AVGO earnings comprehensive analysis

AVGO Q3 FY2026 — Executive Report

Reported 02 Sep 2026 after the close. Stock: $354.91 (as of 13:23 ET today).

The one-line version: an operationally spectacular quarter with the largest forward guidance any semiconductor company has ever put in print — and the stock is down anyway because Q4 revenue guidance missed by a hair and gross margin is structurally deteriorating.

---

1. The print vs. the reaction

Metric	Q3'26 actual	Consensus	Surprise
Adj. EPS	$3.32 (+96.5% YoY)	$3.16	+5.1%
Revenue	$29.59B (+85.5% YoY)	$29.36B	+0.5%
Q4 rev guide	$34.80B (+93% YoY)	$35.03B	−0.7%

That 0.7% guidance shortfall is the entire story of the sell-off. AVGO is −3.4% over the two sessions since the print, and −15.1% from its 13 Aug high of 417.82.

Segment detail: AI semis $16.7B (+221% YoY, +54% QoQ), of which XPUs were 73%. AI networking +150%+. Non-AI semis $4.2B (+5%, still flat/stagnant). Infrastructure software $8.8B (+29%).

---

2. Fundamentals — genuinely excellent, and the operating leverage is real

Revenue nearly doubled YoY; the last bar is company guidance, not actual.

Metric	Q3'26	Q3'25
Operating margin (GAAP)	53.9%	36.9%
EBITDA margin	61.4%	50.7%
FCF	$13.7B (46.2% of rev)	$7.0B (44.0%)
Net income	$13.1B	$4.1B
Net debt / EBITDA	0.69x	1.69x

Two things I want to flag that most summaries will miss:

1. Deleveraging is dramatic. Net debt fell from $53.5B to $35.4B YoY; they paid down $5.6B of long-term debt in Q3 plus $1.5B after quarter-end. Net debt/EBITDA under 0.7x. This balance sheet is no longer a risk factor — it's now dry powder.
2. R&D intensity halved. R&D fell from 19.1% of revenue to 9.78%. In dollars that's roughly $3.06B → $2.90B — i.e. R&D spend went slightly down in absolute terms while revenue nearly doubled. Management frames this as "operating leverage." For a company promising three MTIA generations, three OpenAI XPU generations and successive TPU nodes, flat R&D dollars is a legitimate question, not a virtue. Nobody on the call asked it.

Margin mechanics (the actual bear point): Q4 consolidated gross margin guided to ~73% vs 78% a year ago — a 500bp YoY compression, driven by rising HBM/memory content inside XPUs. Management's answer was blunt: "stop focusing on gross margin… look at operating margin." Q4 operating margin guided to 66%, flat YoY. That defence works only as long as opex stays flat — see point 2 above. Note also: GAAP gross margin per the filings is 69.1%, not the 75% non-GAAP figure cited on the call.

---

3. Earnings call — the substance

Hock Tan did something CEOs almost never do: guided two full fiscal years of a single revenue line.

• FY26 AI revenue raised to $58B (from $56B), +186% YoY.
• FY27 AI revenue ≈ $115B — "we have secured the supply."
• FY28 AI revenue ≈ $230B — "we have secured the supply to meet this outlook."
• "On target to exceed $30 EPS in FY2028."
• $350B of AI semiconductors to be shipped to six customers over two years.

Customer detail:

• Google: long-term agreement, "multi tens of billions of dollars of TPUs annually." Now shipping TPU v8i in production, ahead of the MediaTek v8t.
• Anthropic: 1GW Ironwood in 2026 → 5GW of TPU v8i in 2027 → incremental 10GW in 2028. Becomes Broadcom's largest XPU customer in 2027.
• OpenAI: "Jalapeno" shipped; 1.3GW in 2027, >5GW in 2028; third-generation XPU already in development. Claimed to beat Grace Blackwell Ultra on perf/watt "at half the cost."
• Meta: three MTIA generations through 2027, 3GW line of sight.
• Content economics: $20–30B of Broadcom content per gigawatt, described as stable across generations (chip ASPs rise, but so does power per chip).

The three things that should worry you:

1. Vendor financing of customers. The "XPV platform" with Apollo and Blackstone — first $35B tranche closed in June for Anthropic. Broadcom provides "modest residual value guarantees… contingent liabilities we view as low risk." Melius asked directly about maximum off-balance-sheet exposure (~$29B disclosed in the last 10-Q) and got a non-answer: "we don't have anything to announce today." Tan's justification — comparing OpenAI and Anthropic to "two geniuses in the middle of Mongolia" who need help getting to college — is charming and is not a credit analysis. A chip company underwriting its own customers' demand is a circularity risk, full stop.
2. Bottlenecks are not silicon. Land, power and shell, plus substrates and HBM. Tan explicitly said the 30GW of aggregate 2027–28 demand will not all deploy in those fiscal years — the $115B/$230B numbers are "judged" down from actual demand. So the guidance is conservative on demand but hostage to construction schedules Broadcom does not control.
3. Concentration. Six customers; four of them "just simply going to be huge." Two of those four are pre-profit AI labs that Broadcom is helping to finance.

