Operating vs Finance Lease API: ASC 842 Data

Oracle's balance sheet carries $37.9B of recognized lease liability. Its lease footnote discloses another $260.0B the company has already signed but has not started yet, nearly double Oracle's entire $130.1B of funded debt. That figure comes from one field in Oracle's FY2026 10-K, tagged in XBRL as leases not yet commenced, and none of it sits on the balance sheet.

/api/footnotes/leases turns the full ASC 842 lease footnote into structured JSON for any filer that tags it: right-of-use assets, the lease liability split by operating and finance classification, the payment maturity schedule, and this not-yet-commenced disclosure, cited to the exact SEC accession number. This post covers what the disclosure actually requires, how to pull it via API, the AI-datacenter-driven growth in commitments still off five hyperscalers' balance sheets, and a worked classification comparison between Walmart and Delta Air Lines.

What ASC 842 requires: operating vs finance lease classification

ASC 842, in effect for public companies since 2019, requires lessees to put substantially every lease on the balance sheet: a right-of-use asset and a matching lease liability, the present value of the lease's future payments. Before ASC 842, only leases meeting stricter capital-lease tests appeared on the balance sheet at all, operating leases were a footnote-only commitment. That changed in 2019, but the operating versus finance distinction did not disappear. It moved from an on-balance-sheet-or-not question to a classification-and-presentation question.

A lease is classified as finance (the term ASC 842 uses for what used to be called a capital lease) if it meets any one of five tests: the lease transfers ownership of the asset by the end of the term, it grants a bargain purchase option, its term covers the major part of the asset's remaining economic life, the present value of payments equals substantially all of the asset's fair value, or the asset is so specialized it has no alternative use to the lessor once the term ends. A lease meeting none of the five is an operating lease. Classification still matters after ASC 842 because it changes the income statement, not just the balance sheet: operating lease cost is one straight-line expense, while finance lease cost splits into asset amortization plus separately disclosed interest, the same shape as debt.

Getting a company's lease data via API

Call /api/footnotes/leases with a ticker, CIK, CUSIP, or FIGI. period is annual or quarter. The response returns operating and finance leases side by side, each with its own right-of-use asset, current and noncurrent liability, and payment ladder. Oracle's real FY2026 response, trimmed:

// GET /api/footnotes/leases?symbol=ORCL&period=annual&limit=1
{
  "period": "2026-05-31",
  "operating": {
    "rightOfUseAsset": 29690000000,
    "liabilityCurrent": 3542000000,
    "liabilityNoncurrent": 26648000000,
    "liabilityTotal": 30190000000,
    "discountRate": 0.057
  },
  "finance": {
    "liabilityCurrent": 620000000,
    "liabilityNoncurrent": 7081000000,
    "liabilityTotal": 7701000000,
    "discountRate": 0.04
  },
  "leaseNotYetCommenced": {
    "total": 260000000000,
    "operating": 260000000000
  },
  "dateFiled": "2026-06-22"
}

/api/footnotes/leases requires the Professional plan. Each payment ladder's total minus discountExcess reconciles to the discounted liability client-side, every figure in this post's underlying data file was checked against that identity before use. A minority of filers mis-tag the discount rate as a whole percent instead of a fraction; /api/footnotes/leases re-scales those and flags the entry, so a raw discountRate in the response is always a fraction of 1.

Leases not yet commenced: the AI datacenter growth signal

A minority of filers also tag leaseNotYetCommenced: fixed payments for leases a company has already signed but that have not started, so no right-of-use asset or liability exists on the balance sheet yet. It is a pure footnote disclosure. Among filers who tag it, five names dominate by scale, each building out datacenter capacity for AI compute: Amazon, Microsoft, Oracle, Meta, and Alphabet.

