Financebotresearch desk研究台

AVGO › analyze

AVGO · Analyze

ACCUMULATE Semiconductors

A portfolio-specific passage was removed from the public build.

1. What This Updates

The baseline was written June 12, 2026, three trading days after the Q2 FY2026 print (June 3) — the same quarter this pass still relies on as the latest reported data. No new quarterly print has landed since the baseline. Q3 FY2026 results are due Wednesday, September 2, 2026 — six days from this analysis (Broadcom IR). Q3 guide standing at print time: revenue ~$29.4B, non-GAAP operating margin ~67%, AI semiconductor revenue growing >200% YoY, FY2026 AI semiconductor revenue $56B (unchanged), FY2027 AI revenue guide ">$100B" (reiterated).

Event list since June 12 (what actually happened, not what the calendar implies): 1. Sector-wide AI-capex de-rating. SOX fell ~28.6% peak (6/22) to trough (7/29); AVGO round-tripped $495 high → as low as the mid-$350s → back to ~$371 today. Price is essentially flat (-2.7%) since baseline despite QQQ running well ahead over the same window. 2. Google/Marvell TPU deal disclosed 8/19 (signed 7/29) — Google can buy up to $12.2B of Marvell stock tied to a custom-silicon agreement spanning inference accelerators, storage/network/memory controllers and near-memory compute attached to the TPU ecosystem. AVGO fell ~5% same day; MRVL popped ~8-14%. 3. CFO transition, effective June 12, 2026 (the baseline's own publish date) — Amie Thuener (ex-Alphabet VP/Corporate Controller & Chief Accounting Officer) succeeded Kirsten Spears, who stays on as advisor through ~March 2027. Pre-announced in April; orderly succession. 4. A fourth hyperscaler XPU customer substantively confirmed. CEO Hock Tan said in August that a fourth hyperscaler had placed >$10B in AI-rack orders; widely reported as OpenAI, though a Broadcom exec joked on CNBC the PO hadn't technically landed on his desk yet. Baseline had this as an unconfirmed rumor. 5. New off-balance-sheet credit exposure surfaced. Reporting (BofA, Seeking Alpha, ZeroHedge, Yahoo Finance) describes Broadcom lending its investment-grade credit standing to senior tranches of SPV debt (Apollo/Blackstone-arranged, ~$35B closed June 2026, another ~$30B+ reportedly being marketed) financing Anthropic's chip leases on Google TPUs. Broadcom CDS reportedly widened on the disclosure. This is new, real, and not yet verified against the 10-Q — flagged as an open item below, not adjudicated as a settled fact. 6. Nvidia's blowout Q2 print (8/27, today) triggered a broad chip-sector rally; AVGO +4.7% intraday alongside it. Marvell also beat on data-center revenue the same day — the two headline events landed on the same news cycle. 7. Institutional base grew modestly across BlackRock (+3.2%), Vanguard, State Street (+2.9%), FMR (+16.3%); insider activity remained routine 10b5-1 selling (no cluster, no buys of size).

