Broadcom (AVGO) Deep Dive — Thursday, September 3, 2026

The $230 billion raise the market refuses to price — and why I'm holding the $342 flush all the way to the December 9 print.

30-Second TLDR

1) What Happened: A Beat, a Two-Year Raise, and a −4% Lesson in Maxed Expectations

Broadcom just delivered the strongest two-year demand signal in the AI silicon trade's history, and the market's only question was why the next 90 days aren't even stronger. For the record, the company: a $1.74 trillion market cap (≈$1.8T at Wednesday's close) built on custom AI accelerators (XPUs) for Google, Anthropic, OpenAI and Meta, Tomahawk-class data-center networking, and a VMware-anchored infrastructure-software annuity that is now ~30% of revenue at very high margin. Last night it printed fiscal Q3 (quarter ended August 2): adjusted EPS $3.32 vs $3.24 expected, revenue $29.59B vs $29.36B expected, up 86% year over year. GAAP net income more than tripled to $13.09B; non-GAAP net income rose 95% to $16.37B. AI semiconductor revenue hit $16.7B — up 221% year over year, up 54% sequentially, 56% of total revenue — against a ~$16B guide. Infrastructure software grew 29% to $8.75B, a hair below the $8.82B consensus.

Then the stock fell, for three compounding reasons. First, the Q4 revenue guide of $34.8B (+93% YoY) missed the $35.03B consensus by 0.7% — and Morgan Stanley had warned before the print that some holders were whispering $150B+ for FY27 AI revenue, so the company's raise to $115B registered as a disappointment against the whisper even as it raised from $100B. Second, Q4 AI guidance of $21.7B (+236% YoY) came with a consolidated gross-margin guide of ~73%, down from 78% a year ago — memory costs are now visible in the P&L (more below). Third, the market's reflex this week is to sell beats: Palo Alto (−9.3%) and MongoDB (−13.5%) on Tuesday, Broadcom after hours Wednesday, while paying up only for new evidence — Dell +15.8% Tuesday on customer breadth, Snowflake +21% today on an AI-driven raise.

Today's tape: the flush filled. AVGO opened at $356.66, traded down to $342.33 — inside the $330–348 limit zone I published before the print (the branch framework is here) — then recovered to ~$353, down 3.9%, its lowest session since April. The stock is now up only ~2% year to date against SOXX's +65%, and down ~23% over three months even as guidance moved higher. Macquarie upgraded to Outperform this morning with a $490 target (from $437), arguing Google insourcing risk "has played out and seems priced in"; 47 of 51 covering shops sit at buy or strong buy. That divergence — a near-unanimous sell-side and a stock at five-month lows — is the setup.

2) The FY28 Arithmetic: Six Customers, Secured Supply, and ~$30 of EPS That Actually Adds Up

The $230B FY28 number is not a hope — it is a customer-by-customer deployment schedule with the supply already locked, and the EPS bridge to >$30 is arithmetic, not narrative. On the call, Hock Tan named the demand: Anthropic becomes Broadcom's largest XPU customer in 2027, deploying 5 gigawatts of Google's TPU 8i (which Broadcom co-designs and supplies) next year with line of sight to 10 more in 2028 — anchored by the $35B first tranche closed in June. OpenAI deploys 1.3GW of its Jalapeno XPU in 2027 with >5GW of line of sight, its second-generation chip is taping out, and a third is in planning. Meta's MTIA moves to production shipments. And Google alone takes "tens of billions of dollars" of processors per year for the next several years. Six XPU customers, four driving the bulk.

The supply side is the part consensus hasn't priced. Tan's exact words: "In 2027, we have secured the supply to again double AI revenue to approximately $115 billion. Our demand actually exceeds this outlook, and we will work to improve supply… In 2028… we have line of sight for fiscal 2028 AI semiconductor revenue growth to again double to $230 billion. Here again, we have secured the supply to meet this outlook." Wafers, substrates and memory are secured for both years, and a Singapore substrate facility comes online in FY27 to relieve the packaging bottleneck. In a market where memory is in "extreme pricing conditions" (Nvidia's CFO, last week) and supply is the binding constraint everywhere else, Broadcom claims to have pre-bought its way out of it. If true, that is worth a premium, not a discount.

