 2026, reported Sep 2, 2026, is the most lopsided beat-and-fell in the company's modern history. Net revenue landed at $29.59 billion — up 86% year-over-year and above the $29.44 billion consensus. Non-GAAP EPS of $3.32 cleared the $3.16 estimate by ~5%. AI semiconductor revenue alone hit $16.7 billion, +221% year-over-year and +54% sequentially, and now contributes roughly 56% of consolidated sales. Operating income reached a record $20.1 billion on a non-GAAP basis, with non-GAAP operating margin of 67.9% — itself a record — and free cash flow of $13.7 billion.

Hock Tan's forward outlook made the print feel even more outsized. Q4 AI semiconductor revenue is guided to $21.7 billion (+236% YoY), above the $21.34 billion Street consensus, and full-year FY26 AI revenue was raised to $58 billion from the prior $56 billion view — a +186% year-over-year clip. Tan put formal, multi-year numbers on the table: $115 billion of AI revenue in FY27 and $230 billion in FY28, a four-fold expansion that, if delivered, would re-rate the entire semiconductor supply chain. Anthropic was confirmed on track to become Broadcom's largest XPU customer in 2027 and 2028, with a 10-gigawatt commitment; OpenAI's Jalapeño program is on its planned deployment schedule with 5 gigawatts behind it.

The print that should have re-rated the stock instead triggered a ~6.5% drop on Sep 2 and a continued slide through the end of the month, taking Broadcom to roughly $352 — about 28.7% below the $495 52-week high set on Jun 3, 2026. The culprit is mechanical, not fundamental: Q4 total-revenue guidance of $34.8 billion missed the $35.03 billion Street consensus by ~$230 million. Q4 AI semiconductor revenue beat consensus by ~$360 million, so the miss came entirely from the non-AI side — software, networking and the VMware carry, which together represent ~$13.1 billion of the Q4 outlook.

The second anchor was gross-margin compression. Broadcom guided to a Q4 non-GAAP gross margin of ~73%, down from ~75% in Q3, and Q4 non-GAAP operating margin of ~66%, roughly flat year-on-year. The compression is a structural consequence of the AI mix: HBM, advanced packaging and TSMC N3/N2 wafers carry higher cost-of-revenue than legacy networking and software, and the XPU programs are still ramping. Investors had already priced this once in late 2025 — when Broadcom warned that AI-mix dilution would pressure margins through the year — and the renewed focus on it during this print explains the cooling reaction despite the AI beat.

The post-earnings selloff has pushed Broadcom's forward P/E to roughly 19.4x against a trailing P/E near 43.8x — a forward PEG around 0.4, the cheapest the stock has looked relative to its growth trajectory since the early AI buildout. With FY26 revenue now tracking toward $108–110 billion and FY27 AI revenue alone guided to $115 billion, the multiple implies the market is pricing in something close to flat earnings growth into FY27 — an outcome that is hard to reconcile with a forward order book anchored by Anthropic at 10 GW, OpenAI at 5 GW, Google's seventh-generation TPU relationship, and Meta's 2nm program extended through 2029. By comparison, TSMC — the foundry partner actually building these chips — trades at a forward P/E near 21.4x, Alphabet at ~22.8x, and Micron, the second-source HBM supplier, at ~7.0x.

What makes this print qualitatively different from prior Broadcom beats is that Tan gave the market formal, multi-year numbers rather than directional language. FY26 AI revenue of $58 billion (+186% YoY), FY27 of $115 billion (+98% YoY) and FY28 of $230 billion (+100% YoY) represent an implied 4x revenue ramp over two years — a number that has no analog in the merchant-GPU world. The underlying demand is concentrated but contracted: Tan confirmed six XPU customers — Google, Meta, Anthropic, OpenAI, ByteDance and Fujitsu — with three of them now locked in through 2028.

The concentration risk cuts both ways. Top three customers now absorb roughly 73% of Broadcom's AI revenue run-rate; if any of them retrenches, the $230B FY28 number collapses by tens of billions. But the supply-side commitments — Tan said Broadcom has secured HBM and foundry capacity through 2027 — are similarly hard to replicate, and represent a moat that pure-play merchants like NVIDIA cannot fully match. The mix of contracted demand and locked-in capacity is the structural reason the long-run number is anchored rather than aspirational.

Every dollar of Broadcom's $16.7 billion Q3 AI semiconductor revenue translates directly into demand at TSMC for advanced-node wafers, at SK Hynix and Micron for HBM stacks, and at packaging partners for2.5D/3D integration. The leak from Broadcom's backlog into its supply chain is the second-derivative story of this quarter — and the read for those suppliers is materially more constructive than AVGO's price action suggests.

The disconnect between the fundamentals and the tape reduces to a disagreement about customer durability and margin recovery. The bull case is mechanical: forward P/E of ~19.4x against an order book that locks $115 billion in FY27 and $230 billion in FY28, with three of the largest hyperscalers and AI labs as contracted counterparties. If Tan's $230B FY28 AI number is even approximately right, Broadcom is trading at a fraction of its growth-adjusted multiple — and the recent 29% drawdown is the kind of setup that has marked every prior cycle low since 2018.

The bear case is also mechanical: gross margin compression of 210 bps in a single quarter, with another 200 bps guided for Q4, exposes the fact that AI revenue is not margin-accretive at the mix level — it dilutes. Customer concentration near 73% of AI revenue means any one of Anthropic, Google or Meta trimming capex in2027 would knock $20–40 billion off the FY28 number overnight. And the 29% drawdown from $495 is not, by itself, a valuation floor — Broadcom fell more than 30% from its 2024 high before re-rating, and the Q4 guide miss is a fresh reason for systematic sellers to keep pressing. The forward PEG of ~0.4 is the asymmetric setup; the gross-margin glide path is the question that decides it.