Cloud CapEx Translation

Does the CapEx translate?

$358B

combined CapEx from Microsoft, Amazon, Alphabet, and Meta in FY2025.

It rose 45–87% year over year. The cloud revenue it funds grew 20–36%. That gap is the question.

AI CapEx becomes visible before revenue translation becomes fully measurable. In FY2024 the gap widened sharply — and FY2025 data shows it persisting across all three cloud companies.

Interactive explorer

CapEx is outrunning the revenue it funds

Pick a company. The spend (CapEx) is in blue, what it should translate into (cloud revenue) in green, and the lagged cost it creates (D&A) in amber — all in USD millions, from Form 10-K filings.

CapEx YoY (FY2025)

+45%

$44,477M → $64,551M

Intelligent Cloud YoY

+21%

$87,464M → $106,265M

D&A vs the spend

53%

D&A = 32% of Intelligent Cloud revenue

What the filing does — and doesn’t — show

Intelligent Cloud is broader than Azure (it also includes server products and enterprise services). Azure revenue in dollars is not disclosed — Microsoft reports only an Azure growth %. Microsoft restructured its segment reporting in FY2025, restating prior-year Intelligent Cloud figures (FY2024 restated from $105,362M to $87,464M; FY2023 restated from $87,907M to $72,944M). The figures above use the restated basis for year-over-year comparability.

Azure growth %: 2021 50% · 2022 45% · 2023 29% · 2024 30% · 2025 34% — reported, but never in dollars.

The signature metric

The translation gap

For each percent of cloud-revenue growth, how many percent did capex grow? Above 1.0× means capex is outrunning the revenue it funds. Through 2023 the ratio sat near or below 1.0× — cloud revenue kept pace. In FY2024 it broke to 2–3× across the board, and in FY2025 the gap persisted — no reversion yet.

CapEx growth ÷ cloud-rev growth’22’23’24’25
MSFTMicrosoft0.6×2.9×2.1×
AMZNAmazon0.1×capex fell3.1×3.0×
GOOGLAlphabet0.8×0.1×2.1×2.1×
METAMeta Platformsno cloud revenueno cloud revenueno cloud revenueno cloud revenue

Meta has no cloud-revenue denominator, so the gap is undefined — capex with no external revenue to translate into is the structural finding, not a missing cell. The ratio uses total-company capex against segment cloud revenue, so it is a directional growth-gap signal, not a clean cloud-capex-to-cloud-revenue measure (capex also funds non-cloud activity). Microsoft fiscal year ends June 30.

Issuer-level read

Four translation types

The same question lands differently at each company — because what the filing lets you measure is different. That is the analysis, not a data gap.

AMZNDirect

Amazon

Direct cloud translation

Revenue line
AWS
Measurable
AWS revenue growth
Not measurable
AWS-specific CapEx

Cleanest translation test, imperfect attribution.

GOOGLDirect

Alphabet

Direct cloud translation

Revenue line
Google Cloud
Measurable
Cloud segment revenue
Not measurable
Cloud-specific CapEx — spend also supports broader technical infrastructure

Visible revenue translation, broad asset base.

MSFTProxy

Microsoft

Proxy translation

Revenue line
Intelligent Cloud
Measurable
Segment revenue and Azure growth %
Not measurable
Azure revenue in dollars

The AI narrative is large, but the clean revenue denominator is not disclosed.

METAInternal

Meta Platforms

Internal translation

Revenue line
None
Measurable
CapEx and D&A
Not measurable
External cloud revenue

No denominator is the point — monetization must be indirect, through ads, engagement, and AI products.

Same spend dynamic, four different economic forms

FY2025 year-over-year growth: CapEx (blue) jumped 45–87% across all four. The revenue it is meant to fund (green) grew 20–36% — and for Meta there is no cloud-revenue line at all.

Meta has no cloud-revenue bar — its CapEx funds internal AI and recommendation infrastructure, monetized indirectly through advertising. Microsoft fiscal year ends June 30; others end December 31.

Interim signal

Unaudited

Q1 2026: the gap has not closed

Single-quarter figures from the most recent 10-Q filings, compared against the same quarter one year earlier. Unaudited. Not annualized. Not directly comparable to full-year ratios.

Company / PeriodCapEx (qtr)YoYCloud revenueYoYGap estimate
MSFTQ3 FY2026 (Jan–Mar 2026)$30.9B+84%$34.7B+30%2.8×
AMZNQ1 2026 (Jan–Mar 2026)$44.2B+77%$37.6B+28%2.7×
GOOGLQ1 2026 (Jan–Mar 2026)$35.7B+107%$20.0B+63%1.7×
METAQ1 2026 (Jan–Mar 2026)$19.0B+47%no segmentn/a

Meta FY2026 CapEx guidance

$125B–$145B — management guidance from Meta’s Q1 2026 10-Q. Not audited actuals.