The EPS bridge, from the filed statements: nine-month FY26 revenue is $71.09B; add the $34.8B Q4 guide and FY26 lands at ~$105.9B, with AI at $58B (company guide, +186% YoY). FY27: $115B AI plus ~$50B of software and non-AI semis ≈ $165B. FY28: $230B AI plus ~$51B non-AI ≈ $281B. Q3's actual non-GAAP net margin was 55.3% (nine-month: 54.4%); holding ~53–55% through the memory squeeze yields ~$149–155B of non-GAAP net income on ~4.95B diluted shares — ~$30–31 of FY28 EPS, matching management's ">$30." The street's $25.86 implies ~$128B of net income, i.e. ~$240B of revenue, i.e. the street has quietly discounted the company's FY28 AI guide by ~$40B (17%) — and the stock trades on the street's number, not the company's.

What that means in multiples, at $353: 30.4x FY26E EPS (~$11.6, street $11.63), 18.1x FY27E (street $19.53), 13.6x FY28E on the street, and 11.7x on the company's own FY28. Cross-check: Nvidia at $227 trades at roughly 16x my FY28 base-case EPS (~$14, per yesterday's deep dive). My opinion, stated plainly: the faster-growing company with a high-margin software annuity attached is priced at a 25–30% discount to its larger peer on the same horizon. That is the mispricing. My 12-month weighted fair value is ~$390–400 (scenarios in Section 4).

3) The Balance Sheet Is Clean Today — the Canary Is in the Guarantee Language

The cash says the earnings are real: operating cash flow covered GAAP net income at 108%, free cash flow margin hit 46%, and Broadcom paid down $5.7B of debt in nine months. Q3 OCF was $14.2B (+98% YoY); capex was just $0.5B, leaving FCF of $13.7B (+95%), equal to 46% of revenue. Cash rose to $24.0B from $19.6B sequentially. Total debt fell from ~$65.1B last November to ~$59.4B. The $0.65 quarterly dividend (~$3.1B) is a rounding error against that flow. Contrast with Nvidia, where OCF ran at just 45% of net income last quarter with receivables ballooning: Broadcom's cash conversion is, today, the cleanest in the AI complex. Receivables did rise 92% since November (to $13.7B; DSO roughly 36 → 42 days) and inventory doubled (to $4.5B) — but both grew roughly in line with revenue, with inventory days flat. Watch, not alarm.

The canary is forward-looking, and the CFO put it on the record herself. Amie Thuener: "We are empowering two of our most strategic customers, the leading AI labs, to bridge the gap between their current cash flow and the significant upfront investments required for their businesses" — and Broadcom "might provide residual value guarantees that are contingent liabilities to the labs." Tan, on the same theme: "part of it is creating sources of financing." This is the same vendor-financing loop I documented at Nvidia — chipmakers underwriting their own customers' capex — except Broadcom's version is still mostly off balance sheet, running through an AI financing platform reported to involve Apollo and Blackstone, with the company reported earlier this year to have sought up to $100B of AI financing capacity. Translation for a numerate reader: the risk appears first as language, then as a commitments-and-contingencies line, then as receivables that stop turning.

The 10-Q audit (due ~mid-September, before the FOMC) is where readers have the edge. Three checks: (1) receivables growth versus revenue — DSO pushing above ~50 from 42 is flashing; (2) the commitments note — any first appearance of a residual-value guarantee or customer-financing line means downgrade the call to WATCH and halve it; (3) OCF as a share of net income — slipping below ~80% means the gap between earnings and cash is opening. This is the exact audit that surfaced the Nvidia receivables story before the market cared.

4) The Real Risks: Concentration, Memory Costs, and the December 9 Print

The bear case is real and specific: a 56%-of-revenue business concentrated in four customers, gross margin already sliding on HBM content, and a December 9 print where the FY27 bridge has to survive contact with supply timing. The concentration math is uncomfortable — AI is anchored to six XPU customers with four driving the bulk, so one deferred generation (an Anthropic 5GW slip, an OpenAI tape-out stall, Google moving its next TPU fully in-house — Macquarie argues that risk "has played out") makes FY27's $115B unfillable and resets the entire compounding story. The margin math is already visible: Q3 non-GAAP gross margin printed 75.0%, down ~210bp sequentially as XPU and HBM content rose, and Q4 is guided to ~73% against 78% a year ago; operating margin holds at 66% only through opex leverage. Memory pricing is the single most likely source of a miss against the >$30 FY28 path. And the options market is not sanguine: full-chain put/call is 0.66, but September-2027 expiries sit at 2.54 — big money is paying up for long-dated downside into the December 9 report. My read (opinion): that's rented certainty, and it means downside is priced while upside is not — but the signal deserves respect, which is why the position is half size.

Weighted fair value ≈ $393 — about 11% above $353 and ~14% above the $345 mid of today's fills. The kill at $322 sits above the bear target on purpose: it exits the trade before the bear case finishes playing out. $322 is roughly where the market is pricing FY27 AI below $100B — $15B under the company's guide — and below it, I am not arguing with the tape.

