AI infrastructure spending: five companies under cash-flow pressure

AI infrastructure spending: five companies under cash-flow pressure

Filings available on August 30, 2026 show two of these five infrastructure buyers spending more on capital assets than their operations generated in cash. Compare cash coverage, disclosed returns and reporting periods before drawing conclusions.

File number: AI-CAPEX-26

Data cutoff: 2026-08-30

Evidence basis: Official filings and releases

Buyers covered 5 companies

Microsoft、Alphabet、Amazon、Meta、Oracle

Cash-flow deficit 2 companies

Amazon and Oracle had negative free cash flow in the periods shown

Highest CapEx / OCF 174.1 %

Oracle FY2026

Supplier data-center growth 117 %

NVIDIA Q2 FY2027, year over year

Spending growth is outrunning internally generated cash

Farther right means faster capital-spending growth; higher means capital spending consumes more operating cash flow. Bubble area represents capital spending in the disclosed period. The upper-right is not an AI leadership zone; it is where cash flow must prove returns sooner.

Cash-flow pressure across five infrastructure buyers

Five companies plotted by capital-spending growth, capital spending as a share of operating cash flow, and capital-spending scale. Oracle and Amazon spent more on capital expenditure than they generated in operating cash flow.

Capital spending / operating cash flow

This ranks financial pressure, not company quality. Reporting periods differ; the ratio shows how much of each company's operating cash flow is absorbed by capital spending.

Cash-flow pressure ranking

Ranked by capital spending as a share of operating cash flow, highest first.

The first fault line: free cash flow

Alphabet's first-half 2026 capital spending reached about 95% of operating cash flow. Amazon and Oracle exceeded 100%, so current-period operating cash flow alone did not cover their capital spending.

The money has arrived; the returns remain backstage

Block area represents cash capital expenditure in each company's latest reported period. Half-year and full-year figures are not directly comparable; this view shows scale, not a precise ranking.

Cash capital-spending scale

Treemap of cash capital spending in each company's latest reported period. Microsoft is largest on a full-year basis; Alphabet, Amazon, and Meta report half-year figures.

Disclosure gaps: AI operating profit is rarely separated

0 = not separately disclosed; 1 = proxy metric or management run rate only; 2 = independently traceable disclosure. Darker cells make it easier to connect investment with returns. The largest gaps are in AI revenue, AI capital spending, and AI operating profit.

AI return disclosure visibility matrix

Visibility scores for five companies across AI revenue, AI capital spending, AI operating profit, and backlog. AI-specific profit disclosure is broadly absent.

Supplier comparison: NVIDIA data-center revenue

Buyer filings still tend to fold AI revenue, depreciation, and operating profit into cloud or advertising businesses. Supplier filings report data-center revenue directly. NVIDIA recorded $89.023 billion in Q2 FY2027 data-center revenue, up 117% year over year, with a 75% GAAP gross margin. This is a direct supplier revenue measure, not a measure of buyers’ AI investment returns.

What the evidence supports

This does not prove an AI bubble, nor that returns have arrived. The primary evidence supports a narrower conclusion: spending is real, cloud growth is real, buyer-side AI returns remain opaque, and supplier monetization is the most direct.

Reproducible financial table

Free cash flow = operating cash flow − cash capital spending. CapEx / OCF and capital-spending growth are calculated from the same underlying table. Reporting periods are not added together; a zero cloud-growth value means the company has no matching standalone cloud segment, not that the business had zero growth.

Financial base table for five buyers

Evidence ledger

Every source is an SEC filing or official investor-relations material. Original HTML is frozen as a local snapshot. URLs are for traceability; figures on this page come from the snapshots and structured CSV files.

Official source index

Why there is no payback forecast

This report deliberately avoids forecasting an AI payback year. The five companies generally do not disclose AI-specific revenue, capital spending, and operating profit together; without all three, filings cannot support a direct return-on-investment calculation.

Scope and limitations

Microsoft and Oracle use full-year periods; Alphabet, Amazon, and Meta use half-year periods. Ratios in the scatter plot compare cash-flow pressure, while absolute capital spending shows only disclosed scale. Cloud growth is not the same as AI revenue growth.

Research method

Freeze official HTML first, then extract the minimum structured fields. Query IR calculates every free-cash-flow figure and ratio. The disclosure matrix uses a documented 0 / 1 / 2 editorial code and retains source IDs.