Summary

  • Oracle has raised roughly $48 billion of senior unsecured debt across two flagship transactions — an $18.0 billion six-tranche sale completed September 26, 2025 and a $25.0 billion eight-tranche sale plus $500 million floater priced February 2, 2026 — with fiscal 2026 senior note issuance totaling $43 billion.
  • The company announced on February 1, 2026 a plan to raise $45–50 billion of gross proceeds in calendar 2026, roughly half equity-linked, including a $5 billion mandatory convertible preferred and a $20 billion at-the-market program completed during the first quarter of fiscal 2027.
  • Trailing-four-quarter capital expenditures rose from $21.2 billion at Q4 FY2025 to $55.7 billion at Q4 FY2026, while trailing free cash flow reached negative $23.7 billion; fiscal 2027 guidance calls for $90–95 billion of reported capex.
  • Off-balance-sheet lease commitments, mostly datacenter leases with 15–19 year terms, stood at roughly $260 billion at fiscal 2026 year-end — a third ledger column that is disclosed but only partially auditable.
  • The financing ledger is the most externally verifiable dimension of Oracle's AI bet, yet it still cannot answer the central question: whether the contracted backlog behind the borrowing converts to cash on the assumed schedule.

The Two Ledgers

Oracle's public AI narrative rests on a number that grows every quarter and can never be opened up: remaining performance obligations, the contracted backlog that management reports as future revenue. Across five quarters it expanded from $455 billion at August 31, 2025 (Q1 FY2026 results; Oracle IR), to $523 billion in November, $552.6 billion in February 2026, $638 billion in May and $664 billion at August 31, 2026, the last step driven by more than $30 billion of new AI contracts signed in a single quarter (Q1 FY2027 results). No disclosure identifies who the counterparties are, how long the contracts run, or under what terms they can be cancelled or renegotiated.

The company's own behavior acknowledges the difference between what is contracted and what is spent. Every dollar of the backlog eventually has to become a datacenter hall, a power feed and a rack of GPUs — and the spending shows up not in the backlog line but in Oracle's financing documents. Those documents are the subject of this report: bonds priced by investors, equity sold at market, ratings opinions issued by three agencies, and lease commitments disclosed under accounting rules. Where the backlog is a claim about the future, the financing ledger is a record of what Oracle has already had to do, and pay for, today.

Bond Issues With Dates and Coupons

The first large transaction is fully documented in an SEC filing. On September 26, 2025, Oracle consummated the issuance of $18.0 billion of senior notes across six tranches: $3.0 billion of 4.450% notes due 2030, $3.0 billion of 4.800% notes due 2032, $4.0 billion of 5.200% notes due 2035, $2.5 billion of 5.875% notes due 2045, $3.5 billion of 5.950% notes due 2055 and $2.0 billion of 6.100% notes due 2065 (8-K, September 26, 2025). The stated use of proceeds was general corporate purposes, which may include capital expenditures, repayment of indebtedness, future investments or acquisitions and dividends — language that does not tie any tranche to a specific datacenter but leaves the capital free to fund the buildout.

The second transaction came four months later and was, by size, a different class of event. On February 2, 2026 Oracle priced a $25.0 billion bond across eight tranches plus a $500 million floating-rate note (IFR). Combined with the September 2025 offering, these transactions brought fiscal 2026 senior note issuance to $43 billion (8-K) — more debt, in one fiscal year, than Oracle had appeared to need for most of its previous decade of software-led growth.

Two features of the February deal matter more than its size. First, the timing: it was priced one day after Oracle published its financing plan, as detailed below, and the same day two rating agencies acted on the company. Second, the structure: eight tranches stretching to long maturities signal that Oracle is funding assets with multi-decade physical lives — datacenters with 15–19 year lease terms — with multi-decade liabilities. That is a rational duration match, and it is also a commitment: the coupons come due whether or not the AI tenants renew.

The February 2026 Financing Plan

On February 1, 2026, Oracle announced a plan to raise $45–50 billion of gross cash proceeds during calendar 2026 to fund Oracle Cloud Infrastructure expansion (Oracle IR). Approximately half was planned to come from equity-linked and common equity issuance, anchored by a $5 billion mandatory convertible preferred (the 6.50% Series D) and a newly authorized at-the-market program of up to $20 billion. On the debt side, the company said it intended "a single, one-time issuance of investment-grade senior unsecured bonds early in 2026" and that it did "not expect to issue additional bonds during calendar year 2026 beyond this transaction."

The plan's composition is as informative as its size. Roughly half equity-linked means Oracle chose to dilute existing shareholders rather than load the entire buildout onto its balance sheet — a decision consistent with management's awareness that pure-debt financing at this scale would test its investment-grade ratings. The equity program was executed: the $20 billion ATM was completed during the first quarter of fiscal 2027, alongside the $5 billion mandatory convertible preferred.

The plan's debt clause, by contrast, became a testable promise — and, by mid-2026, an ambiguous one. When Oracle reported fiscal 2026 results in June, management indicated fiscal 2027 would involve raising roughly $40 billion via debt and equity (Datacenter Dynamics), reiterating that no additional debt issuance was planned for calendar 2026. Whether future financing continues to honor the "one-time" framing is one of the clearer signals investors can watch: a second investment-grade bond sale within the same calendar year would falsify the stated plan, or require management to explain what changed.

Capital Expenditure: The Gap Made Visible

The financing exists to feed a spending machine that has grown at a pace Oracle itself did not originally plan for. Trailing-four-quarter capital expenditures were $21.2 billion at the end of fiscal 2025 (stockanalysis balance sheet data). As recently as September 2025, on the Q1 FY2026 earnings call, management guided fiscal 2026 capex to approximately $35 billion, saying "given our RPO growth, I now expect fiscal year '26 CapEx will be around $35 billion." That figure was raised repeatedly during the year. By the time fiscal 2026 closed, trailing-four-quarter capex had reached $55.7 billion — roughly 60 percent above the September guidance (Q4/FY2026 results).