2. The Delta Ledger

# Claim (baseline) Type Status Evidence
1 "19.7x forward P/E... extremely cheap" (PEG 0.57) Price 🔄 SUPERSEDED Yahoo's epsForward ($19.50) is the FY2027 consensus, not the imminent FY2026 figure. FY2026 (epsCurrentYear, $11.63) is corroborated bottom-up: Q1 FY26 non-GAAP EPS $2.05 + Q2 $2.44 actual = $4.49 H1; guided Q3 op-margin math implies ~$3.30-3.50 Q3, leaving a Q4 that lands full-year near $11.6-11.9 — matching Yahoo within rounding. True near-term (FY2026, closes in ~2 months) non-GAAP P/E is ~32x, not 19.7x → [[pitfall-vendor-forward-eps-is-the-wrong-fiscal-year]] (same trap that hit ANET/AMD/TSM this week). This also resolves the baseline's own Tension 1 (20x fwd P/E vs 46x EV/FCF) — the gap was partly a data error, not only an amortization/SBC artifact.
2 Fair value: Bear $200-250 / Base $380-480 / Bull $550-700 Price 🔄 SUPERSEDED Re-derived below with the corrected multiple basis and TTM data. Base case narrows and shifts modestly.
3 Entry zone $360-420, strong buy <$360 Price ✅ CARRIED Price has traded inside this exact band for 2.5 months — $356 to $430 — repeatedly touching both edges without breaking either. The zone has been market-tested, which is a stronger form of validation than when it was set.
4 Trim $520-550 (dollar) / later rewritten "26.7x fwd" Price 🔁 REFRESHED Re-set to 25x fwd (FY2027 basis, ≈$487) — modestly below the prior 26.7x (which tracked the raw analyst-mean target) to price in the two new risk vectors (#8, #11) without abandoning the growth case.
5 ROIC ex-goodwill ~47% (FY2025) State ⏳ UNTESTED No new annual print since FY2025; TTM figures are directionally consistent (op income growing faster than invested capital) but a precise TTM ROIC recompute wasn't corroborated against a second source this pass. Carry forward, don't re-assert as new fact.
6 Net Debt/EBITDA 1.4x (FY2025), "manageable" State 🔁 REFRESHED TTM: net debt ~$45.3B (debt $64.9B − cash $19.6B) / TTM EBITDA ~$42.1B ≈ 1.08x — improved, comfortably under the 3x break trigger. EBITDA growth is outrunning the dollar debt balance.
7 "By FY2027, net debt could be under $20B" Trend 📉 DRIFTED Dollar net debt fell only ~$3.7B in the ten months since FY2025 year-end ($49.0B → $45.3B) — an annualized pace that reaches ~$20B closer to FY2032 than FY2027 at this rate. The ratio is improving fast (EBITDA growth), but the dollar claim as literally written doesn't hold up. Not thesis-breaking (leverage ratio is what the break trigger actually tests), but the specific magnitude was too optimistic.
8 SBC = 28% of FCF (FY2025), "a meaningful drag"; watch for <20% in FY2027 Trend 🔁 REFRESHED TTM SBC/FCF = $8.79B / $32.76B = 26.8% — a small improvement, still well above the 20% threshold the baseline itself set as the test for "buybacks create real per-share accretion." Open item carried forward unresolved.
9 FCF 3yr CAGR +18.2% (FY22-25) Trend ✅ CARRIED No new annual print; the underlying figure is unchanged and no new data contradicts the trajectory.
10 TTM/run-rate FCF trajectory (baseline estimated FY2026E $38-45B) Trend 🔁 REFRESHED Actual TTM FCF (Q3 FY25 + Q4 FY25 + Q1 FY26 + Q2 FY26, summed from quarterly statements, not read off one column) = $32.76B, up +21.7% from FY2025's $26.91B annual figure — and this TTM window still carries the two weakest AI-ramp quarters (Q3/Q4 FY25) rather than the two strongest (Q3/Q4 FY26, still to report). The baseline's FY2026E range looks achievable to conservative on current trajectory.
11 Revenue 3yr CAGR +24.4%; "48% YoY" TTM growth context Trend 🔁 REFRESHED TTM revenue $75.46B vs FY2025 $63.89B = +18.1% sequential-window growth, +48% YoY on the latest reported quarter — growth is accelerating into FY2026, not decelerating. Corroborated: TTM revenue sum from quarterly statements ($75,465M) matches Yahoo's totalRevenue field exactly.
12 Dividend CAGR +16.6% (FY22-25) Trend ✅ CARRIED No new annual data; unchanged.
13 Analyst consensus 91.7% Buy+, zero Sell (June 2026) State 🔁 REFRESHED Current averageAnalystRating "1.3 — Strong Buy," 46 analysts, mean target $525.97 (+41.6% from spot). Two downgrades since baseline (Macquarie Outperform→Neutral 6/4; Erste Group Buy→Hold 7/7) — both explainable as post-Q2 valuation calls, not fundamental breaks. Consensus direction intact.
