AVGOBroadcom Inc.

AVGO — dossier, 2026-09-21

Slot names-ai-capex-1, name pass. Drafted unattended by the cloud research routine following .claude/skills/stock-sourcing-pipeline/SKILL.md. Nothing here is human-verified. Every figure carries the document it was read from; anything from background knowledge is marked [background]. Never edited after today — corrections go in scorecard.md's log.

Price reference: $356.96, ROIC.ai print dated 2026-09-21 on volume of 144,831 — a thin print, not a reliable close; the NVDA and AMD prices in their dossiers of this date are 2026-09-18 closes. Treat every multiple below as approximate for that reason. Shares outstanding 4,773,629,865 as of 2026-08-28 (EDGAR dei:EntityCommonStockSharesOutstanding, Q3 FY26 10-Q cover). Implied market capitalisation ≈ $1.70tn.

---

S1 — Source

watchlist (owner, Apple Stocks, 2026-09-20), cluster ai_semis, coverage: sec_domestic, cik 1730168. No Tier 0 promotion: queue.json promoted was empty at the start of this slot.

---

S2 — Kill test

EDGAR XBRL companyconcept. Fiscal years end in late October / early November; FY2025 ended 2025-11-02.

Share count over five years. Split-adjusted for the 10-for-1 [background: the FY2024 jump from 427m to 4,778m in the as-reported series is the split], the count went 4.29bn → 4.85bn, +13%. Almost all of it is one event: 544 million shares issued for VMware. Since then it creeps up on stock compensation — $6,287m of SBC in the first nine months of FY26 alone, against $5,373m a year earlier. Diluted shares were 4,887m in Q3 FY26 and management guides ~4,940m for Q4 despite a buyback. The count is going up, not down.

The leverage, and what has happened to it. The FY25 10-K states the number plainly:

As of November 2, 2025, the aggregate indebtedness was $67,120 million.

By 2026-08-02 the balance sheet shows $57,167m long-term plus $2,252m short-term = $59.4bn. On the Q3 call the CFO said $5.6bn of long-term debt was repaid in the quarter and a further $1.5bn of senior notes at maturity after quarter end, with the remaining gross fixed-rate principal at a 4% weighted coupon and 7.4 years to maturity. Against $27.5bn of FY25 operating cash flow and a Q3 FY26 run-rate far above it, net debt of ~$35.4bn is roughly one year of free cash flow. There is no maturity wall.

Goodwill is the other half of the balance sheet. $97,801m of goodwill and $26,325m of net intangibles at 2026-08-02 — 66% of $188.1bn of total assets, against $99.7bn of equity. A serious impairment would take equity with it. Nothing indicates one; the exposure is structural.

Analyst coverage count: not captured this pass. Thin.

Verdict — survives or dies: Broadcom survives easily on cash flow and is actively deleveraging. It is the only name in this slot paying debt down while the other two raise it. The risk is not solvency, it is that two thirds of the assets are purchase accounting and an accelerating share of the revenue comes from a handful of counterparties.

---

S3 — Filings, at gate-minimum depth

Read this pass: FY25 10-K (Item 1A risk factors and Item 7 MD&A only), Q3 FY26 10-Q (statements and Item 2 MD&A), DEF 14A 2026-03-02 (ownership), and the Q3 FY26 earnings call. One transcript, not two. Not read: the FY24 10-K, older 10-Qs, comment letters (AVGO's EDGAR page shows 15 CORRESP and 11 UPLOAD items — none read), critical audit matters, auditor tenure and fees. The 56 Form 425 filings on the EDGAR page were checked and all date from 2022–2023: they are the VMware merger, not a pending transaction.

The three biggest risks management itself discloses

1. Customer concentration — and it is getting worse fast. FY25 10-K, Significant Customer Information note:

During fiscal years 2025, 2024 and 2023, one customer accounted for 32 %, 28 % and 21 % of our net revenue, respectively. Revenue from this customer was included in our semiconductor solutions segment. One customer accounted for 44 % and 18 % of our net accounts receivable balance as of November 2, 2025 and November 3, 2024, respectively.

