Broadcom Into the Q3 Print: Trade the Sentence, Not the Coin Flip

@dailyanalysts · Tuesday, September 1, 2026 · intraday marks as of 1:23 PM ET (17:23 UTC). Broadcom (AVGO $369.42) reports fiscal Q3 Wednesday, September 2, after the close.

30-Second TLDR

The call: No pre-print position. If Tan RAISES the FY27 AI guide above $100B — buy above $385, target $455, invalidation weekly close < $355. If he HOLDS ">$100B" again — buy the $330–348 flush, target $430, invalidation weekly close < $322.

Why now: AVGO is down ~25% from its June peak while FY27 EPS consensus ROSE from $18.22 to $19.53. The derate already happened; the stock sits exactly on its 200-day (~$369) and outperformed semis today.

The disagreement: Consensus braces for a June redux — flat guide equals punishment. My opinion: a flat guide at $370 is pre-paid pain; the flush is the entry, not the exit, because December is when the raise was always going to come.

The level that changes everything: a weekly close below $322. There, the market is repricing FY27 below $100 billion and every long thesis in this name is dead.

Tomorrow is a referendum on one sentence, not an earnings report

The only line that matters Wednesday night is whether CEO Hock Tan moves the fiscal-2027 AI revenue anchor off "in excess of $100 billion." Everything else on the tape is choreography.

Recall June 3: Broadcom beat adjusted EPS ($2.44 vs $2.40), missed revenue by two-tenths of a percent ($22.19B vs $22.27B), guided Q3 revenue to $29.4B against $28.53B consensus, and said Q3 AI revenue would grow over 200% year-over-year to $16.0 billion. The stock still fell more than 12% the next day (roughly 15% at the intraday worst). CNBC's headline blamed "weak software sales, unchanged AI chip forecast" — but the software miss was a 2% blemish ($7.18B vs $7.32B); the unchanged forecast was the sin. A $250-billion-plus market-value haircut, per the FT's own framing of that day, for one reiterated number.

The setup into tomorrow is different in one crucial way: price. AVGO trades at $369.42, roughly 25% below the early-June intraday peak of $495 (record close $481.57 on June 2), pinned exactly on its 200-day moving average (~$369) — and today, with the Nasdaq 100 down 1.2% on a bond selloff, AVGO is down just 0.25% and holding that line into the print. The options market prices the September 4 weekly $370 straddle for an ~8% move with a 2:1 call/put skew. For context on that skew: MarketChameleon's 12-quarter history says the options market overestimated AVGO's post-earnings move 42% of the time.

The quarter is already known — consensus IS the company's guide

There is no "beat" to buy tomorrow; there is only the bridge to 2027. Consensus for Q3 is $29.43–29.44 billion in revenue with adjusted EPS clustering $3.16–$3.24 (moomoo's panel shows GAAP $2.55) — and the $29.4B revenue figure is not a forecast, it is the company's own guidance, verbatim from the "Third Quarter Fiscal Year 2026 Business Outlook" section of the June 3 release. The quarter prints on the company's number unless something broke in August.

Three numbers actually matter tomorrow night:

1. The FY27 AI sentence. "In excess of $100 billion" — raised, held, or (tail risk) cut. Tan has held it twice since first articulating it at the December 2025 Q4 print.
2. The Q4 guide. The FY26 AI bridge: $10.8B shipped in Q2 (+143% y/y), $16.0B guided for Q3, and Q1 grew 106% y/y — FY26 lands around $53B if Q4 exits near $18B. The Q4 revenue and AI guide is the first measurable test of the FY27 ramp.
3. The Q4 margin guide. Q3 carries a non-GAAP operating income guide of ~67% of revenue. In June, Tan said Broadcom will sell "chips only" rather than complete integrated AI systems — that trade lowers revenue per deal and shifts mix. If the Q4 margin guide slips below ~66%, the AI volume story is being bought with margin, and the multiple has a new problem.

The arithmetic the guide has to bridge: FY27 ">$100B" requires an average above $25B per quarter — a ~39% step-up from the likely Q4 exit rate (~$18B, my estimate). Morgan Stanley models ~$120B ($30B/quarter). That gap is why "hold" reads as a negative at any price, and why December — the print where Tan set the $100B anchor in the first place — is the structural raise event, not Wednesday.