5) The Trade: Hold the Filled Flush, Arm the Breakout, Ignore the Dead Zone

Both published branches did their job; the discipline from here is refusing to trade the middle. The flush leg filled at $342–348 this morning on the gap-down. Between $348 and $385 the expected value is poor — the stock is neither cheap enough to add nor confirmed enough to chase — so the only two permitted actions are the ones already specified:

What would make me add early: a second flush under $348 without news (the $330–348 zone remains valid for the other half of the tranche), or the 10-Q arriving clean on all three audit checks. What would make me exit early: any guarantee line appearing in the commitments note — that converts a demand story into a credit story, and credit stories don't trade at 16x.

6) The Macro Turned Friendlier Today — and the Rest of the Book in One Paragraph Each

The macro headwind that has capped every AI multiple all summer just lost a third of its force before the open. Fed Governor Waller said he is inclined to hold rates at the September 15–16 meeting — "Give disinflation a chance. We can wait one meeting… Hiking 25 basis points, one meeting right now, is not going to bring the CPI down to 2%" — the first FOMC official to break with Chair Warsh's Jackson Hole posture. CME FedWatch hike odds collapsed ~15 points to 48.4%; the 10-year fell to ~4.74% (its 4.80% regime line now has clear air below); the VIX sits at 14.62; the Dow is up ~600 points for its best day in a month; gold at ~$4,540 (+2.8%) is the war premium trading alongside the risk-on tape. ISM Services printed 55.4, the strongest in six months, with prices-paid still pointing the wrong way. The remaining inputs before the FOMC are tomorrow's NFP (consensus ~+53k after July's −23k) and next week's CPI/PPI — a +120k print re-arms the hawkish case, so the book stays deliberately two-sided into September 16.

• XLRE contrarian long (EQ-XLRE-1): filled today at $43.84–44.00 (session low $43.84), now $44.20 — target $46; killed by NFP ≥ +120k or a decisive 10Y close ≥ 4.85%. Tomorrow is its binary.
• IWM short (EQ-IWM-1): $295.04, back inside the $293–297 zone, target $282, dead on a weekly close above $302. It fights the XLRE long by design.
• NVDA (EQ-NVDA-2): $227.34, inside the $222–236 no-trade zone; the $12.93B Hugging Face deal (its largest acquisition ever) is ecosystem breadth, not a trigger — the momentum tranche needs a weekly close above $236.54.
• SPY put hedge (EQ-VIXHEDGE-1): VIX at 14.62 keeps the ≤16 entry window wide open — Oct 16 ~$730 puts, premium ≤1%, covering the FOMC.
• BTC (CR-BTC-1): $81.3K (+5.3%) is above the $80,300 breakout trigger intraday — the add fires on the weekly close, not intraday. Don't chase.
• ETH (CR-ETH-1): survived its kill line — the Sept 2 close was $2,379, $29 above the $2,350 invalidation, after being $41 above at the last mark; now $2,512, twenty-three dollars below Target 1 ($2,535). Book half at target; run the rest toward $2,800 only on the BTC $86K path.
• The tape's verdict on breadth: Snowflake +21% and Dell +7% say the market still pays for new evidence; HPE −4.6% post-print says it still sells soft guides. Broadcom is being sold on the whisper, not the business.

Conclusion — My Highest-Conviction Take: You Are Being Paid to Wait Two Quarters for Proof That Is Already Scheduled

AVGO at ~$353 is the only mega-cap in the AI complex where the company has pre-announced two consecutive years of doubling, pre-bought the supply to deliver it, and converted 46 cents of every revenue dollar into free cash flow — and the market's response was to bid up 27-month downside puts. The honest bear case — four-customer concentration, a memory-cost margin squeeze already visible, and a financing loop that just put "contingent liabilities" in a CFO's mouth — is worth roughly 12x on the street's FY28 EPS. The honest bull case is 16x on the company's. The weighted answer is ~$393, and the trade is already on: hold the $342–348 fills, add only above $385 or into another no-news panic under $348, and let December 9 do the arguing. The single highest-value action between now and then costs nothing: read the commitments note in the 10-Q before the market does. That is where this thesis is proven, or where it quietly stops being an AI story and becomes a credit story.

