Phase Space AI

02 Financial Model Notes

Meta Platforms [META]

Meta Platforms, Inc. [META] — Financial Model Notes

Task 2 · META_Underwriting_Model_2026-07-27.xlsx · 6 tabs, live formulas


1. Structure

Tab Purpose
Assumptions All drivers; blue = input, black = formula. Segment growth, EBITDA margin, capex, asset-class shares, useful lives, CIP lags, tax, SBC, WACC, terminal growth, maintenance capex. Base-rate check block at the foot.
Depreciation Engine The analytical core. Vintage straight-line schedule with a full parallel calculation at a 4.0-year server life, so the useful-life assumption can be flexed live.
Income Statement Family of Apps + Reality Labs → consolidated → EBITDA → less modelled D&A → EBIT → pre-tax → EPS, plus EPS excluding non-operating income and EPS at a 4-year server life.
Cash Flow & BS OCF, capex, economic FCF (SBC at 10.2% of revenue treated as a real cost), cash roll including debt issuance, net debt, leases, and an off-balance-sheet memo line carrying the VIE exposure.
Scenarios Bull / Base / Bear, terminal revenue linked live to the Income Statement, exit EV/NOPAT, discounting, probability weighting, E[R], cash-hurdle comparison, base-rate check.
DCF & Valuation Unlevered FCF, five-year capex fade to maintenance, Gordon terminal value, WACC × g sensitivity grid, comps, football field.

2. Verification

All workbooks closed first, then opened in Excel via AppleScript and computed cells read back. openpyxl writes formula strings without evaluating them, so a readback is the only proof.

Readback (2026-07-27):

Revenue  : $200,966  $250,987  $296,215  $340,952  $382,361  $421,311
EBIT     : $83,278   $103,103  $108,907  $111,159  $111,345  $109,460
D&A      : $18,000   $24,900   $39,201   $55,907   $74,100   $92,769
D&A@4yr  : $23,625   $32,775   $51,739   $73,644   $97,439   $121,874
EPS      : $27.43    $33.56    $35.46    $36.23    $36.50    $36.06
EPS@4yr  : $25.63    $31.02    $31.38    $30.40    $28.74    $26.30
EconFCF  : 18,509    (25,023)  (29,696)  (25,841)  (14,778)  (1,924)
DCF/sh = $78.01   TV% = 177%   Scenario PW = $424.59   E[R] = -28.5%

FY2025A ties to the 10-K: revenue $200,966m ✓ (exact), PP&E depreciation $18,000m ✓ (exact), operating income $83,278m vs $83,276m reported (0.002% residual).

Errors found and fixed during verification (each would have failed silently): the D&A fade formula referenced capex as a signed negative figure and drove terminal D&A negative; the cash roll-forward omitted debt issuance; and the original five-year DCF terminated while capex was still ~1.8x D&A, which is not a steady state — a five-year fade period was added.

3. The Depreciation Engine — calibration

Parameter Value Basis
Servers / network share of capex 58% FY2025 gross PP&E movement by class (R56)
Buildings / shell share 37% same
Other equipment share 5% residual
Server useful life 5.5 years 10-K R45: "5 years 6 months", extended Jan-2025 from 4–5 years
Buildings life 27.5 years 10-K R45: 25–30 years
Other equipment life 7 years 10-K R45: 1–25 years; 7 used as an effective average
CIP lag, servers 1.0 years Fitted
CIP lag, buildings 2.0 years Fitted
Calibration RMSE, FY2022–FY2025 $0.38bn on an $8.7–18.0bn series Grid search — a very tight fit

The CIP lag is the mechanism, not a convenience. Construction in progress was $50,521m at Dec-2025 and $61,017m at Mar-2026 — 22% and ~24% of gross PP&E respectively — depreciating nothing at all. Meta's capex enters the balance sheet roughly two years before it enters the income statement.

Cross-check against the separately-disclosed servers line (R57, recovered from the R-files and not present in XBRL company-facts): reported servers-and-network depreciation was $7,320m (FY2023), $11,340m (FY2024), $13,360m (FY2025). The engine reproduces the total ($18,000m FY2025 ✓) and the servers/other split is consistent with the disclosed sub-line.