10-Q filings: MSFT Q3 FY2026 (10-Q filed May 2026); Amazon Q1 2026 10-Q; Alphabet Q1 2026 10-Q; Meta Platforms Q1 2026 10-Q. Gap estimate = quarterly CapEx growth ÷ quarterly cloud-revenue growth (same-period YoY). Directional only.

So what

Why the gap matters

Margins follow with a lag

Capex hits the income statement later, as depreciation. D&A is already compounding faster than cloud revenue at several of these companies — a drag that shows up after the spending headlines fade.

The bubble question, made concrete

“Is AI capex a bubble?” becomes measurable here: not whether the spend is large, but whether the revenue it funds is keeping pace — and where it structurally cannot (Meta has no cloud line at all).

Someone receives this money

Every dollar of this capex is revenue for the AI infrastructure supply chain — chips, foundry, memory, networking, power. The AI Infrastructure project studies who captures it.

Evidence & sources

Every figure, traceable

All values in USD millions, as filed. Each company-year is taken from that fiscal year’s Form 10-K.

CapEx
Cash flow statement — "Purchases of property and equipment" (gross)
D&A
Cash flow statement — "Depreciation and amortization"
Cloud revenue
Segment note — reportable cloud-segment revenue (AWS / Google Cloud / Intelligent Cloud)
CompanyFiscal yearCapExD&ACloud revenue
MSFTMicrosoftFY2021$20,622M$11,686M$60,080M
FY2022$23,886M$14,460M$75,251M
FY2023$28,107M$13,861M$72,944M
FY2024$44,477M$22,287M$87,464M
FY2025$64,551M$34,153M$106,265M
AMZNAmazonFY2021$61,053M$34,433M$62,202M
FY2022$63,645M$41,921M$80,096M
FY2023$52,729M$48,663M$90,757M
FY2024$82,999M$52,795M$107,556M
FY2025$131,819M$65,756M$128,725M
GOOGLAlphabetFY2021$24,640M$10,273M$19,206M
FY2022$31,485M$13,475M$26,280M
FY2023$32,251M$11,946M$33,088M
FY2024$52,535M$15,311M$43,229M
FY2025$91,447M$21,136M$58,705M
METAMeta PlatformsFY2021$18,690M$7,967Mn/a
FY2022$31,186M$8,686Mn/a
FY2023$27,045M$11,178Mn/a
FY2024$37,256M$15,498Mn/a
FY2025$69,691M$18,616Mn/a

Company Form 10-K filings. CapEx = 'Purchases of property and equipment' from the cash flow statement (gross, before proceeds/incentives). D&A = 'Depreciation and amortization' from the cash flow statement. Cloud revenue = reportable-segment revenue from the segment note. Cross-checked against stockanalysis.com filing tables. Pulled 2026-06-17.

Method & limitations

How to read this

The disclosure asymmetry is the finding

AWS (Amazon) and Google Cloud (Alphabet) report cloud revenue in dollars. Microsoft reports 'Intelligent Cloud' in dollars (broader than Azure) and Azure only as a growth %. Meta reports no external cloud segment at all. The translation is hardest to see exactly where the spend is largest.

Meta is the contrast case, not an outlier to hide

Meta’s capex serves internal AI training and recommendation infrastructure, monetized indirectly through advertising — there is no external cloud-revenue line for the spend to translate into. That structural difference is the sharpest version of the question, so it is shown explicitly rather than dropped.

Limitations

Capex here is the gross cash-flow line and is not AI-/cloud-specific (it also funds fulfillment, offices, devices, etc.). D&A is a policy-sensitive number (e.g., Alphabet extended server useful lives effective Jan 2023, lowering 2023 depreciation). Microsoft figures are fiscal-year (June). Microsoft restructured its segment reporting in FY2025, restating historical Intelligent Cloud figures (FY2024: $105,362M → $87,464M; FY2023: $87,907M → $72,944M) — all years above use the restated basis. Not investment advice.

Microsoft's fiscal year ends June 30; Amazon, Alphabet, and Meta end December 31 — years are not perfectly comparable across companies.

Update cadence

Annual data updated from each company's Form 10-K upon filing (typically February–August for fiscal years ending December 31 or June 30). Interim signals updated from 10-Q filings on a best-effort basis when material new data is available.

Scope

This is descriptive business analysis based on public filings. It does not provide investment advice, valuation targets, or trading recommendations.