Q3 FY26 10-Q, Item 2 MD&A:

Direct sales to one semiconductor solutions customer, which is a distributor, accounted for 50% and 46% of our net revenue for the fiscal quarter and three fiscal quarters ended August 2, 2026, respectively, and 32% and 30% of our net revenue for the fiscal quarter and three fiscal quarters ended August 3, 2025, respectively.

We believe aggregate sales to our top five end customers, through all channels, accounted for approximately 55% and 50% of our net revenue for the fiscal quarter and three fiscal quarters ended August 2, 2026, respectively, and approximately 40% of our net revenue for each of the fiscal quarter and three fiscal quarters ended August 3, 2025.

The arc is 21% → 28% → 32% → 46% → 50% in one customer line, and ~40% → ~55% in the top five, in the space of three years. The cluster file of 2026-09-20 records AVGO at 32% of FY25 revenue; that number is a year stale and understates the position by 18 points.

2. Leverage. Quoted above from Item 1A. Broadcom is the one name in this slot that carries a dedicated "Risks Related to Our Indebtedness" heading.

3. AI customers are pressing on terms — and want to lease rather than buy. FY25 10-K, Item 1A:

our top customers, including our AI customers, may make and have made greater demands on us with regards to pricing and contractual terms, such as seeking to lease AI racks or systems based on our XPUs instead of purchasing, as well as alternative financings for such l

(the quote is cut where the located span ends; the sentence continues past the extraction boundary and was not completed by hand.) This is the same pressure NVIDIA describes and the direct origin of the XPV platform described below.

Latest quarter — Q3 FY26 (ended 2026-08-02), from the 10-Q

Nine months: revenue $71,089m vs $45,872m; net income $29,747m vs $14,608m; diluted EPS $6.09.

Balance-sheet deltas worth naming, 2026-08-02 against 2025-11-02: cash $16,178m → $23,975m; trade receivables $7,145m → $13,707m (+92%); inventory $2,270m → $4,523m (+99%); long-term debt $61,984m → $57,167m. Receivables and inventory are both close to doubling in three quarters while revenue over the same nine months grew 55%. Working capital is expanding faster than revenue — consistent with a ramp, and worth watching for exactly that reason.

Guidance and structure from the Q3 FY26 call (2026-09-02)

Machine transcript; wording is mangled in places and the numbers below are what it renders.

• Q3 AI semiconductor revenue $16.7bn, +221% y/y, +54% q/q, 56% of total revenue (49% in Q2). XPUs were 73% of AI revenue and up over 3.5x y/y; AI networking up over 2.5x.
• Q4 FY26 guide: consolidated $34.8bn (+93%), AI $21.7bn (+236%), semiconductors $26.1bn, infrastructure software $8.7bn, gross margin ~73%, operating margin ~66%, diluted share count ~4.94bn, capex $1.4bn.
• FY26 AI revenue ~$58bn (+186%), raised from a prior ~$56bn.
• FY27 AI revenue ~$115bn, with management stating the supply is secured; FY28 ~$230bn, also supply-secured; and "on target to exceed $30 in earnings per share in fiscal 2028."
• Six XPU customers. Named: Google (Ironwood TPU v7 in volume, next generation in production, long-term agreement for "multi tens of billions of dollars" of TPUs annually), Anthropic (1 GW 2026, +5 GW 2027, line of sight to +10 GW 2028, expected to become the largest XPU customer in 2027), OpenAI (first-generation accelerator shipped, 1.3 GW in 2027, over 5 GW in 2028, second-largest), Meta (three MTIA generations to end-2027, 3 GW through 2028).
• Q3 non-GAAP EPS $3.32 (+96%); non-GAAP gross margin 75%, down 210bps sequentially on XPU memory content; operating margin 67.9%. Free cash flow $13.7bn, 46% of revenue.
• Capital returns: $3.1bn of dividends at $0.65/quarter; $5.6bn of long-term debt repaid.
• A Singapore fab is to be deployed for substrate from fiscal 2027, addressing a supply bottleneck.

