Broadcom's AI Revenue Grew 221% — and It's Still the "Frustrating" AI Laggard

① THE FILTER — what we screened out, what we kept

We scanned 40+ analyst actions on AVGO after its Sep 2 Q3 FY2026 print, the results, and the segment filings.

We cut: the "why couldn't it match Nvidia" hand-wringing.
We kept the hard stuff:

• Q3 FY2026 (reported Sep 2): revenue $$29.59B (+85.5% YoY)**, gross margin 74%, operating margin **54.3%**, net income$$13.09B, EPS $$2.68, **free cash flow$$13.67B. AI revenue $16.7B (+221% YoY), guided to double annually through 2028.

• Yet the stock fell ~5% — a beat-and-fade like Nvidia/Marvell, and Broadcom has been "a frustrating stretch" (up just ~6% YTD vs. the chip sector's ~60%).

• Consensus Strong Buy / Moderate Buy (35–49 analysts). Avg target ~$$500–533 (+40–49% upside)**, high *$$600, low $350 (a new street-low from DA Davidson).

📊 BULL vs BEAR — the analyst split

Post-earnings targets split (Cantor →$$600, BMO $$575 up; DA Davidson →$$350, Truist $$520, TD Cowen →$475 down; UBS downgraded to Hold). The book is bullish (89% Buy) but the reaction shows the "great quarter, high bar" tension. Implied upside to consensus is still ~40%.

② CORE LOGIC — the one-page thesis & the expectation gap

The thesis in one line: Broadcom is the #1 custom-AI-silicon house (ahead of Marvell) — designing the chips for Google, Meta, ByteDance and OpenAI — plus a VMware software cash machine, yet it trades like the AI trade's afterthought.

What the market is really betting on (the expectation gap):

Broadcom's AI business is booming (+221%) and it makes the custom XPUs that hyperscalers use to reduce Nvidia dependence — arguably the best "anti-Nvidia" AI play. But the stock has lagged the whole chip sector all year and fell again on guidance. The expectation gap: bulls see a 40%+ discount to fair value on a company with 54% operating margins and a doubling AI ramp; the market keeps treating it as "not Nvidia." The debate is whether the custom-ASIC + VMware combination finally gets its re-rating.

• Bull case: Custom AI silicon (Google TPU, Meta MTIA, OpenAI inference chip, ByteDance) is the fastest-growing, stickiest slice of AI compute; VMware adds ~42% high-margin recurring software; 54% operating margins and $13.7B quarterly FCF. At ~20x forward with AI doubling to 2028, the ~40% target upside looks compelling.

• Bear case: Custom-chip revenue is lumpy and timing-dependent; expectations are elevated; VMware integration/pricing draws scrutiny; and the stock has repeatedly disappointed despite great numbers ("frustrating stretch").

Edge vs. the crowd: Broadcom is the "own the anti-Nvidia trade + software moat" name. Pair it with Marvell (the #2 custom-ASIC house, also faded) and Nvidia (the GPU king that also fell): all three prove that this earnings season, AI hardware's bar is set at "flawless." Broadcom offers the biggest target-vs-price gap of the trio.

③ ACTION SIGNALS — dual watch

A. Catalyst / research window (dates to circle)

• 🔴 Q4 FY2026 earnings — early December 2026. Watch AI/custom-silicon revenue + the "double through 2028" trajectory.

• 🟡 New custom-XPU customer wins (OpenAI, Anthropic, others) — the biggest re-rating lever.

• 🟡 VMware software revenue + margins — the recurring-cash story.

• 🟢 Nvidia / Marvell results — the AI-silicon competitive reads.

B. Earnings-preview watch (what "good" vs "bad" looks like)

⚠️ Expectation note: Broadcom keeps posting monster numbers and getting sold — the bar is brutal. Judge it on the custom-ASIC ramp + VMware cash, and recognize the ~40% target gap exists precisely because the market keeps doubting it.

④ VALUE CHAIN & FOCUS NAMES

Upstream / suppliers

• Fabrication: TSMC; the hyperscalers whose architectures Broadcom turns into silicon

Broadcom's engines

• 🧩 Custom AI silicon (XPU/ASIC, ~58% semis) — Google TPU, Meta MTIA, OpenAI, ByteDance; the AI growth engine

• 🌐 Networking / wireless / storage chips — the broad semi franchise

• 💾 Infrastructure software (~42%) — VMware, CA, Symantec; recurring cash

Downstream / customers & competition

• Customers: Google, Meta, ByteDance, OpenAI, Apple (wireless)

• Competitors: Nvidia (GPUs), Marvell (custom ASICs)

Focus names to track alongside AVGO

• Marvell (MRVL): the #2 custom-ASIC rival — same "beat-and-fade" story.

• Nvidia (NVDA): the GPU king Broadcom's custom chips compete against.

• Google / Meta / OpenAI: the anchor XPU customers.

Sources (free/public): stockanalysis.com/AVGO · MarketBeat AVGO price targets · Broadcom results coverage · Wikipedia. Figures as reported by sources, as of Sep 7, 2026.
🤖 Auto-compiled by AI from free public information. For research/education only — not investment advice.