14 CEO Hock Tan: routine 10b5-1 sales, no alarm State ✅ CARRIED Pattern holds — continued scheduled sales at $325-401 through the period, consistent cadence with prior years. New, not in baseline: director Henry Samueli (co-founder) sold ~$250M in June across a pre-set program plus a ~337K-share gift — large in dollar terms but consistent with his multi-year selling cadence (similar-sized sales recur roughly quarterly back to 2024). Not a new signal.
15 (implicit) CFO stable State 🆕 NEW CFO transition effective 6/12/26 — Amie Thuener (ex-Alphabet CAO) replaces retiring Kirsten Spears, who advises through ~March 2027. Pre-announced, orderly. Mildly favorable: incoming CFO's prior seat was inside Google, AVGO's largest and most contested hyperscaler relationship.
16 Institutional positioning stable-to-growing State ✅ CARRIED BlackRock +3.2%, State Street +2.9%, FMR +16.3%, Vanguard flat-to-up. Broad continued accumulation across the passive/active base.
17 XPU design lock-in / ~70% custom-AI-accelerator share (Structural, moat source 1) Structural 📉 DRIFTED The clearest test result this pass. Google — the most-cited XPU relationship in the baseline — signed Marvell into the TPU ecosystem for inference accelerators, storage/network/memory controllers and near-memory compute (disclosed 8/19). Broadcom retains the core ASIC design, advanced packaging, HBM integration, SerDes and networking/switching — the highest-value BOM components — so this is dual-sourcing of attach silicon, not a core-ASIC replacement. But it is the first documented instance of the baseline's own Scenario A ("Marvell takes a hyperscaler") beginning to happen, where the baseline's own text called this "possible only for NEW programs, not replacements of existing ones" — Google's TPU program is not new. Name the level at which this breaks the moat claim outright: Marvell winning the core TPU ASIC (not attach silicon) at Google, or a second hyperscaler following the same playbook.
18 AI Networking Silicon (Tomahawk) as a second lock-in layer (moat source 2) Structural ✅ CARRIED No contrary evidence surfaced; Marvell's Google deal explicitly does not include core switching/networking silicon, which stays with Broadcom.
19 VMware software stabilizer, 77% op margin, recurring ARR Structural ⏳ UNTESTED No new quarter reported since Q2 FY26 (already in baseline). Nothing this pass confirms or contradicts continuation.
20 Process leadership / TSMC relationship advantage Structural ✅ CARRIED No contrary evidence; not tested directly this pass, no reason to doubt.
21 Scenario C: AI capex pause is "the real risk," 40-60% AI-revenue downside if it fires Structural/risk ✅ CARRIED, tension noted The sector-wide SOX de-rating (-28.6% peak-to-trough) shows the market pricing this risk more aggressively than in June, even though nothing in AVGO's own disclosed numbers confirms capex is actually slowing — Q2 AI revenue still guided/tracked toward the $56B FY26 number. This is a Price-row force wearing a Structural costume: sentiment on the risk moved; the evidence for the risk firing did not.
22 Scenario D: Nvidia total-cost undercut unlikely near-term Structural ✅ CARRIED No contrary evidence.
23 Break trigger: AI semi revenue <$10B for 2 consecutive quarters Judgment ⏳ UNTESTED Cannot test before the Sep 2 print; carried forward as the sharpest near-term falsification test on the table.
24 Break trigger: Google or Meta ends XPU partnership Judgment 📉 DRIFTED, not fired Google added a second supplier for attach silicon rather than ending the Broadcom relationship — the literal trigger has not fired, but the direction it warned against (erosion of exclusivity) has begun. See row 17.
25 4th hyperscaler in "late-stage discussions," unconfirmed Judgment 🔁 REFRESHED Substantively confirmed by CEO commentary (>$10B in AI-rack orders), reportedly OpenAI; one executive's on-air joke leaves the customer's name technically unconfirmed. Resolves toward the bull case.
26 (new) Off-balance-sheet SPV credit exposure to Anthropic — 🆕 NEW, flagged UNTESTED Multiple secondary-source reports (BofA research note relayed via 24/7 Wall St, ZeroHedge, Seeking Alpha, Yahoo Finance) describe Broadcom's credit standing being lent to senior tranches of ~$35B+ SPV debt financing Anthropic's chip leases, with CDS spiking on the disclosure. Not yet corroborated against the 10-Q or an 8-K — no primary source read this pass. This is exactly the kind of claim Phase 2 says should not flip a status on a single vendor-adjacent source. Carried as an open item, not as a settled risk.

