No positions in the markets we measure BY 07:30 ET
CCIR Compute Credit
Index Research
Research · 2026-07-30 · updated 2026-07-31 (Amazon Q2 10-Q) · filings + official transcripts

Five Builders, One Covered Quadrant: Microsoft's FY2026

Five companies are building AI infrastructure at a combined pace above half a trillion dollars a year. Four of them spend more on the build than their cloud businesses bring in. Microsoft is the lone company whose cloud revenue covers its capital spend: $115.9 billion of capex against $214.4 billion of Microsoft Cloud revenue in the FY2026 10-K filed July 29. It is also the lone one whose contracted backlog covers its total obligations: $678 billion against $658 billion of forward commitments, debt, and lease liabilities. This note reads Microsoft's filing against the group's: what each company has sold, what each spends, and which builds are paid for by the clouds they serve. Figures are from filings and releases unless attributed. Quantities are maintained on the AI commitments monitor.

01 Where the five builds stand

Set five of the largest AI builders on two axes: capital spend against the cloud revenue that pays for it, and contracted backlog against contracted obligations. Microsoft is alone in the covered quadrant:

  • Microsoft: $140.6B of trailing-year capital spend, including finance-lease additions, inside $214.4B of Microsoft Cloud revenue. A $678B backlog against $658B of total obligations (forward commitments, debt at face, and balance-sheet leases): 1.03×.
  • Oracle: the highest backlog coverage, 1.45× ($638B against $441B of total obligations), on cloud revenue of $34.0B. Capex of $55.7B runs 1.6× that revenue.
  • Amazon: 0.94× ($496B against $528B of total obligations) per the Q2 10-Q it filed July 31. The backlog grew 36% in the quarter (computed) and its weighted-average life stretched to 6.4 years from 5.5. Debt face reached $133.0B, with a further $25.0B of notes issued in July. AWS revenue of $148.4B against $173.0B of capex.
  • Alphabet: 0.51× ($519.5B against $1,018B), the largest obligations stack of the five, on Google Cloud revenue of $77.6B (trailing year, computed). Capex of $132.4B runs 1.7× its cloud line. The stack is partly construction of the disclosure: Alphabet's purchase figure is the broadest of the five, sweeping in open purchase orders plus supply, energy (take-or-pay to 2054), and content commitments.
  • Meta: no compute revenue and no backlog disclosed, against $741B of total obligations and $92.4B of capex.
Bubble chart of five companies. Horizontal axis: cloud revenue divided by capital spend. Vertical axis: contracted backlog divided by all-in obligations including debt and balance-sheet leases. Microsoft sits alone in the shaded upper-right covered quadrant at 1.5 and 1.03. Amazon at 0.86 and 0.94 and Oracle at 0.61 and 1.45 are hollow, marking negative trailing free cash flow. Alphabet at 0.59 and 0.51. Meta sits at the origin with no cloud revenue and no backlog. Bubble area is proportional to capital spend.
Chart notes & definitions

Trailing-year windows ending June 2026 (Oracle: May 2026). Capex is company-wide, including disclosed finance-lease additions, against cloud-segment revenue. Definitions differ by company (Microsoft Cloud includes Office 365; Google Cloud includes Workspace and TPU sales; Oracle excludes license support), and each company's capex also serves businesses outside its cloud. Backlog = commercial RPO. Obligations (all-in) = leases signed but not yet commenced plus purchase obligations, plus debt at face value and balance-sheet lease liabilities (operating and finance): the components all five disclose. Construction commitments, disclosed only by Microsoft, are excluded. Meta's finance-lease liabilities are not broken out in its Q2 10-Q, slightly understating its stack. The ratio is a conservative screen, not a coverage finding: backlog is contracted revenue only, while debt and commenced leases are also serviced by revenue no backlog captures. Hollow = negative trailing free cash flow on a uniform basis (operating cash flow − capex − finance-lease additions). All ratios are our calculations from the filings.

