Broadcom AVGO

Reference table

1. What the business actually is

The vendor's own company description is stale and should be discarded. It says Broadcom's "operations are segmented into four primary divisions: Wired Infrastructure, Wireless Communications, Enterprise Storage, and Industrial & Other" and that it "maintains a significant team of 19,000 full-time staff." The four-division structure is the FY2018 reporting format — the vendor's own seg_prod block shows it was abandoned after that year — and the 19,000 headcount contradicts the profile.employees field of 33,000 inside the same payload. The 10-K is unambiguous: "We have two reportable segments: semiconductor solutions and infrastructure software." We use the filing.

Revenue by segment (from seg_prod, which reconciles exactly to reported revenue and matches the filing's structure):

The software line went from $7.637B to $27.029B in two years — that is the VMware acquisition (closed 2023-11-22) plus the conversion of that base to subscription. The 10-K attributes FY2025 software growth to "strong demand for our VCF product, including license revenue recognized on contracts where customers do not have the right to terminate and the transition to a subscription license model," and semiconductor growth to "strong demand for our networking solutions, primarily custom AI accelerators and AI networking products."

One disclosure materially changes how the software line should be read. The 10-K states: "In fiscal year 2025, we included upfront license revenue of $7,800 million within products revenue. To conform to the current year presentation, we reclassified $4,601 million of upfront license revenue from subscriptions and services revenue to products revenue for fiscal year 2024." $7.8 billion — 12.2% of total company revenue and 28.9% of the software segment — is point-in-time upfront license recognition, not subscription. That is a lumpier, less annuity-like revenue stream than the word "software" implies, and it is the mechanism by which a software segment can grow 25.8% in a year.

Geography, FY2025 (from seg_geo, which also sums exactly to $63.887B): Asia-Pacific $35.896B (56.2%), Americas $18.939B (29.6%), EMEA $9.052B (14.2%). The 10-K adds the detail that matters for policy risk: "In fiscal years 2025 and 2024, 17% and 20%, respectively, of our net revenue came from shipments or deliveries to China (including Hong Kong)," while noting that end demand is frequently located elsewhere.

Customer concentration is the defining structural feature and it is getting worse. The 10-K: "Direct sales to one semiconductor solutions customer, which is a distributor, accounted for 32% and 28% of our net revenue for fiscal years 2025 and 2024" and "we believe aggregate sales to our top five end customers, through all channels, accounted for approximately 40% of our net revenue for each of the fiscal years 2025 and 2024." The 10-Q updates the second figure: "approximately 45% of our net revenue" for the May 2026 quarter and the two quarters then ended, against 40% a year earlier.

Headcount 33,000 (profile). No contradicting figure was located in the extractable 10-K text; the DESC field's 19,000 is rejected.

One structural note on the fiscal calendar. Broadcom operates on a 52- or 53-week year ending the Sunday closest to 31 October. FY2025 ended 2025-11-02 and was a 52-week year; FY2024 was a 53-week year, which flatters FY2024 comparisons slightly and is worth remembering when reading the +23.9% FY2025 growth rate.

2. Three earnings numbers, a 70x headline, and which one we value on

Trailing four-quarter GAAP diluted EPS is $6.00. At $418.16 that is a 69.7x multiple, and the vendor's own priceToEarningsDilutedRatioTTM of 69.69 confirms the arithmetic. Quoting that number without explaining it would be dishonest in either direction.

Here is what suppresses it, taken from the filings rather than asserted.

(a) Amortisation of acquisition-related intangibles: $8.201B in FY2025. The 10-K discloses depreciation expense separately — "Depreciation expense was $574 million, $593 million and $502 million for fiscal years 2025, 2024 and 2023" — and total depreciation and amortisation for FY2025 was $8.775B. The residual, $8.201B, is amortisation of purchased intangibles: 12.8% of revenue and 35.5% of pre-tax income. The independent cross-check is the balance sheet: net intangible assets fell from $40.583B to $32.273B over the year, a decline of $8.310B, consistent with a small amount of additions. This charge is non-cash and is a direct consequence of purchase accounting for VMware. The 10-K confirms the run-off is already beginning: amortisation "recognized in operating expenses decreased $1,213 million, or 37%, in fiscal year 2025 ... primarily due to full amortization of customer-related intangible assets from previous software acquisitions other than VMware."

(b) Stock-based compensation: $7.568B in FY2025, 11.6% of revenue. The 10-K: "Total stock-based compensation expense was $7,568 million and $5,670 million for fiscal years 2025 and 2024." The 10-Q shows the run-rate rising: $2,092M in the May quarter against $1,771M a year earlier, and $4,268M for two quarters against $3,051M. There is $20.106B of unrecognised cost still to be expensed over a 3.0-year weighted service period, and $13.527B of restricted units vested in the first half of FY2026 alone.

(c) A tax benefit, not a tax charge. FY2025 carries a $397 million benefit from income taxes on $22.729B of pre-tax income, "primarily due to the recognition of uncertain tax benefits from expiration of statutes of limitations and audit settlements, and excess tax benefits from stock-based awards." Trailing effective tax rate: 3.8%. This flatters GAAP earnings in the opposite direction from (a) and (b), and it is not sustainable — the 10-K warns that Singapore's global minimum tax "will become effective in our fiscal year ending November 1, 2026" and that "we expect a material impact from the enactment of these laws on our consolidated results of operations and cash flows for our fiscal year 2026."

The three lenses, at $418.16:

The reconciliation checks out. For FY2025 the four earn_cal prints imply roughly $6.82 of EPS against GAAP's $4.77; the vendor's FY2025 consensus epsAvg was $6.748. That confirms that every est row in this file is on the adjusted basis, not GAAP — a trap worth naming, because comparing FY2027E EPS of $19.546 against trailing GAAP of $6.00 would imply a growth rate that does not exist.

Our position, stated plainly. We value on the analyst basis because that is what the estimates are built on and it is the only internally consistent way to compute a forward multiple — but we do not accept that stock compensation is not an expense. It is 11.6% of revenue, it is rising, and it shows up in the share count: diluted shares went from 4.836B (February 2025) to 4.876B (May 2026) despite $8.450B of buybacks in two quarters. Where the distinction matters to the verdict — in the fair-value work in Section 6 — we haircut the multiple rather than the earnings, and we say so.

3. Growth, margins, and the contracted book

FY2024 is the VMware integration year — gross margin fell 590 basis points and operating margin collapsed to 26.1% under amortisation and restructuring, and GAAP EPS fell to $1.23. FY2025 is the recovery, and FY2026 is the acceleration.