The XPV platform — the structural fact. From the CFO's prepared remarks: a platform established in June 2026 in partnership with Apollo and Blackstone to enable more than 20 GW of compute infrastructure for OpenAI and Anthropic by the end of 2028, with a first $35bn tranche closed in June for Anthropic's 1 GW deployment. Management's stated position is that third-party partners underwrite and capitalise the assets rather than Broadcom financing them, and that Broadcom may provide "modest residual value guarantees, which are contingent liabilities we view as low risk."

This is the third variant of the same structure found in one slot. NVIDIA gives a capped $105bn guarantee behind OpenAI's leases and a $36bn take-or-pay back to the AI clouds; AMD gives 320m warrant shares to OpenAI and Meta; Broadcom co-sponsors a $35bn-and-growing financing vehicle for the same two labs and may backstop residual value. All three sellers of the shortage are now, in different instruments, underwriting their own buyers. The cluster pass of 2026-09-20 drew the line between self-funding sellers and debt-funded buyers; that line has been crossed from the seller side by every name in this slot, in the four months since.

Of the three disclosures, Broadcom's is the thinnest. NVIDIA's guarantee is in a 10-Q with a cap, a trigger and a termination condition. Broadcom's XPV exists, this pass, only in call commentary with the adjectives "modest" and "low risk" attached and no number.

Proxy skim (DEF 14A filed 2026-03-02)

All 13 current directors and executive officers as a group: 89,467,014 shares, 1.9%. Individually disclosed lines are far smaller — Charlie B. Kawwas 814,716, Kirsten M. Spears 187,237 — so the group total is concentrated in the CEO. Compensation structure and what the bonus is paid on: not extracted this pass — thin.

Insider pattern (12 months to 2026-09-21)

The one piece of open-market insider buying found anywhere in this slot: director Harry L. You bought 1,000 shares at $373.57 on 2026-06-11 and 1,000 at $325.13 on 2025-12-18, both direct, neither under a 10b5-1 plan, holding 38,466 after the later purchase. Small, but it is the only non-zero line across NVDA, AMD and AVGO.

Red-flag sweep

• Going concern: no flag. Newest examined filing is the 10-Q filed 2026-09-10; coverage there starts 2026-06-09.
• Executive changes: CFO transition. Kirsten M. Spears notified the company on 2026-03-30 of her retirement and resigned as both Chief Financial Officer and Chief Accounting Officer effective 2026-06-12; Amie Thuener was appointed CFO from the same date. Her first full quarter as CFO is the one in which XPV closed its first $35bn tranche and single-customer concentration reached 50%. Stated here as a fact and a watch item, not an allegation. Also on the board since 2024-02-05: Kenneth Y. Hao, Chairman and Managing Partner of Silver Lake, as an independent director.
• Auditor change / restatement / NT 10-K: not run this pass — the efts.sec.gov full-text sweep was dropped for time. Thin. AVGO's 15 CORRESP and 11 UPLOAD items make it the name in this slot where that sweep would be most worth running.

---

S5 — Ownership

Who is on the other side: almost nobody, and less than two months ago. Short interest fell 28% between 2026-06-30 and 2026-08-31 to 1.1% of shares out. There is no visible bear. Capital Group's 8.2% is the only large discretionary holder, and it is long.

---

S8 — Valuation and sizing (model; both decisions are ⏳ OPEN (user))

Broadcom guides on a non-GAAP basis. The GAAP-to-non-GAAP gap is very wide — Q3 FY26 GAAP operating margin 53.9% against 67.9% non-GAAP — because ~$2.0bn a quarter of acquisition intangible amortisation and ~$2.1bn of stock compensation are excluded. The SBC exclusion is real dilution, and the share count above shows it arriving.