Ledger count: 26 claims — 8 carried (5, 9, 12, 14 [as-carried], 16, 18, 20, 22) · 7 refreshed (6, 8, 10, 11, 13, 4, 25) · 4 drifted (7, 17, 21, 24) · 2 superseded (1, 2) · 0 retracted · 3 untested (5, 19, 23) · 2 new (15, 26).

3. How the Close Calls Were Decided

Row 17 (XPU lock-in) — the central judgment of this pass. The Fundamentals/Sentiment forces (Google-Marvell deal, real and disclosed via SEC filing) point one way; the Moat force (Broadcom retains the highest-value BOM components — core ASIC, packaging, HBM, SerDes, switching) points the other. These are not contradictory once decomposed: Google is diversifying the periphery of its TPU stack while Broadcom keeps the center. The baseline's own stress test named this exact scenario and called it low-probability for existing programs — the new evidence shows it happening at the edge of an existing program, not the core. That is a genuine erosion of the "near-monopoly" framing (worth a full status downgrade to DRIFTED) but not yet a Structural break (would require the core ASIC role itself to move) — hence DRIFTED rather than SUPERSEDED, with the specific level named that would force the next downgrade.

Row 21 (AI capex pause risk) — sentiment vs. evidence. The Sentiment force (SOX -28.6%, AVGO round-tripping $495→mid-$350s→$371) says the market is pricing a much more severe capex-slowdown scenario than in June. The Fundamentals force (revenue accelerating to +48% YoY, TTM FCF +21.7%, guidance unchanged) says nothing in AVGO's own numbers confirms that slowdown yet. Per the framework's own instruction, a market-price move is a Price-row force, not Structural evidence — it was recorded against the price/fair-value claims (row 2), not used to downgrade the Structural moat claims.

Row 1 (wrong-year forward P/E) — single vendor field vs. three corroborations. Before treating this as SUPERSEDED, three independent checks were required and obtained: (a) bottom-up quarterly math from actual Q1/Q2 non-GAAP EPS plus guided Q3 margins lands within rounding of Yahoo's epsCurrentYear; (b) the analyst mean target ($525.97) divides almost exactly by epsForward ($19.50) at ~27x, showing the Street itself is using the FY2027 figure as its forward-multiple anchor, not treating it as near-term; (c) the same trap fired independently on ANET, AMD and TSM this week, evidenced against each company's own guidance — a repeated pattern, not a one-off read. Three corroborations clears the bar this command sets for a status change.

Row 26 (SPV credit exposure) — deliberately left UNTESTED rather than adjudicated. Every source found is a secondary aggregator quoting the same underlying BofA note and SEC filings; none was read directly. Per Phase 2's rule, a claim resting on unverified vendor-adjacent reporting does not get a status change on the strength of repetition across outlets that may all be citing one source. This is flagged prominently as the top item for the next pass rather than folded into a conviction cut on thin sourcing.