Quarterly, with the gross-margin line taken from the filings rather than the vendor (see Section 8):

Read the last two rows against the vendor's version and the story inverts. The vendor reports Q1 FY26 gross profit of $12.663B (65.6%) and Q2 FY26 of $14.919B (67.2%) — a picture of margin compression from FY2025's 68.0%. The 10-Qs state the opposite: "Gross margin was $13,157 million for the fiscal quarter ended February 1, 2026" and "$15,415 million for the fiscal quarter ended May 3, 2026 ... As a percentage of net revenue, gross margin was 69% ... and 68% for each of the fiscal quarter and two fiscal quarters ended May 4, 2025." The vendor understates gross profit by $494M and $496M respectively — a consistent ~$0.5B per quarter defect confined to the FY2026 quarters (the FY2025 quarters reconcile to the dollar). The filing wins: gross margin is EXPANDING, from 68% to 69%, and operating margin has reached a record 49.0%.

The forward evidence — and it is not in the vendor payload in any form.

The 10-K, as of 2025-11-02:

> "Certain multi-year customer contracts in our semiconductor solutions segment and infrastructure software segment, including contracts where customers do not have termination rights, contain firmly committed amounts and the remaining performance obligations under these contracts as of November 2, 2025 were approximately $33.3 billion. We expect approximately 35% of this amount to be recognized as revenue over the next 12 months."

The 10-Q, seven months later, as of 2026-05-03:

> "Certain multi-year customer contracts ... contain firmly committed amounts and the remaining performance obligations under these contracts as of May 3, 2026 were approximately $164.6 billion. These commitments include obligations under a long-term contract for custom AI accelerators entered in the fiscal quarter ended May 3, 2026. We expect approximately 30% of this amount to be recognized as revenue over the next 12 months."

$33.3 billion to $164.6 billion in two quarters — a 394% increase — with the cause named in the footnote. Thirty per cent of $164.6B is approximately $49.4 billion of contracted revenue in the year to May 2027, against total FY2025 revenue of $63.887B. This is the single most important disclosure in the file and it does not appear in any vendor field.

Two supporting disclosures from the filings. The 10-Q notes "Inventory increased to $4,328 million at May 3, 2026 from $2,270 million at November 2, 2025 primarily to support higher expected shipments for custom AI accelerators" — a 91% inventory build with a stated cause, which is what a real ramp looks like. And the 8-K filed 2026-07-06 states that Broadcom and Apple "have agreed to expand their long-standing technology collaboration through 2031 by entering into new multi-year long-term agreements for Broadcom to develop and supply a range of custom ASIC silicon products for use in multiple generations of Apple products." A named, dated, five-year design win outside the hyperscaler concentration.

Consensus forward path (analyst counts in brackets): revenue $105.647B FY2026 (32), $174.884B FY2027 (35), $234.230B FY2028 (27), $290.277B FY2029 (13). Analyst-basis EPS $11.590 (27), $19.546 (28), $26.136 (13), $32.323 (5). The FY2029 and FY2030 EPS rows rest on five analysts each and are not used for anything; the FY2030 revenue row of $139.000B is LOWER than FY2029's $290.277B and is rejected outright as corrupt (Section 8).

4. Balance sheet and capital returns — clean where it matters, ugly where it counts

The vendor's net debt is correct, which is unusual enough to state. netDebt of $48.958B equals total debt $65.136B less cash and short-term investments $16.178B exactly — and short-term investments are genuinely zero on this balance sheet, so nothing was discarded. capitalLeaseObligationsCurrent and capitalLeaseObligationsNonCurrent are both 0.00, and the 10-K corroborates: Broadcom has operating leases only — "We have operating leases for our facilities, land, data centers and certain equipment. Operating lease expense was $182 million, $187 million and $91 million for fiscal years 2025, 2024 and 2023." $182 million of annual lease expense on a $63.9B revenue base is immaterial, there are no finance leases, and there is no double-count. We checked specifically because the VMware financing was large; the balance sheet is clean.

The filing does refine the debt figure. The 10-K states liquidity requirements include "interest and principal payments related to our $67,120 million of outstanding indebtedness with $3,152 million principal amounts payable within 12 months." The balance sheet carries $65.136B — the $1.984B difference is unamortised discount and issuance costs. We use the filing's principal. The 10-Q updates it: "As of May 3, 2026, the aggregate indebtedness was $66,720 million" with "$2,252 million principal amounts payable within 12 months," against $19.628 billion of cash — "our primary sources of liquidity as of May 3, 2026 consisted of: (i) $19,628 million in cash and cash equivalents." Net debt at the most recent balance-sheet date is therefore approximately $47.1 billion, modestly better than the fiscal-year-end figure.

Debt composition and market view:

The debt trades below par ($62.505B fair value against $66.720B principal), which means the coupons sit below current market rates — a mild positive. The company has been actively managing it: the 10-Q discloses $4,500M of new senior notes issued and $3,650M repaid or redeemed in the February quarter, and a further $1,250M redeemed in the May quarter; the 8-K of 2026-06-11 announced cash tender offers for certain debt securities, and the 8-K of 2026-06-18 announced their "expiration, upsize and results." Interest expense fell from $3.953B (FY2024) to $3.210B (FY2025). Interest coverage is 11.0x. This is not a leverage problem.

What is a problem is the asset side. Goodwill $97.801B plus intangibles $32.273B equals $130.074B of $171.092B total assets — 76%. intangiblesToTotalAssetsTTM is 0.704. Total equity is $81.292B, so net tangible equity is NEGATIVE $48.782B and tangible book value per share is NEGATIVE $8.10. Retained earnings are only $9.761B (against $0.00 a year earlier) because the company distributes essentially everything it earns. Property, plant and equipment is $2.530B — smaller than a single quarter's research spend. This is a balance sheet of intellectual property and purchase accounting, and the vendor's quality model reads it correctly: rating B, overall score 3, with debt-to-equity, price-to-earnings and price-to-book sub-scores all 1.

And then the item that is in no ratio at all. The 10-Q's subsequent-events note:

> "On June 8, 2026, we arranged for an investor partner to take on certain agreements to purchase AI racks based on custom AI accelerators designed by us and the related lease agreements with a customer that enable access to compute capacity. In connection with the arrangement, we entered into a backstop agreement with the investor partner for the customer's lease obligations over 5-year terms. The backstop will increase over time as the AI racks are deployed and decrease as the customer makes payments on its lease obligations, with a maximum exposure of $29 billion. In the event of default by the customer, we have various remedies, including the assumption of the lease or effecting a sale of the AI racks, which would reduce our maximum exposure."

$29 billion — 43% of the entire on-balance-sheet debt load — of contingent exposure to one customer's ability to pay for compute capacity. The risk factors were updated to match: "we have entered into and may enter into financing arrangements where we provide a backstop or a guarantee to support our strategic initiatives or other corporate purposes, which could increase our fixed obligations, have a negative impact on our financial condition and reduce our financial and business flexibility." This is vendor-financing risk in a new costume, it is disclosed nowhere in the payload, and it should always be quoted alongside the $164.6B order book, because it is part of how that book was won.