The derate already happened — while the estimates went the other way

You are not being asked to pay June's price for tomorrow's risk. The stock is down ~25% from the peak and 4.9% over the past month while the S&P rose; it is up just 7.4% YTD against the S&P's 12.5%. Meanwhile the 44-analyst panel reads 7 Strong Buy / 37 Buy / 4 Hold / 0 Sell, with an average target of $526 — 42% above spot — and FY27 EPS consensus has DRIFTED UP from $18.22 to $19.53 over ninety days, with three upward revisions in the last 30 and zero cuts. Price down a quarter, earnings up 7%: that divergence is the entire opportunity.

The multiple map (my arithmetic on the $19.53 consensus): $370 = 18.9x FY27; the June record close of $481.57 = 24.6x a lower estimate; $348 = 17.8x; $330 = 16.9x; $430 = 22x; $455 = 23.3x. BNP Paribas Exane's street-high $675 is ~34.5x. The Fool's YCharts lens (31x forward on next-twelve-month EPS) says the same thing from a different denominator: near the low end of this year's range. Morgan Stanley's Joseph Moore, in the July 14 note that framed the whole debate, said he was "surprised by the year-to-date underperformance" — AVGO +13% YTD at that point versus the semiconductor ETF's +90%.

Two primary-source footnotes worth more than they appear. First, from the Q2 cash flow statement: Broadcom's repurchase program bought $7.85B of stock in the quarter ended February 1 and just $600M in the quarter ended May 3 — the company's own bid largely left the tape right as the stock peaked. Second, insider activity: no Form 4 sales filed in the 30 days before this print. The June cluster (co-founder Henry Samueli's scheduled sales at $377–388 plus gifts; other executives at $379–401 in early July) is routine in size and above today's price. Neither is a red flag; both belong in the file.

The bear case is real — it's Google, not the quarter

The reason this stock cannot re-rate is that its growth story is one customer's sourcing decisions, and that customer is visibly building alternatives. Morgan Stanley attributes roughly $80 billion of its ~$120 billion FY27 AI estimate to Google-TPU-related revenue. Three fronts, all recent:

• MediaTek — The Information reported in March that MediaTek is working with Alphabet on the next generation of custom processors. Moore's answer (July 14): Broadcom retains "roughly 80%" of Google's TPU business long-term; MediaTek is "real, but not disruptive" — its CoWoS packaging capacity for 2nm TPU production is unproven at volume, EMIB packaging is unproven at Google's scale, Broadcom has locked HBM supply, and MediaTek's own stated long-term target is just 15–20% of the market.
• Marvell — On August 19, Marvell disclosed a Google warrant for up to 58.97 million shares at $206.58 (about $12.2B), unlocking against purchase targets through fiscal 2033, covering products that "attach to the [TPU] ecosystem": AI inference accelerators, storage and network interface controllers. Broadcom fell ~5% that day. Note the strike: with MRVL at $208.47, the warrant is already in the money.
• NVIDIA's NVHBM — yesterday, Counterpoint's Neil Shah laid out the sharpest structural point of the three: NVHBM moves the memory controller into the HBM base die behind a proprietary die-to-die link, letting NVIDIA sell fabric, rack architecture, software and the memory subsystem to companies building custom accelerators — without designing the accelerator. Amazon's Annapurna Labs (Trainium4) is the first disclosed customer and MediaTek has joined NVLink Fusion. Per Shah, Broadcom is "the only major custom chip design company without a direct NVIDIA partnership" and "lacks a comparable base-die offering."

The concentration math: at 80% retention (the bull case) the story holds; a slide to 60% share removes roughly $16 billion of FY27 revenue — about 8% of the growth story — and the targets here come down ~10%. The macro leg compounds it: hyperscalers are spending 102% of cloud revenue on capex; UBS cut its 2027 AI industry revenue outlook to $1.3 trillion with hyperscaler capex growth slowing to 25% in 2027 and 6% in 2028, and trimmed its AVGO target to $470 from $485, seeing "modest" Q3 upside and doubting this report "will sway the overarching debate." And the tape pays nothing for patience right now: the 10-year yield is at its highest since January 2025, oil is above $92 after the U.S. struck Iran over the Hormuz attacks, and the market prices roughly two-in-three odds of a September 16 hike.

