What the biggest cloud and chip companies actually spend on infrastructure, taken straight from their quarterly filings. Updated weekly.
Capital expenditure by quarter, with revenue and R&D for context. Pick a different measure to change what the table shows.
Cash spent on property and equipment in the quarter. For the big cloud providers this is mostly data centres, servers and chips.
| Trend | Filing | ||||||
|---|---|---|---|---|---|---|---|
AmazonAMZN | Q2 2026 | $54.2B | +22.6% | +68.4% | 27.0% | SEC | |
AlphabetGOOGL | Q2 2026Derived | $44.9B | +25.9% | +100.1% | 37.5% | SEC | |
MicrosoftMSFT | Q2 2026Derived | $35.8B | +16.0% | +109.6% | 39.8% | SEC | |
MetaMETA | Q1 2026 | $19.0B | -11.2% | +46.8% | 33.7% | SEC | |
OracleORCL | Q2 2026Derived | $16.5B | -11.5% | +81.6% | 86.0% | SEC | |
MicronMU | Q2 2026Derived | $7.83B | +22.5% | +166.4% | 18.9% | SEC | |
CoreWeaveCRWV | Q1 2026 | $7.70B | +89.5% | +446.9% | 370.3% | SEC | |
TeslaTSLA | Q2 2026Derived | $5.79B | +132.2% | +141.8% | 20.5% | SEC | |
IntelINTC | Q2 2026Derived | $2.56B | -29.7% | -28.0% | 15.8% | SEC | |
AppleAAPL | Q2 2026Derived | $2.46B | +24.6% | -29.1% | 2.2% | SEC | |
NvidiaNVDA | Q1 2026 | $1.76B | +36.8% | +43.2% | 2.2% | SEC | |
AMDAMD | Q2 2026Derived | $0.81B | +107.7% | +186.5% | 7.0% | SEC | |
BroadcomAVGO | Q1 2026Derived | $0.23B | -7.6% | +60.4% | 1.0% | SEC | |
SpaceXSPCX | – | – | – | – | – | Not enough quarters | SEC |
The AI CapEx Tracker follows quarterly capital spending at the companies building and supplying AI infrastructure, sourced directly from SEC filings.
Capital expenditure is the clearest public signal of how much conviction these companies have. Marketing language changes quarter to quarter. Spending commitments do not.
Reading it as a share of revenue matters more than the raw number. A company spending thirty per cent of revenue on buildout is making a very different bet from one spending five per cent, even if the absolute figures look similar.
If you are deciding whether to build on hosted AI services or run your own hardware, this is the supply side of that question. Sustained heavy investment means more capacity and, historically, falling prices per unit of compute.
Related Tool
The spending above is the supply side. This tool works out the cost side for your own project, comparing hosted APIs against renting GPUs and buying hardware.
What that buildout costs you per token today, across 50+ providers.
Live cloud GPU rental rates, the clearest downstream signal of added capacity.
Find the break-even point between your own hardware and cloud API spend.
What all this compute actually buys in model capability.
It refreshes weekly, but the underlying data only moves when a company files. Most file within about six weeks of a quarter closing, so expect a lag of roughly one to two months after the quarter ends.
It has not filed yet, or it reported the period in a form we cannot split into a single quarter. Blank means no filed figure, never zero.
No, and that is worth being clear about. Capital expenditure covers all property and equipment, including offices and retail space. For the big cloud providers the large majority is data centre capacity, but these filings do not break out an AI-only line.
Nvidia designs chips and has them manufactured elsewhere, so it does not carry the cost of building fabs. Its role here is as the supplier the others are spending with, which shows up in its revenue rather than its capital spending.
They are private, so they file nothing with the SEC and there is no verified figure to show. Their spending still shows up here indirectly, because it runs through the cloud providers and chip suppliers that are listed. SpaceX only became trackable when it listed in 2026.
Because Tesla, SpaceX and Apple are shown for context but left out of the total. Their capital spending is real, it just mostly buys car plants, rockets and retail space rather than AI capacity.
Yes. It is public filing data. Cite SEC EDGAR as the source, and link back here if the calculation is useful to you.