Capital returns. Dividends of $11.142B in FY2025 (against $9.814B and $7.645B in the two prior years); the board declared a quarterly dividend of $0.65 per share on 2026-06-02, payable 2026-06-30 — $2.60 annualised, a 0.62% forward yield against a trailing dividendPerShareTTM of $2.54 and a 0.61% trailing yield. Payout is 40.1% of adjusted earnings and 41.4% of free cash flow. Buybacks were $6.310B in FY2025 and $8.450B in the first two quarters of FY2026 (25 million shares), with $10.1 billion of authorisation remaining at 2026-05-03.

But the share count went the wrong way. Diluted shares: 4.836B (Q1 FY25) → 4.826B → 4.860B → 4.889B → 4.888B → 4.876B (Q2 FY26). Up 0.8% year over year despite $8.450B of repurchases, because $13.527B of restricted units vested in the same half-year. Net buyback yield is approximately zero. Total shareholder yield is therefore the dividend alone, ~0.6% — and this is the single clearest place where the "stock compensation is not a real expense" argument fails an empirical test.

Enterprise value. enterpriseValueTTM reports $2,034.709B. Rebuilt (market cap $1,989.430B + total debt $65.136B − cash $16.178B) it is $2,038.388B, a −0.2% residual — the cleanest EV in this batch, and we accept the vendor figure. On the filing's principal debt it would be $2,040.372B. evToEBITDATTM of 48.4x is arithmetically defensible but rests on a GAAP EBITDA that carries the amortisation issue of Section 2; we do not lean on it.

Contingencies are immaterial. The 10-Q states management does not believe pending legal proceedings "taken individually or as a whole, will have a material adverse effect" and that "no material amounts have been accrued or disclosed ... with respect to loss contingencies." $1.662B of unrecognised tax benefits sits outside the contractual-commitments table because timing is unpredictable. No valuation adjustment is made.

5. Cash flow — very high quality, with one line worth watching

This is the argument for the business in one table. Free cash flow converts at 97.4% of operating cash flow (freeCashFlowOperatingCashFlowRatioTTM 0.974) because capital expenditure is 1.0% of revenue. Operating cash flow grew 38.0% in FY2025. incomeQualityTTM of 1.15 says operating cash exceeds reported net income. A fabless designer with a 49% operating margin and no factories is close to the ideal cash-generating structure, and it is why the market is willing to look through a 70x GAAP multiple at all.

Two qualifications, both from the filings.

First, receivables factoring flatters operating cash flow. The 10-K: "Total trade accounts receivable sold under the factoring arrangements were $7,401 million, $5,900 million and $3,975 million during fiscal years 2025, 2024 and 2023" and "we present cash proceeds as cash provided by operating activities." The programme has grown by $3.4B in two years, and that growth sits inside the operating cash-flow line. It is a legitimate accounting treatment, but it means part of the FY2025 improvement is a financing decision. And it is now reversing: the 10-Q attributes rising receivables partly to "lower receivables sold through factoring arrangements" — i.e. a headwind to FY2026 operating cash flow that will not be labelled as such.

Second, capital expenditure is going up. The 10-K states flatly: "We expect capital expenditures to be higher in fiscal year 2026 as compared to fiscal year 2025." The 10-Q confirms "higher capital expenditures in the current year fiscal period." From a base of 1.0% of revenue this is not a threat to the model, but the free-cash-flow conversion rate of 97.4% should not be extrapolated.

Free cash flow yield is 1.65%; price to free cash flow is 60.7x. Those are not attractive numbers on trailing figures and we do not pretend otherwise — they are a statement about the price, not the business.

6. Valuation — priced in or room?

At $418.16 (market cap $1,989.4B, enterprise value ~$2,038B):

The whole valuation question is whether FY2027 happens. At the trailing multiple this is an expensive stock by any measure. At FY2027 consensus it is 21.4x — cheaper than the market's mega-cap software names — but that requires revenue to roughly triple from FY2025's $63.887B in two years. The $164.6B contracted book is the evidence; Section 6a quantifies exactly how much of the gap it actually covers.

Peer context. The vendor peer set is mixed: ASML ($659.8B), Micron ($1,008.2B), AMD ($845.6B), Taiwan Semiconductor ($2,163.2B) and Analog Devices ($185.2B) are legitimate comparisons; FormFactor ($9.4B), Power Integrations ($3.6B), Qorvo ($8.4B) and Semtech ($12.0B) are not comparable to a $1.99 trillion franchise, and Meta Platforms appears in the peer list, which is a classification error. No peer multiples are supplied in the file, so no peer-multiple comparison is drawn.

The vendor's scoring model rates Broadcom B, overall score 3, with return-on-equity and return-on-assets sub-scores of 5 but debt-to-equity, price-to-earnings and price-to-book sub-scores all 1. Translated: excellent returns, expensive and levered. That is a fair summary and we do not argue with it.

6a. What today's price assumes (the inversion)

At $418.16 — 21.4x FY2027 consensus EPS of $19.546, and 36.1x FY2026E — the price embeds roughly the following falsifiable claims:

• Revenue reaches $105.647B in FY2026 (+65.4%) and $174.884B in FY2027 (+65.5%). (Consensus-derived: 32 and 35 analysts.) The first half of FY2026 delivered $41.498B, so the second half must produce $64.149B. Consensus and management's own guide put the August quarter at ~$29.2B; the implied November quarter is therefore approximately $34.9B — versus $22.187B in the May quarter and $18.015B a year earlier. (Our number, derived by subtraction.) This is the most fragile assumption in the price. It is not guided by the company and it is not covered by the contracted book on any reasonable phasing.
• The $164.6B contracted book converts on schedule. (Filing-derived, 10-Q 2026-06-09: 30% within twelve months ≈ $49.4B.) But note the coverage honestly: consensus for the four quarters from August 2026 through May 2027 totals roughly $146B; $49.4B of that is contractually committed. The order book underwrites about one third of the near-term revenue line — genuine evidence, and materially less than the headline implies. (Our number.)
• The effective tax rate does not normalise faster than earnings grow. (Our number, from the filings.) The trailing rate is 3.8%; the FY2025 income statement carries a $397M tax benefit; and the 10-K states management expects "a material impact" from Singapore's global minimum tax in FY2026. Consensus EPS of $11.590 and $19.546 embeds some normalisation, but the size is not disclosed anywhere in this file.
• Gross margin holds near 69% through a mix shift toward lower-margin semiconductors. (Filing-derived: the 10-Q states the growth benefit is "partially offset by a higher mix of semiconductor solutions net revenue, which has a lower gross margin than infrastructure software.") Semiconductor solutions is growing far faster than software, so the mix headwind builds mechanically.
• The $29B backstop is never drawn. (Filing-derived, subsequent-events note.) It is contingent on one customer's ability to meet five-year lease obligations on compute capacity.

6b. The return bridge (why the multiple moves)

Expected return over 12-24 months decomposes as: EPS growth (+42.7% in FY2026 and +68.6% in FY2027 on consensus) + multiple drift (we assume material COMPRESSION, from 21.4x FY2027E to roughly 24x — which is expansion against the forward year but a large compression against the 36.1x being paid on FY2026) + shareholder yield of ~0.6% (dividend only, because the buyback does not shrink the count).