4. Thesis Persistence and Conviction Delta

Persistence: Of the 12 Structural + Trend claims in the ledger (rows 5, 7-12, 17-22), 8 survived as CARRIED or REFRESHED, 2 DRIFTED, 2 UNTESTED, 0 RETRACTED/SUPERSEDED → ~67% clean persistence (75% if the 2 UNTESTED rows are excluded from the denominator rather than counted against it). This is high enough that the core thesis — dominant custom-silicon design partner plus a high-margin recurring software layer — is intact. It is not the ~90%+ persistence that would justify raising conviction; the two DRIFTED rows (Marvell/Google dual-sourcing, the net-debt pace claim) are real, not cosmetic.

Conviction: 8.5 → 8.0. Driven by: - Down: Row 17 (Google/Marvell dual-sourcing at the exact customer the baseline cited first) and row 26 (new, unverified but credible-enough-to-flag credit exposure to Anthropic's ability to pay) are both genuinely new negative information, not re-weighted old facts. - Up (partial offset): Row 25 (4th hyperscaler substantively confirmed, >$10B orders) and row 10/11 (FCF and revenue growth both accelerating ahead of the baseline's own estimate pace) are genuinely new positive information. - Neutral but important: Row 1 (true near-term multiple is 32x, not 19.7x) doesn't change what the business is worth, but it removes a "this looks cheap" argument the baseline leaned on in its Bogle section — the position is better justified on FCF/growth durability than on a headline multiple that was measuring the wrong year.

No claim in this ledger was RETRACTED — nothing in the baseline was wrong when written; two things were superseded by better data (both Price rows, which always re-derive) and two Structural/Trend claims drifted against the thesis without breaking it.

5. What Is Genuinely New

  • CFO transition (row 15) — orderly, mildly favorable given the incoming CFO's Alphabet background.
  • Fourth hyperscaler substantively confirmed (row 25) — was a rumor in the baseline, now backed by the CEO's own order-size disclosure.
  • Off-balance-sheet SPV/vendor-financing credit exposure to Anthropic (row 26) — a risk vector the baseline had no way to see in June; it postdates the baseline's Scenario A-D stress test entirely and doesn't map cleanly onto any of the four named scenarios. This deserves a dedicated primary-source check (10-Q, the credit-rating agencies' commentary on the CDS move) before the next print, not a conviction cut on today's secondary sourcing alone.
  • Sector-wide AI-capex repricing (SOX -28.6% peak-to-trough) as market context distinct from anything company-specific — worth carrying forward as the backdrop against which the Sep 2 print will be read.

6. Updated Verdict

ACCUMULATE at $371.54 — in the validated $360-420 zone, six days before a binary earnings event. Conviction [8.0] (was [8.5]).

Broadcom's core thesis is intact and, on the hard numbers, strengthening: TTM revenue growth accelerated to +48% YoY, TTM FCF is up +21.7% to $32.76B, leverage improved (Net Debt/EBITDA 1.4x → ~1.08x), the dividend growth story continues, and a fourth hyperscaler customer has moved from rumor to substantively confirmed. Against that, two new and real risks surfaced since June: Google has begun dual-sourcing attach silicon to Marvell inside its TPU program — the first documented instance of the baseline's own top competitive-risk scenario — and Broadcom's credit standing is now reportedly extended into off-balance-sheet debt financing a single AI lab's chip leases, a risk vector that did not exist in the baseline's framework and has not yet been independently verified.

The single most important fact for anyone acting on this report: nothing here reflects Q3 FY2026 results. The print lands September 2, six days out. A position taken today is a position taken into that event, not around it — the framework's own instruction from the current watchlist entry stands: act deliberately before the print, or wait deliberately for it; don't drift into a third month of "in zone, unresolved."