02 Backlog vs cloud revenue

Customers have signed $678 billion of commercial contracts Microsoft has not yet delivered. That is up 84% in a year, with a weighted-average duration of about 2.3 years and roughly 30% expected to convert to revenue within twelve months. Azure grew 43% in the fourth quarter against the company's own 39–40% guide, and Microsoft Cloud revenue reached $214.4 billion for the fiscal year. For two years the backlog held near twice the annual cloud run-rate. It now stands at more than three times. The backlog is growing faster than the infrastructure that serves it.

Commercial RPO vs Microsoft Cloud revenue · $B at fiscal year end
$269B $137B 2.0× FY2024 $368B $169B 2.2× FY2025 $678B $214B 3.2× FY2026 commercial RPO Microsoft Cloud revenue
Chart notes & definitions

Commercial RPO from each fiscal year's 10-K ($269B FY24, $368B FY25, $678B FY26). Growth percentages are computed, not printed. Microsoft Cloud revenue per the company's releases. The ~30%-within-12-months figure applies to total-company RPO.

03 Backlog vs contracted obligations

The chart below is the forward slice of the obligations stack: the AI-era commitments signed ahead of delivery, before debt and commenced leases are added. On that slice alone the coverage runs higher: Oracle 2.3×, Amazon 1.7×, Microsoft 1.3×, Alphabet 0.6×. The all-in ratios above temper each by the balance sheet. Meta discloses no compute revenue backlog at all against $628 billion of forward commitments: four contracted books and one merchant position.

Paired horizontal bars per company of contracted revenue backlog in gold against contracted obligations in blue. Oracle 638 billion dollars of backlog against 273 billion of obligations, 2.3 times. Amazon 496 against 286, 1.7 times. Microsoft 678 against 523, 1.3 times. Alphabet 520 against 896, 0.6 times. Meta shows no backlog against 628 billion of obligations.
Chart notes

Backlog = commercial RPO. Obligations = leases signed but not yet commenced plus purchase obligations, latest filing each (Microsoft, Amazon, Alphabet, and Meta as of June 2026; Oracle May 2026). Roughly $68B of data-center leases Meta signed in July and $19B of purchase commitments Oracle entered after its fiscal year-end are not yet in these figures. Obligation constructs differ by issuer: Alphabet's purchase figure is its MD&A total including open purchase orders and supply, energy, and content commitments, the broadest of the five, while Oracle disclosed only unconditional obligations, primarily power. Ratios are our calculations from the filings.

The ratio is one dimension of a sold book. Tenor and counterparty are the others, and the four books differ more on those than on the ratio:

  • Microsoft: the shortest and broadest book, with ~2.3-year weighted duration and ~30% converting within twelve months. Roughly 45% was tied to OpenAI as of December per the company's January disclosure, with the fourth quarter's growth attributed to clients other than AI model developers.
  • Amazon: the book lengthened to a 6.4-year weighted life from 5.5 in March. A $138B OpenAI arrangement, disclosed by Amazon, equals about 28% of the June backlog. And the Q2 10-Q discloses an expansion of the Anthropic commitment announced in the quarter of more than $100B over ten years. Both arrangements include obligations tied to the performance of AWS chips.
  • Oracle: the longest and most concentrated. ~12% converts within twelve months, 34% in months 13–36, 34% in months 37–60, the remainder later, attributed only to "certain significant cloud contracts," unnamed. OpenAI, though, has published a partnership with Oracle exceeding $300 billion over five years: up to roughly 47% of the book from a single counterparty.
  • Alphabet: a faster book, ~50% within 24 months. No concentration figure disclosed.

The shares are our calculations from the cited disclosures.