Be precise about what is happening to the multiple here, because it is the opposite of the usual story. When earnings grow 65% and the price grows 12%, the multiple falls hard. From $418.16, if FY2027 EPS lands at consensus and the multiple simply holds at today's 21.4x, the price is $418 — flat. Our base case of $470 requires 24.0x FY2027E. That is not a demand for re-rating enthusiasm; it is a demand that the market pay roughly the same forward multiple a year from now that it pays today, once FY2027 becomes the near year rather than the far one.

The cross-check that makes 24x tolerable: $470 is 18.0x FY2028E of $26.136 and 40.6x FY2026E of $11.590. The first is undemanding; the second is a reminder that this remains an expensive stock on the year in progress.

The cross-check that argues for restraint: $470 sits 7.2% BELOW the street consensus of $506.50 and 9.6% below the median of $520. We are more conservative than 59 covering analysts, and the reason is stated: we treat $7.568B of stock compensation as a cost, we do not extrapolate a $34.9B November quarter that nobody has guided to, and we apply a discount for a $29B contingent exposure and 45% customer concentration.

The bull case at $590 requires 30.2x FY2027E — essentially the street high of $582 — and needs the November quarter to land, the tax rate to stay benign, and the order book to grow again at the September print.

6c. Variant perception (where we differ, what would surprise)

• We have a genuine informational edge on the order book, and it is the reason to own this at all. The vendor payload contains no remaining-performance-obligation field. The 10-Q filed 2026-06-09 discloses $164.6 billion against the 10-K's $33.3 billion, with the cause named. Any screen or model built on the vendor file alone cannot see the strongest forward evidence in the company. Watchable number: the RPO disclosure at the 2026-09-03 print — level and the 12-month conversion percentage, currently 30%.
• We differ from the vendor, and from anything built on it, on the direction of gross margin. The vendor's FY2026 quarterly gross profit understates the filings by ~$0.5B per quarter, turning a margin expansion from 68% to 69% into an apparent compression to 65.6%/67.2%. The filing wins and the sign flips. Watchable number: gross margin percentage as stated in the 10-Q text, not as computed from the vendor's gross-profit line.
• We differ from the street on the destination, and we are lower. Base $470 against consensus $506.50 — a 7.2% gap. The disagreement is concentrated in one place: the implied ~$34.9B November quarter, which is a residual of the annual consensus rather than a guided or contracted figure. Watchable number: whatever the company guides for the November quarter on 2026-09-03. If it guides at or above $33B, our discount is wrong and we raise the base.
• We differ from the more enthusiastic framing on what the buyback does. $8.450B of repurchases in two quarters and the diluted share count still rose 0.8% year over year. The correct shareholder yield for this company is the dividend — about 0.6% — and not a penny more. Watchable number: diluted shares outstanding at each print against 4.876B.
• Positive surprise that would force a re-rate: the order book rising again at the September print — from $164.6B to, say, above $200B — with the 12-month conversion share holding at 30%; or a second named multi-year design win of the Apple type outside the hyperscaler set; or the November-quarter guide landing above $33B. Watchable numbers: RPO; the top-five customer concentration percentage (if it falls while revenue grows, the franchise is broadening, which is the single most bullish thing that could happen).
• Negative surprise that would break the thesis: any disclosure that a major custom-accelerator customer is dual-sourcing or insourcing a programme; the top-five concentration passing 50%; a drawdown or increase on the $29B backstop; an effective tax rate stepping to the high teens without an offsetting revenue beat; or the RPO falling without a matching revenue conversion. Watchable numbers: customer-concentration disclosure in each filing; the subsequent-events and commitments notes.

Synthos fair values

All three anchors are multiples of FY2027 consensus analyst-basis EPS of $19.546 (28 analysts), cross-checked against FY2026E of $11.590 (27) and FY2028E of $26.136 (13). Not a discounted cash flow — stated arithmetic.

• Bear ~$330 — 16.9x FY2027E. Cross-checks: 28.5x FY2026E; 12.6x FY2028E. The scenario: the November quarter misses the implied $34.9B; a major accelerator customer is disclosed as dual-sourcing; the tax rate normalises toward the mid-teens; and the market re-rates a business whose top five customers are 45% of revenue and whose tangible book is negative. Note $330 is 15.9% below the 200-DMA of $366.52 and below the street's lowest target of $400. −21.1%.
• Base ~$470 — 24.0x FY2027E. Cross-checks: 40.6x FY2026E; 18.0x FY2028E; 57.8x trailing analyst-basis EPS. Sits 7.2% below the street consensus of $506.50 and 9.6% below the median of $520. The scenario: FY2026 lands close to consensus, the contracted book converts at the disclosed 30%, gross margin holds near 69% against the mix headwind, and the tax rate normalises gradually rather than abruptly. +12.4%, plus ~0.6% dividend.
• Bull ~$590 — 30.2x FY2027E. Cross-checks: 50.9x FY2026E; 22.6x FY2028E; essentially the street high of $582. The scenario: the order book grows again in September, the Apple agreement is the first of several non-hyperscaler wins, operating margin holds above 49%, and the market pays a premium multiple for the only merchant supplier of custom accelerators at scale. +41.1%.

Base is 12.4% above spot; asymmetry roughly 1.9:1 to the upside (21% down, 41% up). The destination supports ownership. A 6.61% single-session gap on a sector-wide move argues against paying for all of it tonight. Buy — Tactical, staged.

7. Knowledge base — 22 claims, thin, and the bears are the better voices

Tracked knowledge-base claims naming Broadcom: 26 raw hits, of which 22 are entity-level. Searches covered AVGO (case-sensitive, 11 matches), Broadcom (9), VMware (2) and Hock Tan (0) across all 51,928 records.

Discards and weighting, stated in full because the lane is small enough that the arithmetic matters. Four hits were thesis- or category-level only — the company was named inside a claim about something else (an artificial-intelligence exchange-traded-fund basket, a hyperscaler accelerator battle, an enterprise-server product, a video-compression patent dispute) — and are excluded from the entity count. One claim tagged to a quarantined-misattribution source is excluded outright under the standing attribution policy. Eight of the remaining 21 are management-lane (skill 0.5) and half-weighted: six from Broadcom's own investor communications, one from a competing chip company's management, one from the management of a large model-developer customer. That leaves thirteen independently-weighted claims. We report this honestly: this is a thin lane and it does not carry the verdict.

Stance mix: 9 bullish, 7 neutral, 6 bearish across 11 distinct sources, spanning 2017-12-12 to 2026-07-08. Net conviction mixed.

The bull lane is dominated by management framing, and we half-weight all of it:

> 2026-06-03 · AVGO_mgmt (skill 0.5) · bullish · conviction 92

> "Q3 semiconductor revenue from AI expected to grow over 200% YoY to $16.0 billion on custom accelerator and networking demand."

