Phase Space AI

Financial Model Notes

Marvell Technology [MRVL]

Marvell Technology, Inc. [MRVL] — Financial Model Notes

Derivations for every figure in MRVL_Research.md and MRVL_Valuation.md. Computed figures are marked derived with the arithmetic shown.

---

1. Quarterly revenue reconstruction

Marvell files no Q4 10-Q, so Q4 revenue is derived as the fiscal-year total less the three reported quarters, then cross-checked against the Q4 earnings release.

Prior year, for the YoY comparison:

| Q1 FY2025 | 2024-05-04 | 1,160.9 | 10-Q | | Q2 FY2025 | 2024-08-03 | 1,272.9 | 10-Q | | Q3 FY2025 | 2024-11-02 | 1,516.1 | 10-Q | | Q4 FY2025 | 2025-02-01 | 1,817.4 | derived: 5,767.3 − 1,160.9 − 1,272.9 − 1,516.1. Cross-check: Q4 release states "$1.817 billion" ✓ |

TTM revenue (Q2 FY26 → Q1 FY27) = 2,006.1 + 2,074.5 + 2,218.7 + 2,417.8 = $8,717.1m. Matches scan_v3. Prior-year TTM (Q2 FY25 → Q1 FY26) = 1,272.9 + 1,516.1 + 1,817.4 + 1,895.3 = $6,501.7m. TTM YoY growth = 8,717.1 / 6,501.7 − 1 = +34.1%. Latest-quarter YoY = 2,417.8 / 1,895.3 − 1 = +27.6% ✓ matches scan_v3. Q-annualised run-rate = 2,417.8 × 4 = $9,671.2m (+10.9% on TTM).

---

2. Operating income and gross profit

TTM EBIT = 290.1 + 357.8 + 404.4 + 339.4 = $1,391.7m → TTM operating margin 16.0%. TTM gross profit = 1,010.6 + 1,069.8 + 1,147.9 + 1,260.8 = $4,489.1m → TTM gross margin 51.5%. Latest quarter: EBIT margin 339.4 / 2,417.8 = 14.0%; gross margin 1,260.8 / 2,417.8 = 52.1%. FY2026: 1,322.9 / 8,194.6 = 16.1% — the scan_v3 figure, correct for its period.

The prior-year losses, and what they do to the margin-delta metric

scan_v3 reports op_margin_delta_pp: 28.6, which reproduces as 16.1 − (−12.5) = 28.6. Arithmetically correct, materially misleading: FY2025's loss includes a large goodwill impairment concentrated in Q3 FY2025, where operating income was −$702.8m on $1,516.1m of revenue and $349.4m of gross profit. The underlying move is from roughly break-even to 16%, not 28.6pp of operating leverage.

Q2 FY2027 guidance implies a 17.0% GAAP operating margin: 2,700 × 52.6% (guided GAAP gross margin midpoint) − $960m (guided GAAP opex) = $460m. This is the basis for the 17.0% terminal margin used in the reverse DCF.

---

3. Net cash / net debt

At 2026-05-02, from the Q1 FY2027 10-Q balance sheet:

scan_v3 reported −$1,379.1m, reproducible as 3,843.6 − 4,961.3 − 261.4, i.e. it included operating-lease noncurrent and omitted operating-lease current. A treatment difference, not an error of fact; variance $261.4m, or 0.16% of EV.

The material caveat: $2.0bn of the $3,843.6m cash arrived on 2026-03-31 from NVIDIA for the Series A Convertible Preferred — inside the reporting quarter. Net financial debt would be −$3,117.7m without it. Counting that cash while using a pre-conversion share count credits the proceeds twice. The two internally consistent pairs are:

The 915m pair is used because it matches how the company will report EPS from Q2 FY2027.

EV = 915.0m × $174.36 + $1,117.7m = $159,539m + $1,118m = $160,657m.

---

4. Share count — the defect and its resolution

At $174.36 the 40.2m difference is $7,009m of equity value and, with net debt unchanged, $6,748m of enterprise value — a 4.4% understatement of the denominator in every multiple the screen computed.

The preferred is common equity for valuation purposes. 10-Q Note 10, verbatim: "The Company's Series A Preferred Stock represents a second class of common stock for purposes of computing net income per share under the two-class method. This determination reflects that the Series A Preferred Stock does not have any material preferential rights relative to the Company's common shares, and its rights and privileges are identical to common shares, except for certain voting rights." Holders receive dividends on an as-converted basis and participate pro rata in liquidation. There are no redemption or preemptive rights.

NVIDIA's economics: paid $2.0bn; 21.778m shares at the $174.36 spot are worth ~$3,796m — a ~90% gain before conversion. As-converted dilution to common is 21.778 / 847.3 = 2.6%.

EPS cross-check

Q1 FY2027 net income $34.5m ÷ 893.3m diluted = $0.0386 against filed diluted EPS $0.04 ✓. FY2026: net income $2,670.1m, diluted EPS $3.07 → implies ~870m diluted WA ✓ consistent with the pre-preferred count.

Not dual-class in the ordinary sense, but the two-class EPS method now applies, which is precisely why the scan's single share-count field is wrong.

---

5. Data-centre end-market series

Sourced from the quarterly earnings releases, each of which states current quarter, year-ago quarter and prior quarter. No derivation.