My opinion: Morgan Stanley's 80% is the right 2027 base case. The operational hurdles are real, and Google itself expanded its Broadcom deal in April — alongside Anthropic — while Meta extended through 2029 on a one-gigawatt commitment, OpenAI's first custom chip is Broadcom-built, and Apple's expanded partnership is valued above $30 billion. Six core customers, per Tan in June. But the direction of share at the edges is one-way; only the pace is uncertain. That is exactly why the flush-buy below carries a $322 invalidation instead of a comfortable one.

The trade: let the binary resolve, then pay for certainty

My math says owning the print itself is negative expected value; the edge is in the branches. The pre-print EV model (mine, assumptions in Appendix B): ~30% chance of a raise (+8–15%), ~50% hold with a solid Q4 guide (−5 to −12%), ~20% hold with a soft guide (−12 to −18%), under 5% cut. Weighted, that is roughly −4% at $369 — with the 2:1 call skew adding fuel to the unwind branch. The ~8% straddle prices the same binary fairly (options overestimated the move 42% of the last 12 quarters). There is no edge in the coin flip; there is edge in the aftermath.

This explicitly updates my August call on AVGO. That structure (preferred entry above $385 after the guide; pre-print $355–375 for small event size only; target $455; flat guide = invalidation) treated a hold as a walk-away. The discipline stands — no coin-flip ownership at $369. But the flat-guide outcome has changed character: at $330–348 the market would be paying 17–18x for a rising $19.53 estimate with the December catalyst 14 weeks out. That is an entry, not an exit. If you bought the June flush and are underwater, the honest instruction is the same: Branch B is the add zone, December is the proof date, and $322 weekly is the line where I am wrong.

Bull, base, bear

• Bull, ~30%: FY27 AI raised (say ">$110B") and Q4 revenue guide ≥ $33B with AI ≥ $18B. Day one $400–425 — the call skew amplifies a raise — and $455 within weeks. Measurable trigger: the explicit raise sentence in the outlook.
• Base, ~50%: Q3 lands on the guide ($29.4B, ~$3.20, AI ~$16B), FY27 held, Q4 solid (revenue ~$32–34B, AI $17–19B, margin ≥ 66%). Day one −5 to −12% ($325–352), then a grind back toward $430 into December. The June precedent says the first day is the worst day.
• Bear, ~20%: FY27 held AND Q4 AI guide below ~$16.5B, or a second software miss, or a margin guide under 65%. Price $305–325 — and the flush is NOT buyable; stand down until December. An outright cut below $100B is a <5% tail and means exit everything the same night.

What would prove me wrong

• A weekly close below $322 in any branch. The market is then pricing FY27 materially below $100B; both branches are dead.
• A Q4 AI guide below ~$16.5B. The FY26 bridge is cracking; the flush is not a value entry.
• Any FY27 AI number below $100B. Thesis dead on arrival, regardless of price.
• A quantified Google/MediaTek TPU share announcement. The 80% retention is the load-bearing competitive assumption; a 60% number removes ~$16B of FY27 revenue and both targets fall ~10%.
• Bull-side falsifier: a raise that still closes the week below $355. Good news being sold in a 4.8%-yield tape is its own information; Branch A exits.
• Macro override: a decisive 10-year break above 4.80% into the September 16 FOMC compresses both targets roughly 10–12% on the same earnings (23x → 21x), whatever the guide says.

What to do

• Tonight: nothing. Do not own the coin flip. The August structure allowed small event size at $355–375 — we are at $369.42 — but my EV math says pass, and the call now reflects that.
• Tomorrow after the close: open the release and go straight to the Business Outlook section — that is where the June sin and any redemption live. Find the FY27 sentence, then the Q4 revenue/AI/margin guide, then check infrastructure software against ~$7.3B.
• Branch A: buy the open above $385; weekly close below $355 is the exit; $455 target.
• Branch B: work limits at $330–348; if the tape gaps under $322, cancel and reassess at the December print.
• The rest of the story, one sentence each: Marvell is the clean winner of Google's diversification and owns the mirror-image concentration; NVIDIA just extended its moat into the custom-chip world without building a custom chip; Alphabet is the actual decision-maker for AVGO's FY27 — its sourcing language matters more than Tan's; Micron's memory supercycle is the other side of the same capex trade.
• Position conflicts, disclosed: this call pays in the risk-on branch — the same side as my NVDA long and the opposite side of my hawkish core (IWM short, SPY puts, which expire worthless if AVGO rips). That two-sided structure is deliberate.