What Would Prove Me Wrong (measurable, dated)

• Weekly close below $322 — the market is repricing FY27 AI below $100B; exit both branches regardless of the model.
• The December 9 print: Q4 AI revenue below the $21.7B guide, or Q1 FY27 guidance that annualizes FY27 AI under $100B — the bridge is broken; my target drops to ~$336 (13x street FY28 EPS) and the exit rule fires first.
• The 10-Q (mid-September): a residual-value-guarantee or customer-financing line in commitments, OR DSO above ~50, OR OCF below ~80% of net income — downgrade to WATCH, halve the position.
• Q4 non-GAAP operating margin below 64% (guide is ~66%) — the memory squeeze is structural, not mix; cut FY28 EPS toward street and the target ~10%.
• A named lab defers a generation (Anthropic 5GW slips to 2028, OpenAI gen-2 tape-out stalls) — FY27 $115B is unfillable; bear case activates.
• Macro override: two consecutive 10Y closes at or above 4.85% — the four AI-semi longs in the book compress 10–12% together, this one included, and the regime rules take precedence over any single-name thesis.

Appendix — Check the Work

Contents: A. Data snapshot with timestamps · B. The model, assumptions, and the load-bearing input · C. Open-book status and disclosed conflicts · D. Sources

A. Data snapshot (intraday, Thursday Sept 3, 2026, ~12:03 PM ET / 16:03 UTC)

Equities (Finnhub pull, 16:03 UTC): AVGO $352.98 −3.88% (open $356.66, low $342.33, prev close $367.24) — cross-checked vs Investing.com $352.51 (−4.01%, 12:09 ET) and the CNBC quote header ($352.35), agreement within 0.2%. NVDA $227.34 +1.31% (Investing $227.86, +1.54%). SNOW $370.64 +21.19%. MRVL $211.84 +2.60%. AMD $456.86 −0.04%. SPY $773.14 +1.04%; QQQ $717.59 +1.18%; IWM $295.04 +0.35%; XLRE $44.20 +1.07% (session low $43.84). Indexes (Investing.com board, 12:09 ET): S&P 500 7,742.26 +0.99%; Dow 53,643.96 +1.10% (~+582 pts); Nasdaq Composite 26,575.23 +1.36%. Also on that board: VIX 14.62 (−3.82%); 10Y 4.742% (−5.2bp); 30Y 5.226%; WTI $91.36 (+0.38%); Brent $95.57 (−0.06%); gold $4,539.70 (+2.83%); DXY 98.92 (−0.63%).

Crypto (Finnhub, 16:03 UTC; CoinGecko series): BTC $81,277 (+5.30%; 16:01 UTC tick $81,287 — above the $80,300 weekly-close trigger); ETH $2,512 (+5.25%; Sept 2 daily close ≈ $2,379 from the hourly series, above the $2,350 kill line). Fear & Greed 65 ("Greed"). AVGO fundamentals (Finnhub): market cap ~$2.0T at a stale mark — computed ~$1.74T at $353 on 4.937B non-GAAP diluted shares; P/E TTM 68.4x (GAAP, amortization-depressed); 52-week range $289.96 (Mar 30) – $495.00 (Jun 3); dividend yield 1.88%; beta 1.48; ROE 36.4%.

Macro: ISM Services 55.4 for August (from 54.1; forecasts 54.2–54.3), the strongest in six months, with TheStreet flagging prices-paid as inflation pressure. CME FedWatch September hike odds 48.4%, down ~15 points on Waller's remarks (provider readings had run 63–68% over the prior week). Fed funds target 3.50–3.75%; July CPI headline 3.7%, core 3.3%; Waller cited 3-month core PCE at 3.05%, down from 4.76% in February. NFP Friday 8:30 AM ET, consensus ~+53k (Dow Jones, via CNBC). FOMC September 15–16. ECB expected to hike September 9 (Reuters poll).

Insiders: no AVGO insider filings in the 30 days before the print (last activity July 14; before that, programmatic June sales at $377–401 — Samueli et al.). No pre-print red flag.

B. The model, assumptions, and the load-bearing input

Revenue build: FY25A $63.9B (AI ~$20.3B implied by the $58B/+186% FY26 guide). FY26E = 9M actual $71.089B + Q4 guide $34.8B ≈ $105.9B, AI $58B. FY27E = AI $115B + software ~$37B + non-AI semis ~$12.5B ≈ $165B. FY28E = AI $230B + non-AI ~$51B ≈ $281B. EPS: FY26E ~$11.6 (9M non-GAAP EPS $7.81 + Q4E ~$3.78 at the 66% op-margin guide; street $11.626). FY27E street $19.532 → 18.1x at $353. FY28E at 53–55% non-GAAP net margin on $281B ≈ $30–31; at the street's 54%-margin-equivalent, $25.86 → the ~$40B FY28 AI discount. Shares: ~4.94B diluted (per the release, 4.887B GAAP / 4.937B non-GAAP). Scenario targets: 16x × $30 = $480 (25%); 15x × $28 = $420 (50%); 12x × $21 = $252 (25%); probability-weighted ≈ $393.