4. Base-rate checks

Family of Apps revenue growth. FY2025 +22.4%; Q1-2026 +33% (helped by a weak dollar — constant currency was +29%); consensus Q2-2026 +26.6%. The reference class says digital-advertising growth at $200bn scale decelerates in every cycle. The model assumes 25% → 10% over five years — deceleration, i.e. WITH the base rate. No override is claimed.

Operating margin. FY2025 was 41.4%. ~$2.9bn of FY2025 operating income came from the January-2025 extension of server useful lives — 20.9% of the entire $13.9bn year-on-year increase. The model does not assume the margin holds: it falls to 26.0% by FY2030, driven entirely by the depreciation engine, not by an assumed demand deterioration. This is stated explicitly so the base case is not mistaken for a bear case imposed on the business.

Capex. FY2025 $69.7bn (+87% YoY). FY2026 guided $125–145bn (+94% at the midpoint; the guide was raised from $115–135bn on the Q1-2026 call). The model uses $135bn for FY2026 then decelerates to $183bn by FY2030. Critically: reported capex EXCLUDES the ~$27bn Hyperion campus, which sits in an unconsolidated VIE — so the model's capex line, like the company's, understates the compute being procured.

Depreciation. FY2025 total PP&E depreciation was $18.0bn against $69.7bn of capex — a ratio of 0.26x. The model shows this converging toward 1.0x by roughly FY2031 regardless of any assumption about demand. This is the least discretionary line in the model.

5. Duration-matched, two-tranche treatment

Per references/regime-change-test.md. Meta discloses no customer RPO — it states explicitly that it does not disclose unsatisfied performance obligations for contracts of a year or less, and advertising is short-cycle. There is no contracted revenue tranche. Duration matching is therefore applied to the cost and obligation side:

Tranche Contracted duration Model treatment
Total contractual commitments ($131.0bn) Ladder: $30.6bn 2026, $22.2bn 2027, $21.2bn 2028, $19.8bn 2029, $19.4bn 2030, $17.9bn thereafter Modelled at the disclosed ladder
Hyperion lease commitment ($12.31bn) Commences 2029, four-year initial terms Outside the FY2026–30 income statement almost entirely — flagged as a post-horizon obligation
Hyperion RVG ($28bn threshold) Decreasing over time Not modelled — not probable per the company. Carried as an off-balance-sheet memo line only.
Long-term debt ($58.7bn) Ladder disclosed Interest modelled
Advertising revenue None disclosed Fully merchant; mean-reverting growth throughout

Contracted share of revenue by forecast year: 0% in every year.

The two-tranche finding that matters, and it cuts bullish: total contractual commitments of $131.0bn are LESS than a single year of guided 2026 capex ($125–145bn). The overwhelming majority of the FY2027–2030 programme is not contracted. Modelling Meta as if it were locked into $180bn a year is modelling an obligation that does not exist. The model therefore treats FY2028–30 capex as management's choice, and the Bear scenario explicitly includes a capex cut path.

6. Known limitations

  1. There is no separable AI revenue line and there cannot be one. Meta's AI capex improves ad ranking; the revenue effect is diffuse and unattributable. The TAM work in the research document reframes this as a required return rather than a market size, which is the only falsifiable formulation available.
  2. FY2027 consensus EPS is inferred, from a secondary source's "~18x 2027 estimates" against the spot price, not pulled from a primary consensus feed (Alpha Vantage exhausted). Directionally reliable, not precise, and labelled as such wherever it drives a conclusion.
  3. The tax rate is normalised at 19%. FY2025 reported 30% including the $15.93bn OBBBA/CAMT charge; Q1-2026 reported −23% including the $8.03bn reversal. Neither is a run-rate. The normalisation is an assumption.
  4. The DCF has terminal value at 177% of enterprise value and is labelled as having low information content. The reverse DCF is the informative version.
  5. The RVG is not modelled. It is a $28bn contingent liability that the company states is not probable. Modelling it would be inventing a liability; ignoring it entirely would be negligent. It is carried as a memo line and discussed in the research document §4.
  6. Comps use trailing reported figures, not NTM consensus. Alphabet's FY2025 was not retrievable under the standard XBRL tag; FY2024 is shown and flagged.