Management's own FY28 figure is "exceed $30 in earnings per share", which the model above lands on independently from the revenue guide — the two are consistent, which is reassuring about the arithmetic and says nothing about whether the revenue arrives.

EV: market cap ~$1.70tn plus net debt ~$35.4bn → EV ≈ $1.74tn, ≈ 10.4x FY2027E revenue and ≈ 6.0x FY2028E revenue.

• Bear: FY27 AI revenue lands at $85bn rather than $115bn because one named customer's gigawatt schedule slips, and gross margin keeps sliding on memory content. Revenue ~$138bn, EPS ~$13, and the multiple compresses because the FY28 number stops being credible. A 40%+ drawdown from here follows arithmetically.
• Base: the guide is met. ~$17.7 FY27 EPS at ~20x today, ~$30 FY28 at ~12x.
• Bull: demand exceeds the secured supply, as management said it already does, and the Singapore substrate capacity unlocks in FY27. FY28 revenue above $300bn, EPS above $32.

The honest summary of the valuation: like NVIDIA, Broadcom is cheap on management's own out-year numbers and not cheap on trailing ones. The gap between ~30x FY26 and ~12x FY28 is the question — it prices the FY27 and FY28 AI guides at a discount, and the reason is visible in S3: 55% of revenue in five customers, four of whom are also being financed into their purchases by the vendors selling to them.

Liquidity: average daily volume 20,841,544 shares (FINRA, 2026-08-31 settlement) at ~$357 ≈ $7.4bn traded per day. No sizing constraint.

⏳ Buy price: OPEN (user). Size: OPEN (user). Both are owner-only under the skill.

---

S9 — Monitoring

• Next earnings: Q4 FY2026 ends ~2026-11-01; Broadcom has reported that quarter in the second week of December in recent years. Derived from the reporting cadence, not from a sourced calendar — ROIC.ai's earnings calendar is behind a paid plan. Thin.
• Filing feed: Tier 0 EDGAR sweep (watch_edgar.py, 06:40 UTC) covers cik 1730168.
• Leading indicator between quarters: the "Net Revenue" paragraph of each 10-Q's Item 2, which restates the single-customer and top-five-customer percentages every quarter. It has moved 32% → 46% → 50% and is the most informative single sentence Broadcom files. Second: whether the XPV residual-value guarantees ever appear as a quantified item in Note 10 rather than as call commentary.
• Trigger that converts this to a buy: owner's call. The condition that would settle the central objection is the opposite of the concentration trend — a quarter in which AI revenue tracks the $115bn FY27 path while the top-five share stops rising, meaning the growth is broadening across the six XPU customers rather than deepening into one or two. The adverse case is the guide being met with concentration still climbing, which is the same revenue bought at a worse risk.

---

Access and connector notes for this run

• SEC EDGAR: reachable, with the required User-Agent: ddboard-research gorkem@celebico.com.
• Equibles: reachable. 22 calls across the three names of this slot against the 25-call cap.
• ROIC.ai: reachable for prices (though AVGO's latest print is thin, see the header); list_earnings_calendar is gated behind a paid plan.
• 3spread (ladder rung 4b, added to the skill by commit c07b237 at 14:06 +0300 today, while this slot was running): NOT available. python ledger/research/tools/threespread.py coverage AVGO exits with THREESPREAD_API_KEY is not set (environment or .env). The cloud environment has neither the variable nor, presumably, api.3spread.com on its allowed-domains list. S5 and the insider pattern for all three names in this slot therefore ran on Equibles, which rung 4b names as the fallback.
• EDINET DB / eu_data: not needed for three sec_domestic names; not called.
• Not called, per the skill: Alpha Vantage, Bigdata.com, FMP.

Process support, not investment advice.