For anyone holding the S&P 500. Big tech's AI spending is now propping up the index. This page asks one thing, from signed contracts: how long before the music could stop?
Each figure is the remaining performance obligation a company reports in its own accounts — revenue under signed contract that has not been delivered yet. Every link below opens the actual filing on the SEC's site.
Pulled automatically from the SEC's XBRL company-facts API
(data.sec.gov), tag RevenueRemainingPerformanceObligation.
Nothing here is typed in by hand.
How to read the gauge: the SECURE / MEDIUM / EXPOSED bands are a convention chosen for this page, not a market standard. Over 3 years of cover spans a real part of the five-to-six-year life these data centres are depreciated over; under 1.5 years is shorter than the capex programmes already in motion. Backlog also covers all cloud contracts, not AI alone, and it is revenue while capex is spend — so read this as "how long the committed money lasts at today's rate", not as an exact expiry date.
Think of it as a household. The business earns money each year. Building data centres is the spending. Under $1 per $1 earned, they can pay from income. Over $1, they dip into savings or borrow.
| Company | Verdict | Spends this much building for every $1 its business earns | Cushion |
|---|---|---|---|
| ORCL | spending more than it earns | $1.61 | 1.3 years of cash left |
| AMZN | spending more than it earns | $1.07 | 10.6 years of cash left |
| GOOGL | pays its own way | 71c | could spend 40% more and still cope |
| META | pays its own way | 69c | could spend 46% more and still cope |
| MSFT | pays its own way | 63c | could spend 58% more and still cope |
| NVDA | pays its own way | 5c | spends only a fraction of what it earns |
NVDA is in this list for contrast: it sells the chips rather than building data centres, so it barely spends anything. Its customers are the other five.
Cash in the bank divided by how fast it is going out. This is how long before they must borrow again — not how long before they fail. All of them can and do borrow easily.
| Company | Cash + short-term investments | Burning per year | Years of cash | Capex cut needed to break even |
|---|---|---|---|---|
| ORCL | $37B | $28.7B | 1.3 | 38% |
| AMZN | $123B | $11.6B | 10.6 | 7% |
Each of these pays for a slice of the total buildout. If one stops, that slice stops with it — and the orders it would have placed disappear.
| If this stops building… | Orders that disappear | Share of the buildout |
|---|---|---|
| AMZN | $173B a year | 34% of all the building stops |
| GOOGL | $132B a year | 26% of all the building stops |
| MSFT | $116B a year | 23% of all the building stops |
| META | $89B a year | 17% of all the building stops |
Nvidia is the one to watch. It sells the chips to all of them and generates $134B of cash a year while spending almost nothing ($7B) itself. Its earnings are their spending. If any large buyer pauses, that shows up in Nvidia first.
Some of the money goes in circles. About 52% of the named deals are between companies that also invest in each other — a chip maker funds a customer, the customer buys its chips. That revenue is real, but it is not new money from outside.
None of this is a prediction. It is what the money looks like today.
Above the red line, building costs more than the business brings in.
| Company | TTM capex ($B) | TTM op. cash flow ($B) | capex ÷ OCF | Flag |
|---|---|---|---|---|
| ORCL | 75.7 | 46.9 | 1.612 | above 1.0 |
| AMZN | 173.0 | 161.4 | 1.072 | above 1.0 |
| GOOGL | 132.4 | 185.7 | 0.713 | |
| META | 89.3 | 130.3 | 0.686 | |
| MSFT | 115.9 | 182.9 | 0.634 | |
| NVDA | 7.4 | 134.4 | 0.055 |
Single quarters jump around — Google pays most of its tax in one quarter, which dents that quarter's earnings. The yearly figures above are the fair read.
| Company | Quarter ending | Capex ($B) | Op. cash flow ($B) | Free cash flow ($B) |
