MARKETCRASH.NET
AI capex · SEC filings · credit spreads

AI Debt — The Build-Out Running on Borrowed Money

The biggest AI builders spend hundreds of billions of dollars a year on datacentres. Most of that comes out of cash flow, but not all of it. This page shows who is exposed, from their own SEC filings, and what the credit market charges the weakest borrowers over the strongest.

CONTAINED0.8xCASH-FUNDED

the five builders' debt together, in years of their combined capital spending — the reading this page is scored on, and the one shown on the dashboard.

The Build-Out in Three Numbers

What the largest AI builders spend, what they owe, and what the credit market charges weak borrowers over strong ones. Readings as of Sep 17, 2026.

Capital spending per year
$586.4bn
All 5 builders combined, trailing four reported quarters
Long-term debt
$459.5bn
All 5 builders combined, latest reported balance sheets
CCC minus AAA spread
10.43 pp
Percentage points, as of Sep 15, 2026 — all-sector rating buckets
Debt ÷ capex, all builders together
0.8x
CASH-FUNDED — the whole build-out on one ratio. Individual builders range from 0.3x to 1.6x; the table below ranks them.

Why this can crash the market

The AI boom is now a borrowing story, not only a stock story. Hundreds of billions of dollars a year in datacentre construction flow through the revenue of chip makers, builders and power companies. That works only while credit markets keep lending. If lending stops, the weakest builder breaks first, the spending stops second, and every revenue line built on that spending breaks third.

Who Is Borrowing to Build

Capital expenditure and long-term debt as each company reported it to the SEC. The periods differ between companies, and sometimes between one company's capex and its debt, so every figure carries its own date.

The 5 Builders, Ranked by Leverage

Source: SEC EDGAR company facts (XBRL), from each filer's own quarterly reports

MOST LEVERAGED: ORCL 1.6x
BuilderCapex per yearLong-term debtDebt ÷ capex
Oracle
ORCL
$75.7bn
ttm to Aug 31, 2026
$117.7bn
as of Aug 31, 2026
1.6x
LEVERAGED
Meta
META
$89.3bn
ttm to Jun 30, 2026
$83.7bn
as of Jun 30, 2026
0.9x
CASH-FUNDED
Amazon
AMZN
$173.0bn
ttm to Jun 30, 2026
$128.9bn
as of Jun 30, 2026
0.7x
CASH-FUNDED
Alphabet
GOOGL
$132.4bn
ttm to Jun 30, 2026
$98.2bn
as of Jun 30, 2026
0.7x
CASH-FUNDED
Microsoft
MSFT
$115.9bn
ttm to Jun 30, 2026
$31.1bn
as of Jun 30, 2026
0.3x
CASH-FUNDED
All 5 combined$586.4bn$459.5bn

Sorted by debt-to-capex, most leveraged first — the ratio is the point of this page: it is the number of years of construction, at the current rate, that each company already owes.

Full explanation of the AI Capex vs Debt ranking →

What the Credit Market Charges the Weakest Borrowers

If the financing of the AI build-out breaks, it shows up here first: in what the market charges to lend to weak borrowers instead of strong ones.

CCC Minus AAA Credit Spread

Source: FRED BAMLC0A1CAAA and BAMLH0A3HYC — ICE BofA AAA and CCC option-adjusted spreads, daily • Latest: Sep 15, 2026 • The last three years

WEAK CREDITORS PUNISHED
CCC minus AAA spread, percentage points
10.43▲ 0.05 (0.48%) vs prevSep 15, 2026

This chart has a limit. It is not a technology-sector measure and not a credit default swap, because single-name CDS quotes are licensed and cannot be republished here. AAA and CCC are all-sector rating buckets, used as a proxy: cash-rich hyperscalers borrow at one end, leveraged datacentre builders at the other. A wider gap means the market charges the weakest borrowers more for the money the build-out needs.

Full explanation of the CCC Minus AAA Credit Spread chart →

How to Read This Page

Every capex and debt figure comes from the company's own SEC filings, with the period printed underneath. The spread chart is a proxy, as stated above. This page gives no bubble probability and no timing, because neither can be computed from this data. It gives the exposure itself, ranked, so you can see who owes what. Read it with the oil shock monitor and the rest of the risks section . An expensive market absorbs an outside shock much worse than a cheap one.

Frequently Asked Questions About AI Debt

Is the AI boom a bubble?

This page computes no bubble probability, because that number cannot be measured. What can be measured is how the boom is financed: how much the biggest builders spend each year, how much debt they carry, and what the credit market charges weak borrowers over strong ones. All three come from SEC filings and Federal Reserve data, and all three would move first if the financing stopped working.

Who is funding AI datacentre construction?

The largest builders fund it in very different ways. Most pay for the whole build-out from operating cash, and their debt equals a fraction of one year of spending. One borrows heavily: its long-term debt equals more than two years of construction at the current rate, so its build-out continues only while the credit market keeps lending. The table on this page ranks the builders by that ratio, from their own SEC filings.

What happens if AI capex stops?

The spending goes to chip makers, server vendors, construction firms and power suppliers, so a stop would hit every revenue line built on it. Companies that funded construction with debt would still carry that debt after the spending ends. This is why the debt-to-capex ratio matters: it shows how many years of building each company already owes. The credit spread shows what the market charges the weakest borrowers, and a wider gap is the first sign of financing stress.

Is the credit spread on this page an AI-specific measure?

No. The chart shows the gap between AAA and CCC corporate borrowing costs across all sectors. It is not a technology index and not a credit default swap, because single-name CDS quotes are licensed and cannot be republished. The rating buckets work as a proxy: AAA is about where cash-rich hyperscalers borrow, and CCC is about where the most indebted datacentre builders borrow.

Where do the capex and debt figures on this page come from?

Every figure comes from the company's own SEC filings: capital expenditure summed over the last four reported quarters, and long-term debt from the latest balance sheet. Reporting periods differ between companies, and sometimes between one company's capex and its debt, so each number is printed with the period it covers. Nothing is estimated to force a common date.

Why does Oracle stand out from the other AI builders?

Its long-term debt equals more than two years of datacentre construction at its current spending rate. The other large builders carry debt worth less than one year of theirs. Oracle finances the build-out with borrowed money in a way Microsoft, Amazon, Alphabet and Meta do not, so its ratio is the one that depends on the credit market staying open.

Get told when the financing cracks

The credit spread and the builders' debt, next to the crash-risk score and the other risk indicators. One email a week when the score moved.

At most one email a week, and only in weeks something actually moved. No account, one click to unsubscribe, and the address is never shared.

Explore Related Indicators

Risks — What Could Crash the Market

The section hub — external shocks that reprice US equities, tracked with live data

Oil Shock

Tanker transits through Hormuz and Bab el-Mandeb, against the Brent price

Treasury Dump

Japan's trillion-dollar Treasury holdings, the yen and the 10-year yield

Debt Spiral

The interest bill on the US debt, against the post-war record share of spending

Market Crash Risk

The composite crash-risk score and the signals behind it

Fear and Greed Index

Market sentiment from volatility, credit spreads, momentum and safe-haven demand

Recession Probability

The live combined recession risk assessment

Methodology

Every formula, source series and threshold behind these numbers

Company figures sourced from SEC filings as noted above. For educational purposes only. Not investment advice.