TTM data centre (Q2 FY26 → Q1 FY27) = $6,492.4m = 74.5% of TTM revenue. Prior-year TTM = 880.9 + 1,101.1 + 1,365.8 + 1,440.6 = $4,788.4m. Contribution to TTM growth = 1,704.0 / 2,215.4 = 76.9%.

Segment-reporting change to note. Marvell states: "Beginning in the fourth quarter of fiscal 2026, the Company consolidated revenue previously reported separately as enterprise networking, carrier infrastructure, consumer and automotive/industrial end markets into a new communications and other end market… The composition of our data center end market remains unchanged." The data-centre series above is therefore continuous across the reclassification; the non-data-centre series is not.

AI revenue is not disclosed. The data-centre end market contains custom compute (XPUs), electro-optics, PAM4 DSPs, data-centre interconnect, switching and storage. No split is given and none is asserted here.

---

6. Corrected trailing CAGR

scan_v3 base: cagr_base_revenue = $5,919.6m = FY2025 revenue by Marvell's labelling, period ended 2023-01-28 (10-K). Endpoint: TTM $8,717.1m at 2026-05-02. Interval 1,190 days = 3.258 years.

(8,717.1 / 5,919.6)^(1 / 3.258) − 1 = 12.6%

scan_v3 reports 11.449588% — bit-identical to scan_v2, whose TTM endpoint was $8,393.7m. The CAGR was not recomputed when TTM was corrected. Understated by 1.2pp. Same fingerprint on AVGO (2.0pp) and AMD (1.7pp).

---

7. Own-multiple history

Daily EV/Sales, 2022-03-10 → 2026-07-28, 1,099 trading days. Alpaca split-adjusted daily bars; TTM revenue as known at each date, lagged to filing dates; current shares (915m) and net debt held constant.

Momentum: 12-1 = +249.7%; trailing 12-month total return +129.4%.

EV/EBIT is declared UNIDENTIFIED as a valuation anchor. The series spans 291 trading days and straddles two years of operating losses, so the distribution crosses zero: the 25th percentile is −86.5x and the median +74.9x. A percentile drawn from a distribution containing negative multiples carries no information. Per valuation.md this is declared rather than papered over with a peer median. For reference: today's 115.4x sits at a nominal 59.5th percentile of that unusable distribution.

---

8. Reverse-DCF parameterisation

assets/reverse_dcf.py, EBIT basis. Solved: revenue CAGR. Fixed: EV $160,657m; revenue $8,717.1m; 5 years; WACC 10.0%; terminal margin 17.0%; exit multiple 20.0x.

Result 54.2%, implying FY2031 revenue of $76,100m — an 8.7x increase from $8.7bn.

Terminal margin justification: 17.0% is above TTM 16.0%, above FY2026 16.1%, and well above the latest quarter's 14.0%. It equals the Q2 FY2027 guide implied by the company's own gross-margin and opex guidance. Deliberately not conservative, so the FAIL cannot be attributed to the margin input.

Exit multiple: 20.0x, the growth-matched anchor from Tier-1 (n = 537). At a 17% margin that is 3.4x EV/Sales, below Marvell's own historical minimum of 5.62x — a stringent exit, which is why the sensitivity extends to 2.0x the anchor (40x EBIT = 6.8x sales, comfortably inside the historical range). It still fails there, requiring 34.3% against a 27.6% latest quarter.

Terminal value is 100% of modelled EV, above the 60% threshold, so the reverse DCF is the primary long-horizon output and sensitivity runs over the exit multiple.

---

9. Accounting-quality figures

GAAP EPS of $3.07 exceeding non-GAAP EPS of $2.84 for FY2026 is the diagnostic: for a company carrying $942m of annual intangible amortisation and ~$600m of SBC, that inversion is only possible via a one-off gain, and Marvell's non-GAAP definition explicitly excludes "gain on sale of business." $1.8bn of the $2.67bn GAAP net income is the divestiture gain, booked in a line an automated screen would read as recurring. This memo uses operating income throughout, never net income.

---

10. Data-quality flags

1. Share count 874.8m against the company's own guided 915m diluted — understates equity value by $7.0bn (4.6%) and EV by $6.75bn (4.4%). The correct figure is published in the same release the revenue came from.
2. growth_trend: "ACCELERATING" — data-centre YoY fell +98% → +21% across five quarters and total revenue +63% → +22%. The label is derived from a two-point comparison of latest-quarter YoY against a stale CAGR. Correct characterisation: decelerated sharply, now re-accelerating off a lower base, with company guidance forecasting further acceleration.
3. revenue_cagr_demonstrated stale, carried from scan_v2; understates by 1.2pp.
4. op_margin_delta_pp: 28.6 measured off an impairment-depressed base (FY2025 −12.5% including a Q3 goodwill impairment). Arithmetically right, materially misleading.
5. Terminal margin capped at 15.6% from 16.1% — nearly harmless here, decisive for AVGO under the same rule.
6. The $2.0bn NVIDIA preferred proceeds sit in the cash used for net cash while the share count excludes the as-converted shares — the proceeds are effectively counted twice.
7. margin_period = FY2026, so margins are one quarter stale.
8. entity_public_float $64,119,895,583 is a stale cover-page dollar amount.
9. Net income is unusable as a scale check for FY2026 without adjusting for the $1.8bn divestiture gain — any screen ranking on net income or P/E would materially misprice this name.