Appendix — checking the work

A. Data snapshot (intraday, Sept 1, 2026, 17:23 UTC / 1:23 PM ET, via market-data handler): AVGO $369.42 (−0.25%, day range $362.00–371.40, prior close $370.34; 52-week $287.17–495.00); SPY $761.70 (−0.70%); QQQ $708.15 (−1.20%); NVDA $218.24 (−1.15%); MRVL $208.47 (−1.51%); GOOGL $335.46 (−1.15%). Macro same day: 10-year yield at its highest since January 2025; WTI at its highest since July after U.S. strikes on Iran following the Hormuz tanker attacks; oil benchmarks above $92. Technical reference set (BigGo, Aug 26): 20-day SMA $393.63, 50-day $387.46, 200-day $369.22; support $356.50, resistance $407.50; April golden cross (50>200) technically intact.

B. Models and arithmetic. (1) Pre-print EV (my opinion): 30% × +11.5% (raise) + 50% × −8% (hold, solid) + 20% × −15% (hold, soft) + 5% × −20% (cut tail, weighted at half) ≈ −4.3%. Load-bearing input: P(raise) = 30%. At P(raise) = 50%, EV flips to roughly +1% and pre-print becomes buyable — that is the dial, and everything else in the model is secondary. Tan's pattern of setting the next-year anchor at December prints (not Q3) is why I set it at 30. (2) Multiple map: price ÷ FY27E $19.53: $330 = 16.9x, $348 = 17.8x, $370 = 18.9x, $430 = 22.0x, $455 = 23.3x, $481.57 = 24.6x, BNP $675 = 34.5x. (3) FY27 bridge: >$100B ÷ 4 = >$25B/qtr average; Q3 actual guide $16.0B; assumed Q4 exit ~$18B (my estimate, not guidance); step-up from exit rate to FY27 average ≈ +39%; Morgan Stanley $120B implies $30B/qtr. (4) Q2 FY26 primary-source table (June 3 release): revenue $22,187M (+48%); semiconductors $15,009M (+79%), infrastructure software $7,178M (+9%); GAAP EPS $1.91, non-GAAP $2.44; adjusted EBITDA $15,244M (69% of revenue); FCF $10,262M (46% of revenue); Q3 guide $29.4B revenue / ~67% non-GAAP operating income / ~68% adjusted EBITDA; cash $19.6B against ~$64.9B total debt, interest expense $776M/qtr; buybacks $600M in Q2 after $7,850M in Q1 (cash flow statement); dividend $0.65/qtr. Hock Tan, in the release: "in Q3 we expect semiconductor revenue from AI to grow over 200 percent year-over-year to $16.0 billion." Bookings context (Q2 call, via TIKR/CNBC): over $30B in AI bookings against $10.8B shipped, visibility to 2028; "the bookings that are coming are not for immediate delivery."

C. Sources. Primary: Broadcom Q2 FY26 results release (June 3, 2026) — outlook and reconciliation sections; SEC EDGAR, AVGO Form 4 filings. Coverage and analysis: CNBC on the June 3 print; TIKR on the selloff and bookings; Motley Fool on post-earnings history; moomoo consensus panel; BigGo on the $60B backlog and technicals. Competition: Morgan Stanley/MediaTek note (July 14); CNBC on the Marvell-Google warrant (Aug 19); Benzinga/Counterpoint on NVHBM (Sept 1). Options and targets: MarketRebellion IV report (Aug 31); MarketChameleon earnings-move history; Investopedia on the UBS trim (Aug 31); KuCoin flash on UBS 2027 AI outlook (Aug 24); 24/7 Wall St on the analyst panel and BNP $675 (Sept 1); 24/7 Wall St on hyperscaler capex (Aug 22). Macro: CNBC on U.S. strikes on Iran; CNBC on the 10-year yield; CNBC on WTI. Customer/deal context: CNBC on the April Google/Anthropic expansion; CNBC on Meta's 1GW commitment.

Opinions are marked as such throughout; everything else is sourced above. Intraday prices carry timestamps in Appendix A. If the print changes any input, the branches reprice — the levels, not the narrative, govern.