16 US companies now spend more on capex than they earn

Capital expenditure exceeding operating cash flow is not, by itself, a problem. It is what a company does when it is building something. It is also what a company does when its earnings have collapsed and its construction schedule has not.

Telling those apart requires knowing who is on the list. So we built the list.

How the screen works

Capital expenditure  ÷  Operating cash flow  >  1.0

Three calls to the SEC’s XBRL frames API return every filer that reported each figure for the period. From there:

Two exclusions do the real work, and both are judgement calls we are stating openly.

Six of the 46 have no ticker. They are wholly-owned subsidiaries that file with the SEC because they have public debt — Public Service Co of Colorado and Northern States Power (both Xcel), Berkshire Hathaway Energy, FirstEnergy Transmission, Cheniere Corpus Christi Holdings, NCL Corporation. Leaving them in double-counts their parents.

Twenty-four of the remaining 40 are utilities. A regulated utility spending more than it earns is not news; it is the business model. Rate base is grown with debt and equity issuance and recovered through rates over decades. Including them would bury the interesting cases under twenty-four companies doing exactly what they are supposed to do.

The sixteen

Together the sixteen generated $69.9B from operations and spent $116.4B on capital assets — a gap of $46.5 billion funded from cash, debt, or equity rather than from the year’s operations.

The same ratio, four different situations

A single number puts these companies on one list. It does not make them one thing.

Building compute. CoreWeave, Oracle, Equinix and Iron Mountain together collected $40.3B and spent $72.6B — 180% — and in each case the spending is on datacentre capacity. Oracle is the starkest: capital expenditure rose 162% year over year, from $21.2B to $55.7B, while operating cash flow rose 54%. The numerator is moving several times faster than the denominator.

Replacing fleets. Norwegian Cruise Line, Southwest, American and U-Haul buy ships, aircraft and trucks. These are long-lived assets bought in lumps, and a single year’s ratio says more about delivery schedules than about the business.

Cyclicals at a trough. Dow, International Paper, Nucor and Boeing are on this list substantially because the denominator fell. Dow generated $1.0B of operating cash flow against $2.5B of capex; the capex is ordinary for a chemicals company, the cash flow is not.

Intel is its own case. It appears at 151%, but capital expenditure actually fell 38.8% year over year, from $23.9B to $14.6B, while operating cash flow rose. It is on the list because it has not yet finished exiting a spending cycle, not because it is entering one.

The ratio is a question, not an answer. Four of these companies are spending because demand is running ahead of capacity. Four are on the list because earnings fell. The screen cannot tell them apart — reading the filings can.

Two things that break this screen if you do not handle them

Companies do not use the same capex tag. 3,512 filers report PaymentsToAcquirePropertyPlantAndEquipment; a separate 526 report PaymentsToAcquireProductiveAssets instead. Seven of the sixteen above — Dow, Ventas, Southwest, American, Liberty Global, Equinix and International Paper — are in the second group. A screen that reads only the standard tag returns no capex for them, and no capex silently becomes zero, which means they never appear on a list of companies spending too much. The tag they use is noted in the data behind every row.

“CY2025” is not calendar 2025 for everyone. The SEC assigns each annual period to the nearest calendar frame. Fourteen of these sixteen closed on or about 31 December 2025. Two did not:

Oracle’s figure therefore includes five months that no other company on this list has reported yet. That does not make it wrong — it is Oracle’s most recent audited year — but it does mean the rows are not measuring an identical stretch of time, and anyone using this table to compare Oracle against Intel should know that.

What would make this wrong

• The screen is as-filed and unadjusted. No normalisation for acquisitions, asset sales, leases, or partner reimbursements. Intel’s gross capex, for example, is not net of any government or partner contributions it may receive.
• REIT capex is not comparable to industrial capex. Ventas, Iron Mountain and Equinix are REITs, where property acquisition and development sit in investing activities in ways that make the ratio structurally higher. They are shown because they meet the test as filed, not because the comparison is apples to apples.
• A $1B floor is arbitrary. It removes small filers where a single project distorts the ratio, and it also removes real companies. A different floor produces a different list.
• One year is one year. Nothing here says whether the spending will earn a return, and no company on this list is being criticised for appearing on it.
• We spot-checked three rows, not sixteen. CoreWeave, Oracle and Intel were verified against their own filings. The remaining thirteen come from the frames data as filed.

Check it yourself

The three API endpoints are linked below and return raw JSON — the same data this table was built from, with no intermediate step. Filter to operating cash flow above $1B, take the larger of the two capex tags, and divide.

If a row here does not match a company’s filing, tell us and it will be corrected on the article and on the corrections log, with the date. The methodology behind reconciling tagged data against filing text is described in our note on how a number gets verified.