Three ways reported capex understates what a company spends

Capital expenditure looks like one of the safest numbers in a filing. It is a single line in the investing section of the cash flow statement, it is audited, and there is no judgement in it — the company either paid for the asset or it did not.

The line is reliable. The inference from it is not. A company can spend heavily on productive capacity while reporting very little capex, and nothing about that is irregular or hidden. It just appears somewhere else in the same statement.

Here are the three cases that come up most often, each with a filing you can open.

1. The capacity sits in a joint venture

Sandisk reported $43 million of capital expenditure in its fiscal fourth quarter of 2026, on revenue of $8,965 million.

$43M ÷ $8,965M  =  0.48% of revenue

For a NAND flash manufacturer that figure is not merely low. Building flash memory requires fabrication plants that cost billions, and 0.48% of revenue does not build one.

The explanation is in the same statement, a few lines down. Sandisk’s fabrication capacity sits in Flash Ventures, a joint venture with Kioxia that is not consolidated. Money that would be capex at a company that owned its fabs instead appears as:

The lending to the joint venture is 2.9 times the reported capex in the quarter and 2.6 times across the year. A reader who stops at the capex line concludes that Sandisk is asset-light. It manufactures flash memory.

None of this is unusual and none of it is concealed. Joint ventures are a normal way to share the cost of a fab. But the accounting consequence is that a standard screen, which reads one line, gets a different answer from a reader who reads the statement.

How to spot it

• Capex far below depreciation and amortisation for a capital-intensive business
• Lines in the investing section referencing notes, loans or advances to an affiliate
• Equity-method losses on the income statement — the tell that an operating entity exists outside the consolidation

2. The asset was leased, not bought

If a company buys a data centre, the cash appears in investing activities as capex. If it signs a finance lease for the same building, the cash appears in financing activities as principal repayment, and capex never sees it.

The economics are close to identical. The presentation is not.

Amazon’s fiscal 2025 statements make the size of this visible:

Operating cash flow                       $139,514M
Purchases of productive assets           −$131,819M
───────────
Free cash flow, capex only                  $7,695M
Principal repayments of finance leases     −$1,557M
───────────
Free cash flow, also deducting leases       $6,138M

In fiscal 2025 the adjustment is $1.6 billion — noticeable but not decisive. It has not always been that small. In fiscal 2021 the same line was $11.2 billion, when the lease-financed build-out was at its peak. A free cash flow series that ignores it is measuring something different in 2021 than in 2025.

There is no correct answer here, only a disclosed one. SEC staff guidance notes that free cash flow has no uniform definition and that whoever presents it should say how it was calculated. Very few publications do.

How to spot it

• A Finance lease principal payments line in financing activities
• Right-of-use assets growing faster than property, plant and equipment
• A company that discusses “capacity” in its narrative but reports modest capex

3. The tag you queried is not the tag they used

The first two cases are about accounting. This one is about plumbing, and it is the one most likely to affect anyone building a screen.

There is more than one XBRL element for capital expenditure. For the CY2025 annual period:

13.0% of filers that reported capex at all used the second tag. Amazon is one of them — it last reported under the standard tag in 2017.

Query the standard tag for Amazon and you do not get an error. You get a nine-year-old value, or nothing. And nothing, in most pipelines, becomes zero somewhere downstream — which turns one of the largest capital spenders in the market into a company that apparently spends nothing.

This failure mode is silent by construction:

• It does not throw. A missing tag is an absence, not an exception.
• The result stays plausible. Zero capex sorts, charts and passes a range check.
• It compounds. Free cash flow, capital intensity, and any valuation built on them inherit it.

How to spot it

Query both tags and take whichever the company actually reported. If neither returns a value, treat capex as unconfirmed rather than zero — and do not compute anything that depends on it. An unconfirmed capex figure means no free cash flow figure, not a free cash flow figure that assumes the company built nothing.

A short checklist

Before using a capex number for anything that matters:

1. Capex ÷ revenue. Below ~1% for a manufacturer is a flag, not a finding.
2. Capex ÷ depreciation. Persistently below 1.0 means the asset base is shrinking, or the spending is somewhere else.
3. Scan investing activities for notes, loans or advances to affiliates and joint ventures.
4. Scan financing activities for finance-lease principal repayments.
5. Check the income statement for equity-method losses.
6. Confirm which tag the company used, if you pulled the figure programmatically.

None of this requires a subscription or a model. It requires opening the cash flow statement and reading past the line you came for.

What would make this wrong

• All three cases are legitimate accounting. Nothing here suggests any company misreported anything. The point is that the reported figure answers a narrower question than readers assume.
• Joint-venture funding is not equivalent to capex. Notes receivable to an affiliate are a loan, recoverable in principle, and consolidating them would be wrong. They are shown as evidence that capital intensity exists outside the capex line, not as a substitute for it.
• The tag counts are for one annual period. Tag usage migrates; InterestExpense moved to InterestExpenseNonoperating over roughly the same span. Check the counts for the period you care about rather than assuming these hold.
• A low capex ratio is not automatically a signal. Software companies genuinely spend little on property and equipment. The checklist above is for deciding whether to look further, not for concluding anything.

Check it yourself

The Sandisk cash flow statement is in Exhibit 99.1 of its 5 August 2026 8-K; the Flash Ventures lines are in the investing section. Amazon’s finance-lease line is in the financing section of its FY2025 10-K. The two XBRL endpoints return raw JSON and can be compared directly.

If a figure here does not match a filing, tell us and it will be corrected on the article and on the corrections log, with the date. A related case — where the standardised data reported borrowings roughly 4,874 times too small — is described in our note on how a number gets verified.