Is AI a bubble? Big Tech's cash flow statements answer part of it
Bubble arguments usually start with prices. We started with the companies' own cash flow statements. They do not say whether AI stocks are overvalued, but they show a clear shift: the AI buildout now consumes nearly all the cash the biggest spenders generate, and the rest is being borrowed.
There is no test that proves a bubble before it bursts. But one question can be answered from filings: can the companies building AI infrastructure pay for it out of the cash their businesses produce, or do they need outside money? The answer changed in 2026.
Spending versus cash generated
We compared purchases of property and equipment with net cash from operating activities, both from each company's cash flow statement, for the April to June quarter. Microsoft's quarter is its fiscal fourth quarter.
| Company | Capex Q2 2026 | Op. cash flow Q2 2026 | Capex Q2 2025 | Op. cash flow Q2 2025 |
|---|---|---|---|---|
| Microsoft | $35.8B | $55.4B | $17.1B | $42.6B |
| Alphabet | $44.9B | $39.1B | $22.4B | $27.7B |
| Amazon | $54.2B | $45.4B | $32.2B | $32.5B |
| Meta | $30.1B | $31.9B | $16.5B | $25.6B |
| Total | $165.1B | $171.8B | $88.2B | $128.5B |
Across the four, capital spending was $165.1 billion against $171.8 billion of operating cash flow in April to June 2026, or about 96 cents of every operating dollar. In the same quarter of 2025, spending was $88.2 billion against $128.5 billion, about 69 cents per dollar. Alphabet and Amazon spent more than their operating cash flow in the quarter.
Free cash flow, in the companies' own words
- Alphabet reported free cash flow of negative $5.9 billion for the quarter in its own reconciliation table. A year earlier, operating cash flow minus capital spending was positive $5.3 billion.
- Meta reported free cash flow of $784 million, down from $8.5 billion a year earlier, after $31.1 billion of capital spending including finance-lease payments.
- Amazon reported trailing twelve-month free cash flow of negative $7.6 billion, against positive $18.2 billion a year earlier, driven primarily, it said, by a $66.1 billion rise in property and equipment purchases.
Filling the gap with outside money
The same filings show the financing. In the second quarter, Alphabet raised $49.6 billion by selling common and mandatory convertible preferred stock, and $20.3 billion of senior notes; its long-term debt rose to $98.2 billion at June 30 from $46.5 billion at December 31. Meta issued $24.9 billion of debt, lifting long-term debt to $83.7 billion from $58.7 billion. Amazon's long-term debt rose to $128.9 billion from $65.6 billion over the same six months. NVIDIA, a supplier rather than a builder, issued $25.0 billion of notes.
The Bank of England noted in July that AI hyperscalers' bond issuance in the first half of 2026 had already exceeded their issuance for all of 2025, and that declining free cash flow increases their dependence on 'accommodative future refinancing conditions.' The Reserve Bank of Australia reports that nearly 30% of US investment-grade corporate bond issuance in the first half of 2026 came from technology-related firms Context: n: not disclosed · first half of 2026 · RBA Financial Stability Review, Oct 2026 · margin: not disclosed.
The other side of the ledger
This is not a picture of companies in distress. Operating cash flow at the four rose from $128.5 billion to $171.8 billion in a year, and revenue grew at every one of them. The Reserve Bank of Australia judges that balance-sheet leverage among listed AI firms remains low, and that hyperscalers can service debt from other businesses such as advertising and e-commerce.
What the data says, and what it does not
The data says the AI buildout has moved from being funded by spare cash to being funded partly by debt and equity markets. That makes the cycle more sensitive to interest rates and investor appetite than it was a year ago. It does not say the investment will fail. That depends on whether AI revenue keeps growing fast enough, which no filing can show in advance.
What it shows / What it doesn't
It shows
- How much of their operating cash flow the four largest AI spenders put into capital spending in April to June 2026, versus a year earlier.
- The companies' own free cash flow figures and the debt and equity they raised.
- What two central banks concluded from similar data.
It doesn't show
- Whether AI stocks are overvalued. Cash flow says nothing about the right price.
- Future returns on the AI investment.
- Off-balance-sheet commitments such as long-term leases, which the Reserve Bank of Australia estimates could be large.
- Results after June 2026.
Why it matters
Bubble debates generate heat; cash flow statements generate numbers. The shift from self-funded to market-funded AI spending is measurable and recent. If you trade AI stocks, rising rates or tighter credit now matter to the AI story in a way they did not in 2025.
Primary sources
- VerifiedAlphabet Q2 2026 results, including free cash flow reconciliation (Exhibit 99.1) (Alphabet, Form 8-K, SEC EDGAR)Cash flow, free cash flow, equity and debt raised, balance sheet.
- VerifiedMeta Q2 2026 results (Exhibit 99.1) (Meta, Form 8-K, SEC EDGAR)Cash flow, free cash flow, debt.
- VerifiedAmazon Q2 2026 results (Exhibit 99.1) (Amazon, Form 8-K, SEC EDGAR)Cash flow, trailing free cash flow, debt.
- VerifiedMicrosoft XBRL company facts (fiscal 2025 and 2026) (SEC EDGAR)Quarterly capex and operating cash flow derived from annual and nine-month values.
- VerifiedFinancial Stability Report press conference remarks, July 7, 2026 (Bank of England)Hyperscaler bond issuance and free cash flow.
- VerifiedFinancial Stability Review, October 2026, Box: Funding the AI investment boom (Reserve Bank of Australia)Investment-grade issuance share and leverage assessment.
In our directories
How we used AI: Capex and operating cash flow were pulled from the SEC's XBRL API and the companies' filed releases, then checked against each release. Ratios were computed by us from those figures.
Not financial advice. For information and education only. AI model outputs and research results can be wrong and do not predict future returns. Nothing here is a recommendation to buy or sell any asset. See our disclaimer and methodology.