The pattern continued into the new fiscal year. First-quarter fiscal 2027 capex was approximately $28 billion in a single quarter, and management guided to $90–95 billion of reported capital expenditures for fiscal 2027, with roughly $70 billion of net cash outlay after $20–25 billion of customer prepayments and timing benefits (Q1 FY2027 results; Q1 FY2027 earnings call transcript). The distinction between reported and net figures is itself a disclosure choice worth monitoring: the gap depends on customer cash arriving before the corresponding capacity is delivered, a timing dependency that quietly couples Oracle's reported capex burden to its customers' payment behavior.

Free Cash Flow: The Price of the Bet

The consequence of spending faster than revenue arrives is visible in one line. Trailing-four-quarter free cash flow turned negative in Q4 fiscal 2025 and reached negative $23.7 billion by Q4 fiscal 2026 (Q4/FY2026 results). In the first quarter of fiscal 2027, free cash flow was negative $5 billion even as operating cash flow set a record at $23 billion — record cash generation, consumed several times over by construction.

This is the arithmetic that connects the two ledgers. A $664 billion backlog is only self-financing if it converts to revenue on a schedule that stays ahead of the capex curve. The gap between the two — funded in the interim by bonds, preferred stock and common equity — is precisely the risk the rating agencies were asked to price. The market's own reaction is on record: Oracle shares fell roughly 11 percent on June 11, 2026, on concerns about the increased capital raise and cash position (CNBC).

The Lease Ledger: Disclosed but Semi-Auditable

Between the fully priced financing transactions and the fully asserted backlog sits a third column: off-balance-sheet lease commitments. Oracle's February 28, 2026 10-Q disclosed $261 billion of lease commitments not yet commenced, the bulk datacenter leases with terms of 15–19 years (Q3 FY2026 10-Q); by fiscal 2026 year-end the figure stood at approximately $260 billion. A separate line of unconditional purchase and supplier obligations was $11 billion in February 2026, mainly datacenter power, and $13.3 billion at year-end, with a subsequent commitment of roughly $19 billion over five years.

These numbers are disclosed, dated and sourced to regulatory filings, which makes them far more auditable than the backlog. But they are not fully priceable: a lease commitment is a future obligation whose ultimate cost depends on commencement timing, escalators, options and termination provisions that Oracle does not disclose in comparable detail. Roughly $260 billion of future lease payments is a real number with real landlords; whether it represents $260 billion of risk or $260 billion of customer-backed capacity depends on the same undisclosed counterparty information that governs the backlog. Meanwhile, a separate layer of project-level debt — $38 billion of senior secured credit facilities for Vantage Data Centers campuses in Texas ($23.25 billion) and Wisconsin ($14.75 billion) tied to the Stargate program — sits off Oracle's corporate balance sheet entirely, on the developer's (Bloomberg).

What the Ratings Say, in Their Own Words

Three agencies have now opined on this structure, and their actions are dated and quotable. On February 2, 2026, Fitch rated the new senior notes BBB and affirmed Oracle's IDR with a stable outlook (Fitch). The same day, Moody's affirmed Oracle at Baa2 but revised the outlook to negative — a split decision, one notch and one direction apart (Moody's). In July 2026, S&P moved further, downgrading Oracle to BBB- (S&P Global).

The divergence is analytically useful. Fitch saw a stable credit with contracted revenue growth; Moody's saw the same growth but flagged the trajectory of leverage and free cash flow; S&P concluded the trajectory was already a downgrade. All three were pricing the same published numbers — $43 billion of fiscal-year debt issuance, negative $23.7 billion of trailing free cash flow, roughly $260 billion of lease commitments — and reached different weights on the offsetting backlog. That disagreement is itself evidence: the financing ledger is solid enough to be rated, and uncertain enough to be rated three different ways.

What the Ledger Verifies — and What It Does Not

The financing ledger establishes, with dated third-party evidence, that Oracle has committed roughly $48 billion of new senior unsecured debt and roughly $25 billion of equity and equity-linked capital to the AI buildout within about a year; that it is spending $28 billion per quarter on construction against negative free cash flow; and that it has contracted future lease payments on a scale larger than its annual revenue by an order of magnitude. None of that is assertion. Every figure traces to a filing, a priced transaction or a regulator-visible disclosure.

What the ledger cannot verify is the other side of every one of those commitments: the assumption that the $664 billion backlog converts to cash on schedule, with counterparties who pay and renew. Management has said the customer base extends beyond OpenAI — Meta has been named, and Oracle cited AMD, Meta, NVIDIA, OpenAI, TikTok and xAI as anchor OCI customers in its financing announcement — but the composition, duration and cancellation terms of the backlog remain undisclosed (Oracle IR). Reuters has framed the situation bluntly as a high-stakes ratings gamble built on AI strategy (Reuters); the rating agencies' split decisions quantify how far reasonable analysts diverge on the same facts.

The financing ledger, in other words, measures the bet, not the outcome. It tells you Oracle has already paid — in coupons, in dilution, in commitments — for a future that exists so far only as a contracted number no outsider can audit. The next falsifiable signals are specific and dated: whether a second calendar-2026 bond issuance appears despite the February plan's "one-time" language; whether lease commitments keep growing faster than revenue recognized from them; and whether the ratings converge or diverge further as the fiscal 2027 capex program of $90–95 billion is funded.

On those observable events, the financing ledger will keep its unusual property: unlike the backlog, it can be counted.