Free Cash Flow: Not a GAAP Number, So Compute It Yourself

Salesforce earned $7.457 billion in fiscal 2026. Salesforce also earned $14.402 billion in fiscal 2026. Both numbers are correct, both describe the twelve months ended January 31, 2026, and they are about as far apart as two figures describing one year of one company can get.^[1]

The first is net income. The second is free cash flow, and it's the number almost every valuation you will ever read actually runs on. It leads the earnings release. It's the input to a discounted cash flow model. It's what people mean when they say a business “prints cash.”

Here's the part that took me an embarrassingly long time to notice. Search Salesforce's fiscal 2026 Form 10-K, the audited annual report, for the phrase “free cash flow.” You get nothing. Zero hits. The most quoted number in the story is not in the audited document at all.^[1] It lives in the press release, where the company writes its own definition.^[2]

Two lines. That is the whole calculation.

Free cash flow, in its most common form, is operating cash flow minus capital expenditures. Both inputs are real GAAP line items you can point at. Operating cash flow sits at the bottom of the first section of the cash flow statement. Capital expenditures sits in the second section, under investing. You subtract one from the other and you're done.

Salesforce publishes the arithmetic itself, in a table titled “Computation of Free Cash Flow, a Non-GAAP Measure.”^[2]

                                          Q4 FY26     FY26      FY25
GAAP net cash provided by
  operating activities                      5,464    14,996    13,092
Capital expenditures                         (141)     (594)     (658)
                                          -------   -------   -------
Free cash flow                              5,323    14,402    12,434

That's it. If you can read two rows of a cash flow statement, you can compute free cash flow for any public company on EDGAR in about ninety seconds, and you never have to take a company's word for it again. Do that once and the number stops being a slide and starts being something you own.

Why it comes out at nearly double net income

Net income is an accrual number. It records revenue when it's earned and expenses when they're incurred, whether or not cash moved. Operating cash flow starts from net income and reverses every place those two things disagreed.

This is also why free cash flow and the price-to-earnings ratio can tell opposite stories about the same company. A software business with heavy amortization from acquisitions looks mediocre on earnings and excellent on cash. A capital-hungry manufacturer looks fine on earnings and terrible on cash. Same reality, two lenses, and the lens is doing a lot of the work.

The stock comp fight, in one company's numbers

Now the argument that has divided tech investors for a decade, and the one that actually matters at these numbers.

Salesforce expensed $3,509 million of stock-based compensation in fiscal 2026, which is $3.509 billion. Because paying people in shares costs the company no cash, it gets added back inside operating cash flow, and nothing downstream takes it out. That single line is 24% of the reported free cash flow.^[1] Subtract it and $14.4 billion becomes about $10.9 billion. That is not a rounding difference. That is the difference between two entirely different companies.

The bull case is straightforward: it's not cash, so a cash-flow measure shouldn't include it. Fine. But the cash flow statement answers that argument two sections further down, in Salesforce's own filing.

That $12.7 billion of buybacks equals 88% of the free cash flow the company reported. You can call stock comp non-cash. The cash used to mop up the dilution it creates is extremely cash. If you want more on why companies do this and what it hides, I wrote about how buybacks actually work separately.

My rule is simple. I compute free cash flow twice: once as reported, once with stock comp subtracted, and I look at the gap as a percentage. Under 5%, ignore it. Over 20%, the reported number is a marketing number. Salesforce is at 24%, and it isn't close to the worst offender in its own sector.

One quarter looks incredible. The next looks broken.

The other thing nobody warns you about is that free cash flow inside a year is nearly meaningless, because working capital moves in lumps. Salesforce sells annual subscriptions. It bills a huge share of them in its January-ending fourth quarter, then collects the cash in the February-to-April quarter that follows. Here's what that does.

The two lines doing most of that work are visible in the filing. Unearned revenue, which is cash customers paid before Salesforce delivered anything, added $2,924 million to operating cash flow for the year. Accounts receivable, which is revenue booked but not yet collected, took $2,160 million back out.^[1] Those two lines alone are worth more than a third of the reported free cash flow, and neither has anything to do with whether the software is any good.

Levered, unlevered, and the mistake that ruins a DCF

There are two versions of free cash flow and mixing them up is the most common valuation error I see.

• Levered free cash flow is measured after the lenders have been paid. Operating cash flow already subtracts cash interest, so the standard operating-cash-flow-minus-capex figure is a levered number. It belongs to shareholders.
• Unlevered free cash flow is measured before any financing. You add back interest, after tax, so the number describes the business regardless of how it happens to be funded. It belongs to everyone who financed the company, debt and equity together.

Salesforce paid $276 million of cash interest in fiscal 2026 at a 22% effective tax rate, so adding back the after-tax cost takes its $14.4 billion to roughly $14.6 billion unlevered.^[1] A trivial adjustment here. For a private-equity-owned business carrying six turns of debt, the same adjustment can be half the number.

The error is discounting a levered cash flow at the weighted average cost of capital, which already accounts for debt. Do that and you count the benefit of leverage twice and hand yourself a valuation that is too high, sometimes by a lot. Unlevered cash flow discounted at WACC gives you enterprise value. Levered cash flow discounted at the cost of equity gives you equity value. If that distinction is fuzzy, the pieces on how a discounted cash flow model is built and market cap versus enterprise value are the two to read next, in that order.

“Free” is doing a lot of work in that name

The SEC has thought about this, and its staff guidance on non-GAAP measures is blunter than anything you will find in an investor deck.

Companies should be aware that this measure does not have a uniform definition and its title does not describe how it is calculated. [...] “Free cash flow” should not be used in a manner that inappropriately implies that the measure represents the residual cash flow available for discretionary expenditures, since many companies have mandatory debt service requirements or other non-discretionary expenditures that are not deducted from the measure.^[3]

Read that against what Salesforce actually did with its $14.4 billion. It spent $12.7 billion on buybacks, $1.6 billion on dividends, and $9.3 billion buying Informatica, the data-management company it closed on in November 2025. To fund that it borrowed $6.0 billion and finished the year with $1.5 billion less cash than it started with.^[1]

So a company with $14.4 billion of “free” cash flow ended the year having drawn down its cash balance and taken on new debt. The cash was real. Free it was not.

What I actually do with the number

Four steps, and none of them involve an investor deck.

1. Pull the 10-K on EDGAR and compute it myself. Operating cash flow minus capital expenditures, from the audited statement. If my number disagrees with the company's headline, the company has redefined something, and finding out what is usually the most interesting thirty minutes of the whole exercise.
2. Compute it a second time with stock comp subtracted. Then look at the gap as a percentage of the reported figure. That ratio, more than almost anything else, tells me who is being paid in real money and who is being paid in dilution.
3. Use trailing twelve months, never a quarter. And read the working capital lines to see whether a good year was a good year or a well-timed collection cycle.
4. Decide levered or unlevered before opening a model. Then match the discount rate to it, every time.

None of this is hard. It's two subtractions and about ten minutes. What surprises me is how rarely anyone does it, given that the entire edifice of modern valuation is balanced on a number that no accounting standard defines and that the company reporting it gets to name.

Go pull a filing for something you own and run the two lines. You will either confirm the headline or find out something the headline was hoping you would not.