04 Free cash flow, uniform basis

On a uniform cash basis (operating cash flow minus capital expenditure minus finance-lease additions), Microsoft generated $42 billion in FY2026, against $67 billion of reported free cash flow. The difference is $24.6 billion of finance-lease additions, non-cash at inception and repaid through financing. The instrument moves where the build appears in the statements, not whether it is covered: counted in full, the build still sits inside the cloud revenue line. And the address moves again next year. On the fourth-quarter call the CFO said more future datacenter leases will shift from finance to operating leases from FY2027, "finance leases are included in capital expenditures while operating leases are not," restating calendar-2026 capital-expenditure expectations from roughly $190 billion in April to approximately $175 billion on the reclassification alone. Same contracts, less of the build in the reported lines.

Five lines of trailing twelve month free cash flow on a uniform basis, operating cash flow minus capex minus finance lease additions, from early 2024 through June 2026. Alphabet ends at plus 51 billion dollars, Microsoft plus 42, Meta plus 39, while Amazon declines to minus 16 and Oracle to minus 29.
Chart notes

Filed cash flows; gross capex, not net of proceeds. Companies' own reported figures differ by basis. Amazon reports −$7.6B trailing free cash flow on its definition (capex net of proceeds, before lease additions). Microsoft reports +$67B before its $24.6B of finance-lease additions. Oracle pre-FY2026 and Meta pre-2025 finance-lease additions are undisclosed and treated as zero. Meta's 2026 and Oracle's FY2026 nine-month additions are apportioned, as labeled on the card. The hollow bubbles in the exhibit above mark the two companies negative on this basis.

05 The swap market's read

The credit market priced these differences in real time this week, as five-year credit-default-swap spreads across the complex swept to records:

  • Oracle: near 215bp per FT/LSEG, beyond its 2008 peak (S&P Global Market Intelligence quoted nearer 200 the same day).
  • Meta and Alphabet: records per Bloomberg and LSEG. Alphabet's was attributed to its negative second-quarter free cash flow (a single-quarter print, computed at −$5.9B from its filed cash flows; its trailing-twelve-month line in the chart above remains positive).
  • CoreWeave: near 855bp per Bloomberg.
  • Amazon: near 68bp per Bloomberg, from 36 at the start of the year. No post-earnings level was in circulation as of this writing.
  • Microsoft: near 53bp per Bloomberg, the tightest of the group.

All levels are as attributed. None are CCIR observations. One market framing, attributed: Seeking Alpha described Oracle's contract as "a liquid hedge on AI capex, OpenAI execution and broader data-center spending narratives". That is the backlog-quality question, stated as a price.

06 Read against the credit record

Microsoft's committed stack funds at lease economics: a 13-year weighted-average remaining term at a disclosed 4.5% discount rate. Standalone GPU-backed borrowers on the compute-credit record fund at spreads of several hundred basis points over it. And unlike Alphabet's backstops, the lease book carries no quarterly filed mark. CCIR carries these quantities with each Microsoft filing on the AI commitments monitor, and the rates the committed deals themselves imply on the implied committed rates table.

Sources: Microsoft Form 10-K for the fiscal year ended June 30, 2026 (acc. 0001193125-26-323660) and FY26 Q4 earnings release; prior-period figures from the FY2025 10-K (acc. 0000950170-25-100235) and FY2024 10-K (acc. 0000950170-24-087843); useful-life and capital-expenditure quotations from Microsoft's official FY26 Q4 (July 29, 2026) and FY26 Q3 (April 29, 2026) earnings-call transcripts. CDS levels as attributed in-text (Bloomberg; FT/LSEG; Reuters and S&P Global Market Intelligence; Seeking Alpha — July 27–29, 2026 quotes). Group-exhibit peer figures: Meta Q2 2026 10-Q (acc. 0001628280-26-050705); Alphabet Q2 2026 10-Q (acc. 0001652044-26-000071); Oracle FY2026 10-K (acc. 0001193125-26-277521); Amazon Q2 2026 10-Q (acc. 0001018724-26-000026, filed July 31 — Amazon figures updated to it that day) and Q2 2026 earnings release. Ratios and the uniform cash basis are our calculations from the disclosed figures. Nothing here is a solvency opinion, a forecast, or investment advice. CCIR publishes reference data and records.