---

4. Technicals (daily)

Reading	Value
Price	354.91
SMA(50)	383.60 — price below
SMA(200)	369.82 — price below
RSI(14)	36.55
03 Sep volume	60.2M vs ~21.3M 10-day avg (2.8x)
03 Sep range	low 342.33, close 357.16

The tape is not confirming the bull case. This was not a one-day earnings dip — AVGO has been rolling over since mid-August, lost the 200-day before the print, and the post-earnings session was a 2.8x-volume down day. That's distribution. RSI in the mid-30s is weak but not washed out. Bulls will point to the 03 Sep intraday reversal off 342.33 (closed 14.8 points off the low) as an early tell; it is one candle.

Levels that matter: 342.33 (post-earnings low / first support), 369.82 (200-day — now overhead resistance), 383.60 (50-day). Reclaiming 370 on volume is the first thing that would falsify the distribution read.

Relative strength is the most damning single data point. Against same-sector, same-market-cap peers, AVGO's quarterly relative performance percentile has collapsed from ~46 in mid-June to 10.7–21.4 through early September. It sat at 10.7 the day before earnings. This stock has been underperforming its own peer group for two months straight while the AI trade broadly worked.

---

5. Valuation — expensive on trailing, cheap on management's own 2028 math

AVGO trades at 2.3–2.8x the semiconductor industry median on every multiple.

So: 45x trailing earnings, or 11.8x FY2028 guidance. The entire investment case reduces to whether you believe the FY28 EPS number. Market cap $1.747T. Dividend yield 0.65%, roughly in line with the sector median (0.49%) and irrelevant to the thesis.

---

6. Alternative data & positioning

• Short volume exploded. 03 Sep: 10.9M shares vs 1.5–2.5M on typical pre-earnings days. — roughly double the normal short share of volume. Some is hedging, but the magnitude is a real shift in bearish activity.
• Retail flipped from seller to buyer. Retail participation jumped from ~0.6% of volume to 2.33% on 03 Sep, and retail net-flow sentiment went from −2 (28 Aug) to +1 → +2. Classic pattern: retail buying an earnings dip while short volume doubles. Take from that what you will; it is not a bullish confirmation signal.
• Options. ATM IV ~35% for the 18 Sep and Oct expiries — post-event crush, no residual stress priced in. But look at where the open interest sits: 13.8K contracts at the $400 call (18 Sep) and 20.2K at the $410 call (16 Oct), both now deeply out of the money with deltas of 0.07 and 0.16. That is a large block of pre-earnings bullish positioning now badly offside — and a source of overhead supply as those hedges unwind.
• Insiders: only selling. Since June: Henry Samueli sold $250M (24 Jun), Mark Brazeal sold ~$32M across five tranches, plus small director sales. Zero open-market purchases apart from a $374K option-related acquisition by Harry You. The Brazeal sales (25,000 shares repeatedly) look programmatic; the $250M Samueli sale does not need to mean anything at his ownership level, but there is no insider buying to offset it.
• Dark pool data is stale — the feed for AVGO ends the week of 27 Jul ($44.6M), so I can't tell you anything credible about institutional off-exchange flow around this print. I'd rather say that than guess.

---

7. Sell-side response

14 actions since 01 Aug. Post-print: 9 targets raised, 4 cut, 1 upgrade (Macquarie, Neutral → Outperform, $490). Mean of the 12 targets published 03 Sep is $528, i.e. ~49% above spot.

Note the dispersion — this is not a consensus view. RBC sits at $400 (Sector Perform, unchanged twice), DA Davidson cut to $350 (Neutral) and TD Cowen cut to $475, while Cantor and Rosenblatt went to $600. A $350–$600 target range on a $1.75T company means the sell side fundamentally disagrees about whether the FY28 numbers are real.

---

Bottom line

The quarter was outstanding and the two-year guidance is extraordinary. But three facts sit uncomfortably against the bull case, and you should not skip past them: gross margin is compressing 500bp YoY with flat R&D dollars defending a multi-generation roadmap; Broadcom is now partially financing the customers whose demand it is guiding on; and the stock has been in the bottom quintile of its peer group for two months, below both the 50- and 200-day, with short volume doubling into the print. "Cheap at 11.8x FY28" is only true if you underwrite management's FY28 number — and management explicitly said it will not update that guidance quarterly, which conveniently removes the feedback loop.

I can't tell you what to do here and this isn't trading advice — position sizing and thesis are yours. Not trading advice.

Want me to create an alert for AVGO reclaiming its 200-day SMA on volume, compare AVGO's performance against NVDA, AMD and MRVL since the print, or scan the semiconductor sector for stocks that beat on earnings but broke below their 200-day SMA?