Combined across these five companies' own two most recent fiscal years (fiscal year ends vary, so the pairing is each company against itself), leases not yet commenced grew from $239.1B to $847.7B, a 255% jump. Oracle led by rate: $43.4B to $260.0B, a 499% increase in a single fiscal year. That tracks Oracle's own disclosure to investors that lease commitments had reached $248 billion by November 2025, reported at the time as largely tied to its data-center and cloud-capacity buildout; the underlying XBRL disclosure carries no use-of-asset breakdown, so treat that characterization as directional, not a dollar-for-dollar reconciliation to the FY-end total. None of that $260.0B carries a right-of-use asset or a liability yet, by definition it has not started.

Microsoft's trajectory does not move in a straight line, which is exactly why the quarterly view is worth pulling separately from the annual one.

Microsoft's not-yet-commenced balance actually declined for three straight quarters, $116.1B in September 2024 down to $92.7B by June 2025, consistent with signed leases converting to recognized right-of-use assets as datacenters came online faster than new commitments were being signed. That reversed hard starting September 2025: $155.1B by December 2025, $196.6B by March 2026, and $329.1B by fiscal year end June 2026, a 255% increase over the fiscal year. New commitment volume overtook the conversion-to-recognized pace by a wide margin.

Worked example: Walmart vs Delta's lease structures

Classification is not just an accounting formality once you compare two real companies. Walmart and Delta Air Lines both carry billions in lease liability, split very differently between operating and finance.

Walmart's total lease liability of $22.3B splits 70% operating and 30.3% finance. Delta's $7.0B splits 88% operating and only 11.6% finance, despite aircraft being the more capital-intensive asset most readers would associate with a finance lease. The payment ladders add another layer: Delta's finance-lease payments are front-loaded (year 1 at $256 million, year 2 at $289 million) then fall to between $49 million and $109 million a year for years 3 through 5, a much shorter runway than Walmart's, whose finance-lease payments stay elevated for years (year 1 at $1,228 million, still $692 million in year 5, plus $6,003 million thereafter).

Neither number is visible from the balance sheet's single lease-liability line. Getting the operating-versus-finance split, and the maturity ladder behind each, requires the footnote-level detail /api/footnotes/leases exposes directly.

Lease liability vs funded debt

Lease liability is real leverage, even though it sits in its own footnote rather than next to the bonds. Oracle's recognized lease liability, operating plus finance, totals $37.9B against $130.1B of funded debt from /api/financials/balance-sheet, the same debt broken into 56 named tranches in Debt Tranche Structure API. The two data paths cross-check exactly: the balance sheet's own operating-lease-liability line matches /api/footnotes/leases' operating liability total to the dollar for the same filing (Oracle's $7.7B of finance-lease liability is disclosed separately and sits outside that check).

Layer in the not-yet-commenced figure and the scale changes again. Oracle's $260.0B of signed-but-unstarted leases is 6.9x its already-recognized lease liability and 2.0x its entire funded debt load, before a single dollar of it touches the balance sheet. A credit read that stops at totalDebt misses both the $37.9B already-recognized lease liability and the larger commitment still ahead of it. Independent analysts and researchers have separately flagged this same industry-wide pattern (aggregate off-balance-sheet AI infrastructure commitments across the largest hyperscalers), but as one-time snapshot estimates. /api/footnotes/leases exposes the same disclosure as queryable, per-company, per-quarter data, with an SEC accession number on every response, not a single point-in-time total.

/api/filings/search-by-accession-number resolves any accession number in this post straight back to the filing on SEC EDGAR, free tier, so none of the figures above have to be taken on faith. Every value was extracted through Arelle, the open-source XBRL processor, the same pipeline behind the sibling Remaining Performance Obligation API, REIT Property Schedule API, and Pension Funded Status API, another obligation that nets against plan assets and disappears into a generic balance-sheet rollup the same way. Coverage of the not-yet-commenced field specifically is thin by design, most filers with no major signed-but-unstarted commitments simply have nothing to tag, so always check that a response is non-empty before building on a given ticker, the same standard StockFit's broader fundamentals API for quant research holds every dataset to.

FAQ