> 2026-06-03 · AVGO_mgmt (skill 0.5) · bullish · conviction 90

> "Q3 FY26 revenue guided to approximately $29.4 billion, an increase of 84% year-over-year."

> 2026-06-03 · AVGO_mgmt (skill 0.5) · bullish · conviction 80

> "Q3 non-GAAP operating margin guided stable at ~67% of revenue, reflecting strong operating leverage."

These are company statements about company results and carry no independent evidentiary weight. They are useful only as a cross-check: the $29.4B guide matches the vendor's $29.248B consensus for 2026-09-03 within 0.5%, which tells us the street has simply adopted the guide. The 67% figure is a non-GAAP operating margin and should not be compared with the 49.0% GAAP margin in Section 3 — the gap is exactly the amortisation and stock compensation of Section 2.

The genuinely independent bull claim is about people, not numbers:

> 2026-07-08 · gavin_baker · bullish · conviction 60 · skill 1.0

> "Only Jensen (Nvidia), Hock Tan (Broadcom), and Lisa Su (AMD) reliably identify and retain exceptional engineers/executives; most other semiconductor companies fail to retain top talent."

The bear lane is where this file earns its keep, and it attacks the correct pressure point:

> 2025-12-09 · invest_like_the_best · bearish · conviction 70 · skill 1.0

> "Broadcom takes 50-55% gross margin on [a hyperscaler's custom accelerator] back-end (~$15B); it becomes economically rational for [that customer] to bring the program in-house."

> 2025-12-09 · invest_like_the_best · neutral · conviction 72 · skill 1.0

> "[The same customer] bringing in MediaTek—a lower-margin Taiwanese ASIC provider—is a warning shot across Broadcom's bow on pricing."

> 2025-10-30 · gavin_baker · bearish · conviction 70 · skill 1.0

> "Most AI ASICs will fail; in the next 3 years a bunch of high-profile ASIC programs will be cancelled, especially if [the hyperscaler] starts selling [its accelerators] externally."

> 2026-04-15 · jensen_huang · bearish · conviction 78 · skill 1.0

> "[Two hyperscalers'] custom-accelerator growth is ~100% [one model developer]—a unique instance, not a trend; there is no abundance of ASIC opportunities."

Take the source of that last claim into account — a direct competitor arguing that the competing architecture has no market — but do not dismiss it, because the accompanying claim is a pricing argument that is checkable:

> 2026-04-15 · jensen_huang · neutral · conviction 70 · skill 1.0

> "Custom ASICs save little: ASIC margins (~65%, paid to Broadcom) rival [the merchant GPU vendor's], and you still must build something better."

The bear case reduces to one proposition: Broadcom's customers are the most sophisticated hardware buyers on earth, they know what they are paying, and the same 65-69% gross margin that makes this stock attractive is the incentive for them to leave. Nothing in the filings falsifies it. The counter-evidence is the $164.6B of firmly committed multi-year contracts and the Apple agreement running through 2031 — both of which say customers are locking in for years rather than preparing to exit. That is the real debate, and it is unresolved.

The highest-skill voice in the file (skill 2.0) is ambivalent and has been for six months:

> 2026-01-11 · jordi_visser (skill 2.0) · bearish · conviction 60

> "Nvidia and Broadcom flat while other semis rally 25%; leadership rotating away from the last three years' momentum winners."

> 2026-02-01 · jordi_visser (skill 2.0) · neutral · conviction 55

> "Large-cap semis (Nvidia, Broadcom, Intel) have already had good runs and can still rise on chip shortage, but energy is the better trade now."

> 2026-04-25 · jordi_visser (skill 2.0) · bullish · conviction 85

> "The shift to agentic/inference [artificial intelligence] (east-west traffic) broadens the semi trade beyond GPUs to memory, CPUs, networking and packaging — Micron, Marvell, TXN benefit; compute shortage runs the rest of the year."

Note that the January claim — Broadcom flat while other semiconductors rally — describes exactly the three-month relative performance in Section 9 (+0.4% against SPY's +7.6%), and that the April claim describes exactly the sector-wide move that lifted the stock 6.61% today alongside Marvell's 12.8% and Micron's 7.6%. The lane has been directionally useful about the sector rotation even where it has been non-committal on the equity.

Two older claims worth recording for perspective:

> 2023-05-23 · business_breakdowns · bearish · conviction 75 · skill 1.0

> "Legacy AV (60% of market) is being pillaged; Broadcom's acquisition of Symantec abandoned most customers, ceding share to CrowdStrike."

> 2017-12-12 · arthur_hayes · bearish · conviction 55 · skill 1.2

> "Absent fundamental technical innovation, chip incumbents trend toward commodity economics — thin margins, high velocity, high volume."

The 2023 claim is a description of the Broadcom acquisition playbook — buy an installed base, raise price, accept churn at the low end — which is precisely what the VMware integration has done, and which the FY2025 numbers show works financially. Whether it works indefinitely is the long-horizon question.

Why Moderate rather than higher or lower. The lane is thin (22 claims, 11 sources, and nearly 40% of it is management framing), and no tracked source has made a dated, falsifiable price call on this equity — the single most valuable thing a knowledge base can contain, and it is absent here. What lifts it above Low is that the bearish claims are specific, from high-skill sources, and aimed at the correct vulnerability. The knowledge base is an input to the risk score, not to the fair value.

8. Data integrity — what we rejected from the vendor file and why

Nine items. Four are direct filing-versus-vendor contradictions, and in every one the filing wins.

1. FILING OVERRIDES VENDOR — quarterly gross profit is understated by ~$0.5B in every FY2026 quarter. The vendor reports gross profit of $12.663B (65.6% margin) for the quarter ended 2026-02-01 and $14.919B (67.2%) for the quarter ended 2026-05-03. The 10-Q filed 2026-03-11 states "Gross margin was $13,157 million for the fiscal quarter ended February 1, 2026"; the 10-Q filed 2026-06-09 states "Gross margin was $15,415 million for the fiscal quarter ended May 3, 2026 ... and $28,572 million for the two fiscal quarters ended May 3, 2026 ... As a percentage of net revenue, gross margin was 69% for each of the fiscal quarter and two fiscal quarters ended May 3, 2026, and 68% ... ended May 4, 2025." The vendor is short by $494M and $496M respectively. The defect is confined to FY2026 — all FY2025 quarters reconcile to the dollar ($10,145M, $10,197M, $10,703M, $12,249M all match), as does the FY2025 annual figure of $43,294M. We use the filing throughout, which reverses the apparent margin compression into a genuine expansion from 68% to 69%.

2. FILING OVERRIDES VENDOR — the payload contains no remaining-performance-obligation field, and this is the most important number about the company. The 10-K discloses $33.3 billion at 2025-11-02 with 35% expected within twelve months; the 10-Q discloses $164.6 billion at 2026-05-03 with 30% expected within twelve months, and names "a long-term contract for custom AI accelerators entered in the fiscal quarter ended May 3, 2026." A 394% increase in the contracted order book in two quarters, invisible to any model built on the vendor file.