Load-bearing input: non-GAAP net margin holds ≥ ~53% through FY28. The memory-cost squeeze has already taken gross margin from 77.1% (Q2) to 75.0% (Q3) to a ~73% Q4 guide; operating margin survives on opex leverage. If net margin instead slips to ~48% by FY28, EPS lands ~$27.3, base fair value drops to ~$360–380, the $430 target is unreachable, and the correct trade becomes "hold the fills to ~$400 and reassess." Secondary load-bearing input: every named deployment (Anthropic 5GW/10GW, OpenAI 1.3GW/>5GW, Meta MTIA, Google's "tens of billions per year") ships on schedule. Where this could be wrong: supply "secured" is management's claim — a substrate or HBM shortfall converts the guide's precision into a timing miss, which is precisely what the December 9 report will test.

C. Open-book status and disclosed conflicts

EQ-AVGO-1 flush leg FILLED today $342–348 (half size), breakout leg armed ($385), kill $322 weekly. EQ-NVDA-2: $227.34, inside the $222–236 dead zone — no action; momentum tranche needs a weekly close > $236.54. EQ-XLRE-1: filled today $43.84–44.00, target $46, killed by NFP ≥ +120k or decisive 10Y ≥ 4.85%. EQ-IWM-1 short: $295.04, inside $293–297, target $282, dead above $302 weekly. EQ-VIXHEDGE-1: window open (VIX 14.62 ≤ 16); SPY Oct-16 ~$730 puts, ≤1% premium, invalidation S&P record close above 7,817. CR-BTC-1: $81.3K, above the $80,300 trigger intraday — add fires on the weekly close only. CR-ETH-1: kill line survived (Sept 2 close $2,379); $23 below Target 1 $2,535 — book half at target. Not re-marked today: AON, FRO, ASML, MRVL ($211.84, inside its $205–222 zone; Oct 6 investor day), OBDC. Disclosed conflicts, deliberate: the put hedge pays in the scenario that kills the AVGO breakout leg and the NVDA momentum tranche; the XLRE long fights the IWM short and OBDC in the hawkish branch; four AI-semi longs (ASML, AVGO, MRVL, NVDA) compress together if the 10Y decisively clears 4.80–4.85%. The book is two-sided around September 16 on purpose.

D. Sources

• Broadcom Q3 FY2026 results statement (PRNewswire — the primary source; full statements) · full-text mirror read via StockTitan
• CNBC — Broadcom Q3 report (LSEG consensus, call quotes, GW deployments, $115B/$230B, EPS >$30 vs $25.86)
• Stocktwits via TradingView — Tan's exact "secured the supply / demand exceeds this outlook" quotes; Moorhead's unresolved-risk list; SOXX +65% YTD comparison
• SiliconANGLE — Thuener "bridge the gap" quote; guarantee language
• Investing.com — why AVGO slid (Morgan Stanley $150B+ whisper warning; market board used for VIX/10Y/indexes)
• 24/7 Wall St. — options skew (Sept-2027 P/C 2.54), 73% vs 78% GM guide, $35B Anthropic tranche, AI XPV platform (Apollo/Blackstone), Dec 9 print date, street EPS ($11.63/$19.53)
• Benzinga — "Broadcom's AI Boom Has A Memory Problem" (margin line vs record revenue)
• CNBC — Macquarie upgrade to Outperform, $490 target, "Google insourcing risk has played out"; 47/51 buys
• CNBC — Waller leans hold; FedWatch 48.4%; "Give disinflation a chance" · Federal Reserve — Waller speech text
• ISM — Services PMI 55.4% (August 2026) · Trading Economics — ISM Services 55.4 vs 54.3 forecast
• CNBC — Nvidia's $12.9B Hugging Face acquisition
• House archive: Nvidia deep dive (Sept 2) · Sept 2 close review (the branch orders) · Broadcom print framework (Sept 1)

Published by @dailyanalysts, September 3, 2026, mid-session (~12:30 PM ET). All prices are intraday marks and are labeled as such; weekly-close invalidations govern. This analysis is opinion on top of sourced facts; it is not personalized investment advice. Open positions disclosed: Broadcom long (flush leg, filled $342–348, half size), Nvidia long, IWM short, SPY put hedge, Aon short (avoid), Frontline long, ASML long, Marvell long, OBDC long, XLRE long (half size), BTC hold, ETH long. Options, shorts and levered positions carry substantial risk of loss. 中文版：Chinese edition.