Valuation (re-derived from scratch, Price rows never carry forward)

  • Graham's Number: Still not applicable — tangible book value remains negative (~-$48.8B FY2025 basis; no new annual print to update it).
  • Bogle Expected Return: Corrected for the true near-term multiple: 0.73% yield + FY2026→FY2027 EPS growth (~68% per the two consensus figures, front-loaded by the AI ramp) is not a clean single-year Bogle input given the ramp's lumpiness — treat as directional support, not a point estimate. The PEG-based "extremely cheap" framing from the baseline is retired; multiple-expansion is not free money here at a true ~32x near-term multiple.
  • DYT: Current yield 0.73% vs. 5yr average 1.80% — still reads "expensive" on this metric, still the same noisy signal flagged in the baseline (structural change from the VMware deal corrupts the comparison window). Yellow flag, not a sell signal, unchanged.
  • DDM: Still not applicable — growth assumptions exceed any reasonable discount rate in the bull case.
  • FCF-based fair value, updated for actual TTM trajectory:
Scenario FY2026E FCF Multiple Implied EV Implied Price/Share
Bear (Marvell erosion + capex pause) $34-38B 25-28x ~$900B-1.05T $190-250
Base (guidance broadly met) $42-46B 33-38x ~$1.5-1.75T $390-480
Bull ($100B+ FY2027 AI on track) $52B+ 42-45x ~$2.3-2.5T $550-700

Fair value range: $390-480 (base case), narrowed and shifted modestly up from the baseline's $380-480 — the growth trajectory is running ahead of the baseline's own estimate, but the two new risk vectors argue against widening the top of the range further.

  • Entry: $360-420 (carried, market-validated over 2.5 months) — spot $371.54 sits in the lower third.
  • Strong buy: <$360.
  • Trim: 25x fwd (FY2027 non-GAAP basis, epsForward $19.50 → renders ≈$487 today) — down modestly from the prior 26.7x to price in the Marvell/Google and SPV-credit risk vectors without abandoning the growth case. Re-derive at the next pass once the Sep 2 print and the SPV-exposure question are both resolved.

Break Triggers (carried, one clarified)

  • AI semiconductor revenue falls below $10B for two consecutive quarters — untested, first testable Sep 2.
  • Google or Meta ends its XPU partnership outright — not fired; watch instead for Marvell winning core TPU ASIC work (not attach silicon) as the next, lower-bar warning sign.
  • VMware software ARR declines YoY.
  • Net Debt/EBITDA rises above 3x (currently ~1.08x — wide berth).
  • FY2027 AI revenue guidance formally withdrawn or cut below $60B.
  • New: confirmation that Broadcom carries direct guarantee (not just credit-enhancement optics) exposure to Anthropic SPV debt in a size that could pressure its own investment-grade rating.

Upgrade Conditions

  • Sep 2 print clears $16B Q3 AI semiconductor revenue with FY2027 $100B+ guidance reaffirmed or raised.
  • Primary-source confirmation that Broadcom's SPV exposure is credit-enhancement optics rather than a direct guarantee, removing row 26 as an open risk.
  • Marvell's Google win stays confined to attach silicon through at least two more quarters with no core-ASIC erosion.

7. What This Pass Did NOT Test

  • Row 5 (ROIC ex-goodwill, ~47% FY2025) — no independent TTM recompute performed; carried forward as the FY2025 figure, not re-asserted as current.
  • Row 19 (VMware 77% op margin, ARR growth) — no new quarter since baseline; genuinely unknown whether this held through Q3.
  • Row 23 (the $10B/quarter break trigger) — cannot be tested before Sep 2 by construction.
  • Row 26 (SPV/Anthropic credit exposure) — flagged, not verified. Every source traced back to the same handful of secondary aggregators; none was read as a primary filing this pass. This is the top item for the next /analyze-from-before or an interim /sentiment check, ideally timed to the 10-Q that will accompany the Sep 2 print.
  • Institutional 13F-level granularity beyond the top-10 holders and options positioning were not pulled this pass — out of scope for a fundamentals-led differential update.

Sources