3. FILING OVERRIDES VENDOR — a $29 billion contingent exposure appears nowhere in the payload. The 10-Q subsequent-events note discloses a backstop of a customer's AI-rack lease obligations entered 2026-06-08 "with a maximum exposure of $29 billion." It is 43% of on-balance-sheet debt and is in no leverage ratio, no km_ttm field and no risk score.

4. FILING OVERRIDES VENDOR — the company description is the FY2018 structure and the headcount contradicts itself. The DESC field describes "four primary divisions: Wired Infrastructure, Wireless Communications, Enterprise Storage, and Industrial & Other" and "19,000 full-time staff." The 10-K: "We have two reportable segments: semiconductor solutions and infrastructure software." The profile.employees field in the same payload says 33,000. We use the filing's segment structure and the profile's headcount, and reject the DESC field entirely.

5. Every est EBITDA, EBIT and net-income row is REJECTED — six of eight rows fail an elementary sanity test. netIncomeAvg exceeds ebitdaAvg in the FY2026, FY2027, FY2028, FY2029 and FY2030 rows, which is arithmetically impossible for a company that pays interest and tax. FY2030 is the most extreme: net income $169.855B against EBITDA of $74.846B. FY2023 is nonsense in a different way — net income $127.176B against EBITDA of $20.927B and EPS of $27.348 in a year when the company actually earned $3.30. In addition, the FY2030 revenue row of $139.000B is LOWER than the FY2029 row of $290.277B, a further impossibility. We use epsAvg and revenueAvg only, and we use no revenue figure beyond FY2029. Every forward multiple in this dive is computed from epsAvg.

6. The est rows are on the ADJUSTED basis, not GAAP, and the file never says so. FY2025 epsAvg of $6.748 against GAAP actual of $4.77 and an implied analyst-basis actual of ~$6.82 settles it. Comparing FY2027E of $19.546 against trailing GAAP EPS of $6.00 would imply a growth rate that does not exist. All forward multiples here are analyst-basis to analyst-basis, and the trailing analyst-basis figure of $8.13 is stated alongside.

7. netDebt is CORRECT and the lease lines are CORRECT — verified, not assumed. netDebt of $48.958B equals total debt $65.136B less cash and short-term investments $16.178B; short-term investments are genuinely zero. capitalLeaseObligationsCurrent and capitalLeaseObligationsNonCurrent are both 0.00 and do not duplicate longTermDebt — the 10-K confirms Broadcom has operating leases only, with expense of just $182 million in FY2025. We checked specifically for the finance-lease double-count that afflicts this feed and it is not present. The filing does refine the debt figure upward to $67.120B of principal (against $65.136B carrying), and updates it to $66.720B at 2026-05-03 with cash of $19.628B — we use the 10-Q figures for the current balance sheet.

8. Enterprise value is sound. enterpriseValueTTM of $2,034.709B against a rebuilt $2,038.388B is a −0.2% residual, the cleanest in this batch. Accepted. We do not lean on evToEBITDATTM of 48.4x because the EBITDA denominator carries the amortisation issue of Section 2.

9. Two minor internal inconsistencies, flagged once. (a) quote.yearHigh of $495.00 and yearLow of $281.87 are intraday extremes; tech.hi52 of $481.57 and tech.lo52 of $289.60 are closing extremes. We use the closing series throughout, so the stock is 13.2% below its 52-week high rather than 15.5%. (b) quote.sma50 of $395.31 and tech.sma50 of $394.87 differ trivially; we use the tech block for consistency.

Not defects, correctly reported: the 69.7x trailing price-to-earnings ratio, the 22.6x price-to-book, the negative tangible book value per share of −$8.10, the 1.65% free-cash-flow yield and the 3.8% effective tax rate all look wrong for a company of this quality and are all correct — they are the honest arithmetic of purchase accounting, heavy equity compensation and a favourable tax posture, and they are explained rather than smoothed in Sections 2 and 4.

Non-equity tripwire: checked and clear. AVGO is common stock, beta 1.462, average volume 25.7M shares, a variable and rising dividend ($2.54 trailing, $2.60 declared forward), and a 52-week closing range of $289.60-$481.57 — a 66% band. Nothing resembling a preferred or baby-bond profile.

9. Technicals — and what a 6.61% day means for the entry

• Price $418.16, UP 6.61% (+$25.93) from $392.23, on volume of 28.1M against a 25.7M average — an active but not extraordinary session. Intraday range $400.68-$422.07; the stock opened at $402.97, i.e. it gapped 2.7% at the open and added another 3.8% through the day.
• This was a sector event, not a company event. On the same session Marvell rose 12.8%, Intel 10.9%, Lam Research 7.9%, Micron 7.6% and AMD 7.0%. Broadcom made no disclosure today. The move is real — it is not a data error — but it contains no new information about this company, and that distinction is the whole of the entry argument.
• −13.2% from the 52-week closing high of $481.57; +44.4% above the 52-week low of $289.60. Maximum drawdown from peak over the trailing year is −13.2% on closing prices.
• Above both moving averages, but not extended: +5.9% over the 50-DMA of $394.87 and +14.1% over the 200-DMA of $366.52. Compare this with the most stretched name in the batch, which sits 22.5% above its 50-day line.
• RSI 59.4 — neutral, not overbought. MACD +1.78, barely positive. Even after a 6.61% day the momentum readings are unremarkable, which is a direct consequence of the flat three months that preceded it.
• Relative performance, and the split that defines this name: 12-month +44.9% against SPY +24.3% and QQQ +30.8% (ahead of both); 6-month +26.3% against SPY +11.1% and QQQ +15.7% (well ahead); 3-month +0.4% against SPY +7.6% and QQQ +7.7% (−7.2 points, a clear lag). The stock did nothing for three months and then made a quarter of a year's move in one afternoon.
• Practical read, and it is the operative point of this dive. Yesterday's close of $392.23 and the 50-DMA of $394.87 sit within 0.7% of each other, 5.6-6.2% below spot. Gaps produced by sector-wide sentiment rather than company news fill more often than gaps produced by earnings. Buying the whole position at $418.16 means paying a 6.6% premium for information you already had at lunchtime. The disciplined structure is an initial tranche now — because the $164.6B order book is a genuine reason to own the shares and the September print could just as easily gap it higher — with the balance reserved for a retracement into the $392-400 band and for the 2026-09-03 result. That is Buy — Tactical by definition, and the "Tactical" is doing real work here.

10. Insiders — consistent, programmatic selling into strength

Every transaction in the window is a disposal. There is not one open-market purchase. That is the honest headline and it should be stated before the mitigation.

The mitigation is real and we do not over-read this table. The Chief Legal Officer's three sales are identical 25,000-share blocks at roughly two-week intervals — the signature of a pre-arranged Rule 10b5-1 trading plan, not a decision made in June. Henry Samueli, a co-founder, sold 56,599 shares against a holding of approximately 36.9 million — 0.15% of his position, which is portfolio housekeeping at a scale that means nothing. The two director sales are small (1,890 and 1,602 shares).