|---|---|---|---|---|
| MSFT | 2023-09-30 | 9.9 | 30.6 | 20.7 |
| MSFT | 2023-12-31 | 9.7 | 18.9 | 9.1 |
| MSFT | 2024-03-31 | 11.0 | 31.9 | 21.0 |
| MSFT | 2024-06-30 | 13.9 | 37.2 | 23.3 |
| MSFT | 2024-09-30 | 14.9 | 34.2 | 19.3 |
| MSFT | 2024-12-31 | 15.8 | 22.3 | 6.5 |
| MSFT | 2025-03-31 | 16.7 | 37.0 | 20.3 |
| MSFT | 2025-06-30 | 17.1 | 42.6 | 25.6 |
| MSFT | 2025-09-30 | 19.4 | 45.1 | 25.7 |
| MSFT | 2025-12-31 | 29.9 | 35.8 | 5.9 |
| MSFT | 2026-03-31 | 30.9 | 46.7 | 15.8 |
| MSFT | 2026-06-30 | 35.8 | 55.4 | 19.6 |
| GOOGL | 2023-09-30 | 8.1 | 30.7 | 22.6 |
| GOOGL | 2023-12-31 | 11.0 | 18.9 | 7.9 |
| GOOGL | 2024-03-31 | 12.0 | 28.8 | 16.8 |
| GOOGL | 2024-06-30 | 13.2 | 26.6 | 13.5 |
| GOOGL | 2024-09-30 | 13.1 | 30.7 | 17.6 |
| GOOGL | 2024-12-31 | 14.3 | 39.1 | 24.8 |
| GOOGL | 2025-03-31 | 17.2 | 36.1 | 19.0 |
| GOOGL | 2025-06-30 | 22.4 | 27.7 | 5.3 |
| GOOGL | 2025-09-30 | 24.0 | 48.4 | 24.5 |
| GOOGL | 2025-12-31 | 27.9 | 52.4 | 24.6 |
| GOOGL | 2026-03-31 | 35.7 | 45.8 | 10.1 |
| GOOGL | 2026-06-30 | 44.9 | 39.1 | -5.9 |
| AMZN | 2023-09-30 | 12.5 | 21.2 | 8.7 |
| AMZN | 2023-12-31 | 14.6 | 42.5 | 27.9 |
| AMZN | 2024-03-31 | 14.9 | 19.0 | 4.1 |
| AMZN | 2024-06-30 | 17.6 | 25.3 | 7.7 |
| AMZN | 2024-09-30 | 22.6 | 26.0 | 3.4 |
| AMZN | 2024-12-31 | 27.8 | 45.6 | 17.8 |
| AMZN | 2025-03-31 | 25.0 | 17.0 | -8.0 |
| AMZN | 2025-06-30 | 32.2 | 32.5 | 0.3 |
| AMZN | 2025-09-30 | 35.1 | 35.5 | 0.4 |
| AMZN | 2025-12-31 | 39.5 | 54.5 | 14.9 |
| AMZN | 2026-03-31 | 44.2 | 26.0 | -18.2 |
| AMZN | 2026-06-30 | 54.2 | 45.4 | -8.8 |
| META | 2023-09-30 | 6.5 | 20.4 | 13.9 |
| META | 2023-12-31 | 7.6 | 19.4 | 11.8 |
| META | 2024-03-31 | 6.4 | 19.2 | 12.8 |
| META | 2024-06-30 | 8.2 | 19.4 | 11.2 |
| META | 2024-09-30 | 8.3 | 24.7 | 16.5 |
| META | 2024-12-31 | 14.4 | 28.0 | 13.6 |
| META | 2025-03-31 | 12.9 | 24.0 | 11.1 |
| META | 2025-06-30 | 16.5 | 25.6 | 9.0 |
| META | 2025-09-30 | 18.8 | 30.0 | 11.2 |
| META | 2025-12-31 | 21.4 | 36.2 | 14.8 |
| META | 2026-03-31 | 19.0 | 32.2 | 13.2 |
| META | 2026-06-30 | 30.1 | 31.9 | 1.7 |
| ORCL | 2023-11-30 | 1.1 | 0.1 | -0.9 |
| ORCL | 2024-02-29 | 1.7 | 5.5 | 3.8 |
| ORCL | 2024-05-31 | 2.8 | 6.1 | 3.3 |
| ORCL | 2024-08-31 | 2.3 | 7.4 | 5.1 |
| ORCL | 2024-11-30 | 4.0 | 1.3 | -2.7 |
| ORCL | 2025-02-28 | 5.9 | 5.9 | 0.1 |
| ORCL | 2025-05-31 | 9.1 | 6.2 | -2.9 |
| ORCL | 2025-08-31 | 8.5 | 8.1 | -0.4 |
| ORCL | 2025-11-30 | 12.0 | 2.1 | -10.0 |
| ORCL | 2026-02-28 | 18.6 | 7.2 | -11.5 |
| ORCL | 2026-05-31 | 16.5 | 14.6 | -1.9 |
| ORCL | 2026-08-31 | 28.5 | 23.1 | -5.4 |
| NVDA | 2023-10-29 | 0.3 | 7.3 | 7.1 |
| NVDA | 2024-01-28 | 0.3 | 11.5 | 11.2 |
| NVDA | 2024-04-28 | 0.4 | 15.3 | 15.0 |
| NVDA | 2024-07-28 | 1.0 | 14.5 | 13.5 |
| NVDA | 2024-10-27 | 0.8 | 17.6 | 16.8 |
| NVDA | 2025-01-26 | 1.1 | 16.6 | 15.6 |
| NVDA | 2025-04-27 | 1.2 | 27.4 | 26.2 |
| NVDA | 2025-07-27 | 1.9 | 15.4 | 13.5 |
| NVDA | 2025-10-26 | 1.6 | 23.8 | 22.1 |
| NVDA | 2026-01-25 | 1.3 | 36.2 | 34.9 |
| NVDA | 2026-04-26 | 1.8 | 50.3 | 48.6 |
| NVDA | 2026-07-26 | 2.7 | 24.1 | 21.4 |
Revenue customers have signed for but not yet paid. This is the audited answer to "how much is really promised".
Left out because the figure is years out of date: AMZN (last reported 2020-06-30).
| Company | Backlog ($B) | As of |
|---|---|---|
| MSFT | 684.0 | 2026-06-30 |
| ORCL | 664.0 | 2026-08-31 |
| GOOGL | 519.5 | 2026-06-30 |
| NVDA | 3.2 | 2026-07-26 |
Cloud contracts are ordinary business, so companies never have to file them. Everything here comes from news reports, not accounts. Treat the numbers as claims.
| Who is paying whom | Announced | Size | Flags |
|---|---|---|---|
| OpenAI → Oracle | 2025-09-10 | $300B | high |
| OpenAI → Microsoft | 2025-10-28 | $250B | circular medium |
| Nvidia → OpenAI | 2025-09-22 | $100B | circular low |
| OpenAI → Amazon | 2025-11-03 | $38B | medium |
| Nvidia → Anthropic | 2025-11-18 | $10B | circular low |