Artificial Intelligence
Hyperscaler capital expenditure, first quarter
How much cash are Microsoft, Alphabet, Amazon and Meta actually putting into the buildout?
129.8USD billion, Q1
This is the capital cycle behind the AI trade, measured in cash that has already left the building rather than in guidance. It is the number revenue has to service eventually, and the one that turns a story about demand into a depreciation schedule.
Two caveats, stated here rather than buried. These companies do not disclose an AI-only capex line, so this is total property and equipment spend: it rose with AI, but it is not exclusively AI, and anyone quoting it as such is adding a precision the filings do not contain. And because it is a first-quarter comparison, it tracks the trajectory of the buildout, not its annual total.
How this is measured
Combined purchases of property and equipment reported by Microsoft, Alphabet, Meta (PaymentsToAcquirePropertyPlantAndEquipment) and Amazon (PaymentsToAcquireProductiveAssets) in their own SEC filings, read from the SEC XBRL company-concept API. We compare the first calendar quarter of each year, and only that quarter: these companies run different fiscal years, so their annual totals do not line up, and Q1 is the one period all four report on the same three-month calendar basis. A year is plotted only once all four have filed. This is total capex, not an AI-only figure — no public filing breaks out AI spend, and we will not estimate one.
All observations
| Date | USD billion, Q1 |
|---|---|
| 31 Mar 2026 | 129.8 |
| 31 Mar 2025 | 71.9 |
| 31 Mar 2024 | 44.3 |
| 31 Mar 2023 | 33.9 |
| 31 Mar 2022 | 35.5 |
| 31 Mar 2021 | 27.4 |
| 31 Mar 2020 | 20.1 |
| 31 Mar 2019 | 14.3 |
| 31 Mar 2018 | 16.1 |