The one observation worth making is about price. Every sale in this table executed between $373.86 and $401.33. The stock closed today at $418.16. Insiders were programmatically selling at prices 4-11% below where the shares now trade — which tells you nothing about their view (10b5-1 plans are price-blind by design) but does tell you that today's level is above where the company's own officers were transacting six weeks ago. Weak evidence, weighted weakly. There is no signal here in either direction, and we would say so equally loudly if the table were full of purchases.

11. Verdict, kill-criteria and flip conditions

Buy — Tactical.

The forward evidence is the best in this batch on a contractual basis. The entry, tonight, is the worst.

What the filings established, and what the vendor could not see. Remaining performance obligations of $164.6 billion at 2026-05-03, against $33.3 billion at 2025-11-02 — a 394% increase in two quarters, explicitly including "a long-term contract for custom AI accelerators entered in the fiscal quarter ended May 3, 2026," with 30% expected to convert within twelve months (~$49.4B against a FY2025 revenue base of $63.887B). Revenue growing 47.9% in the May quarter. Gross margin expanding to 69% — the vendor says it contracted and the vendor is wrong. Operating margin at a record 49.0%. Free cash flow of $26.914B on capital expenditure of 1.0% of revenue. Return on equity 36.4%, return on invested capital 19.5%. An 8-K dated 2026-07-06 extending the Apple relationship "through 2031" across "multiple generations of Apple products." Fifty-nine covering analysts, zero sell ratings.

What you are accepting, stated without softening. Trailing GAAP diluted EPS of $6.00 — a 69.7x multiple. $8.201B of acquisition-intangible amortisation and $7.568B of stock-based compensation, 11.6% of revenue, with $20.106B still to be expensed. A diluted share count that rose 0.8% despite $8.450B of buybacks, so the only real shareholder yield is a 0.6% dividend. Top-five customers at ~45% of revenue and one distributor at 32%, both rising. Goodwill and intangibles at 76% of assets and tangible book value of NEGATIVE $8.10 per share. $66.720B of debt. A $29 billion maximum-exposure backstop on one customer's five-year lease obligations, disclosed in a subsequent-events note and present in no ratio anywhere. A 3.8% effective tax rate that the company's own 10-K warns will be materially affected in fiscal 2026. And a FY2026 consensus whose implied November quarter — roughly $34.9B against $22.187B in May — nobody has guided to.

Why Tactical rather than Core, and why not Watch. Base fair value of $470 is 12.4% above spot and 7.2% below the street's $506.50 — we are the conservative party here, deliberately. That is a real but not commanding margin, and it arrives on the evening of a 6.61% single-session gain in which the company disclosed nothing and five semiconductor peers rose between 7.0% and 12.8%. The reason this is not Watch is that the September print is 30 days away and is a genuine two-sided catalyst: an order book that grows again would justify chasing, and there is no guarantee the gap fills first. The reason it is not Core is everything in the preceding paragraph. Establish the position; do not complete it at $418.16.

Pre-registered KILL criteria — what would break the thesis:

• Remaining performance obligations falling below ~$140B at any print without a matching revenue conversion, or the 12-month recognition share falling materially below 30%.
• Any disclosure that a top-five customer is dual-sourcing or insourcing a custom-accelerator programme. This is the tracked knowledge base's central bear claim and it remains unfalsified.
• Top-five customer concentration passing 50%, or the single-distributor figure passing 35%. Concentration rising while revenue rises is the fragile version of this story.
• A drawdown on, or an increase above, the $29 billion backstop. Vendor financing that starts being used is vendor financing that was needed.
• Gross margin falling below ~66% as the semiconductor mix builds — the 10-Q already names this as the offsetting force.
• The November quarter landing materially below $30B, which would break the FY2026 consensus and, more importantly, the FY2027 base it is built on.
• The effective tax rate stepping above ~18% without an offsetting revenue beat.

Pre-registered FLIP TO BUY — CORE (upsize) criteria:

• A retracement into the $392-400 band (yesterday's close $392.23; 50-DMA $394.87) with the order book intact — this is the most likely path to a full-size position and the explicit reason capital is held back.
• Remaining performance obligations above ~$200B at the 2026-09-03 print with the 30% conversion share maintained.
• A November-quarter guide at or above $33B, which would remove the single most fragile assumption in Section 6a and raise our base case directly.
• Top-five customer concentration FALLING while revenue grows — the clearest possible evidence that the franchise is broadening beyond a handful of buyers.
• A second named multi-year, multi-generation design win of the Apple type outside the hyperscaler set.
• The diluted share count actually declining for two consecutive quarters, which would mean the buyback has finally outrun the equity grants.

Where AVGO fits in the Synthos Framework Portfolio. The artificial-intelligence infrastructure sleeve, at a 2% target weight built in three tranches — approximately one third now, one third on any retracement into $392-400, one third reserved until after the 2026-09-03 print. The house flagship-selection rule — pick forward next-exponentials rather than trailing compounders — is satisfied here in an unusually literal way: the case rests on a contracted order book disclosed eight weeks ago, not on the +44.9% twelve-month return. The house rule that mega-cap incumbents are guilty until proven innocent is also satisfied, and the proof offered is a filing, not a narrative. The counterweight to a larger weight is concentration: this is a 2% position, not a 4% one, because five customers are 45% of the revenue and one of them could take the roadmap in-house.

On overlap within the batch. Broadcom and Meta Platforms (also covered 2026-08-04) sit on opposite ends of the same trade. Broadcom sells artificial-intelligence infrastructure, is up 44.9% over twelve months, gapped 6.61% today, and trades at 69.7x trailing GAAP earnings with a contracted book five times its annual revenue. Meta buys that infrastructure, is down 21.6% over twelve months, has an RSI of 26.2, and trades at 22.2x trailing earnings. They are not redundant; they are two sides of the same capital cycle, and holding both is a deliberate hedge against being wrong about which side captures the economics. Logged as a tracked Synthos call (Buy — Tactical) as of 2026-08-04 at $418.16.

Single biggest risk: that the customers who buy the designs learn to make them. Broadcom earns a reported 69% gross margin selling custom silicon to five buyers who are among the most capable engineering organisations on earth and who represent 45% of its revenue. The tracked knowledge base states the mechanism precisely — at a 50-55% margin on a programme of that size, "it becomes economically rational" to insource — and notes that at least one customer has already introduced a lower-cost second supplier. The $164.6B contracted book and the Apple agreement through 2031 are the counter-evidence, and they are strong. But a contract is a delay, not a moat.

Most fragile assumption in the price: that the November 2026 quarter delivers approximately $34.9 billion of revenue. That figure is not guided by the company, not contracted in the disclosed order book, and not derivable from anything in the filings — it is the residual left when the guided August quarter is subtracted from the annual consensus. It requires 57% sequential growth over the May quarter. Every FY2027 and FY2028 estimate in this dive is built on a base that includes it, and if it is wrong by even 15% the entire forward multiple structure re-rates.

Provenance & disclosures

• Traceability: 22 entity-level knowledge-base claims name Broadcom (26 raw hits; 4 thesis-only mentions excluded — an artificial-intelligence exchange-traded-fund basket claim, a hyperscaler accelerator-battle claim, an enterprise-server product claim and a video-compression patent claim, in each of which the company was named inside an argument about something else; breadth 11 distinct sources; net conviction mixed). Stance mix 9 bullish / 7 neutral / 6 bearish, spanning 2017-12-12 to 2026-07-08. One claim tagged to a quarantined-misattribution source is excluded outright under the standing attribution policy. Eight management-lane claims (skill 0.5) are half-weighted and identified wherever cited — six from the company's own investor communications, one from a competing chip company's management, one from a large model-developer customer's management; none is used as a valuation input. That leaves thirteen independently-weighted claims, which is thin, and the dive says so. Searches covered AVGO (case-sensitive, 11 matches), Broadcom (9), VMware (2) and Hock Tan (0) across all 51,928 records. All quoted claims are verbatim from the stored claim text, with bracketed substitutions only where a house naming restriction applies.
• Data as-of: fundamentals — fiscal Q2 2026 ended 2026-05-03, 10-Q filed 2026-06-09; the last complete fiscal year is FY2025, ended 2025-11-02, 10-K filed 2025-12-18 · estimates 2026-08-04 · prices 2026-08-04 ($418.16, +6.61%; 50-DMA $394.87; 200-DMA $366.52; RSI 59.4) · knowledge-base claims 2026-08-04. Every figure comes from the Synthos vendor data file for AVGO or from the SEC filings in filings/AVGO/; no figure comes from memory, recall or external retrieval.
• Filings consulted: 10-K filed 2025-12-18 (FY2025, year ended 2025-11-02); 10-Q filed 2026-03-11 (quarter ended 2026-02-01); 10-Q filed 2026-06-09 (quarter ended 2026-05-03 — the primary source for this dive); 8-K filed 2026-06-11 (debt tender offers launched); 8-K filed 2026-06-18 (tender-offer pricing, upsize and results); 8-K filed 2026-07-06 (Apple collaboration extended through 2031 for custom ASIC silicon).
• Filing-versus-vendor overrides (four, all resolved in favour of the filing): (1) quarterly gross profit — the vendor reports $12.663B and $14.919B for the February and May 2026 quarters; the 10-Qs report $13,157M and $15,415M, a consistent ~$0.5B understatement confined to FY2026, and the correction reverses an apparent margin compression into a 68%-to-69% expansion; (2) remaining performance obligations — $164.6B at 2026-05-03 against $33.3B at 2025-11-02, disclosed in the filings and absent from every vendor field; (3) the $29 billion maximum-exposure lease backstop entered 2026-06-08 and disclosed in the 10-Q subsequent-events note, absent from every vendor field and every leverage ratio; (4) the company description — the vendor's DESC field carries the abandoned FY2018 four-division structure and a 19,000 headcount that contradicts the 33,000 in its own profile block, against the 10-K's "two reportable segments: semiconductor solutions and infrastructure software." A fifth, minor refinement: the filings state $67.120B of principal indebtedness at fiscal year end and $66.720B at 2026-05-03 with $19.628B of cash, which we use in preference to the carrying values on the annual balance sheet.
• Other data rejections (detailed in Section 8): every est EBITDA, EBIT and net-income row REJECTED — netIncomeAvg exceeds ebitdaAvg in five consecutive forward rows (FY2026 through FY2030, most extremely $169.855B against $74.846B in FY2030) and the FY2023 row is nonsense; the FY2030 revenue row of $139.000B is lower than FY2029's $290.277B and is rejected outright; forward valuation runs on epsAvg only, and no revenue figure beyond FY2029 is used. The est rows are additionally identified as adjusted-basis, not GAAP (FY2025 epsAvg $6.748 against GAAP actual $4.77), which the file never states. Intraday 52-week extremes ($495.00/$281.87) rejected in favour of the closing series ($481.57/$289.60).
• Vendor fields specifically VERIFIED as correct (unusual enough to record): netDebt of $48.958B is arithmetically right and discards nothing, because short-term investments are genuinely zero; capitalLeaseObligations of 0.00 in both current and non-current positions is right and does not duplicate longTermDebt, confirmed against the 10-K lease note showing operating leases only with $182M of annual expense; enterpriseValueTTM carries a −0.2% residual against a rebuild; and seg_prod and seg_geo both reconcile exactly to reported revenue and match the filing's two-segment structure.
• Estimate coverage: FY2026 27 analysts on EPS and 32 on revenue; FY2027 28 and 35 — the primary forward anchor; FY2028 13 and 27; FY2029 5 and 13; FY2030 5 and 18. No conclusion in this dive rests on the FY2029 or FY2030 rows. The FY2027 EPS dispersion is wide ($16.608 low to $22.884 high, a 38% spread on 28 estimates), which is itself a statement about how uncertain the accelerator ramp is.
• Peer note: the vendor peer set contains five legitimate comparables (Taiwan Semiconductor $2,163.2B, Micron $1,008.2B, AMD $845.6B, ASML $659.8B, Analog Devices $185.2B) and four that are not comparable to a $1.99 trillion franchise (Semtech $12.0B, FormFactor $9.4B, Qorvo $8.4B, Power Integrations $3.6B). Meta Platforms also appears in the list, which is a classification error. No peer multiples are supplied in the file, so no peer-multiple comparison is drawn.
• Fair-value caveat: the $330 / $470 / $590 anchors are multiples of FY2027 consensus analyst-basis EPS of $19.546 (16.9x / 24.0x / 30.2x), cross-checked against FY2026E of $11.590 and FY2028E of $26.136. Stated arithmetic, not a discounted cash flow. The base case requires the forward multiple to be broadly maintained rather than expanded, and it sits 7.2% below the street consensus of $506.50 — we are deliberately the conservative party, for the reasons in Section 6b.
• Timing: the last reported quarter was fiscal Q2 2026, announced 2026-06-03 and filed 2026-06-09. The next print is 2026-09-03, 30 days away — inside the tactical horizon but outside the five-day window, so no earnings banner applies; the timing is nonetheless an explicit input to the staged-entry structure and to the Tactical rather than Core designation. Today's 6.61% advance is a genuine, sector-wide session move (Marvell +12.8%, Intel +10.9%, Lam Research +7.9%, Micron +7.6%, AMD +7.0%), not a data error, and it is treated as an entry problem rather than as information.
• Accessibility note: all tables are labelled, all emphasis is carried by text rather than colour, and every figure is stated in full rather than encoded in a chart.
• Not investment advice. Independent research, educational and informational only, never personalised. No recommendation to buy, sell or hold any security is made to any person.
• Version: 2026-08-04-full.