P/E vs EV/EBITDA vs DCF: Which Method Works Best?

Oracle made more money in its 2025 financial year than in 2024. Net income rose from $10.5 billion to $12.4 billion — a good year by the measure most people look at first. Over those same two years its free cash flow went from $11.8 billion to minus $0.4 billion.

Nothing was restated and nothing improper happened. The company simply spent $21.2 billion on capital equipment in a year it had spent $6.9 billion the year before, and cash flow is measured after that spending while earnings are not. Which means a P/E ratio, an EV/EBITDA multiple and a discounted cash flow model would each have told a different story about Oracle that year, and all three would have been arithmetically correct. They were not competing estimates of one number. They were answers to three different questions, and the person asking usually does not realise they have chosen the question.

Most writing on this topic hands you a table of pros and cons and leaves you to pick by temperament, which is not a method. What follows is a test you can run on any company in five minutes, using two numbers from the filings, that tells you which of the three you are entitled to trust. This is a lesson in how to choose a valuation approach, not a view on any of the companies used to illustrate it.

What is each method actually measuring?

Strip away the jargon and the three differ on exactly two things: how they treat debt, and how they treat the cost of staying in business.

P/E is share price divided by earnings per share, and earnings are struck after interest and after depreciation. So a P/E already accounts for the cost of debt and for an accountant's estimate of wear and tear — but not for what the company actually had to spend on new assets this year, and not for the debt itself. EV/EBITDA takes the market value, adds the debt, subtracts the cash, and divides by earnings before interest, tax, depreciation and amortisation. Putting debt in the numerator and pulling interest out of the denominator is what makes two companies with different borrowing comparable; pulling depreciation out is what makes it blind to capital spending. A DCF is the present value of the cash a business will actually generate after paying for its own upkeep and growth. It is the only one of the three that asks what ends up in an owner's pocket, and the only one that requires you to forecast — which is both its virtue and the reason it is so easy to abuse.

Think of a farm. P/E counts the harvest after the bank's interest has been paid. EV/EBITDA counts the harvest before the interest and before anything is set aside for a new tractor. A DCF counts what is left in your pocket after you have actually bought the tractor. In an ordinary year those three numbers sit close together and the argument between them is academic. In a year you buy three tractors, they are not variations on one answer — they are three different answers, and only one of them is about you.

The one test: how far apart are earnings and cash?

Before choosing a method, take two numbers off the last two annual filings — net income, and free cash flow, which is cash from operations minus capital expenditure. The distance between them is the whole argument.

Read the first and last money columns together and let the arithmetic do its own work. NVIDIA earned $72.9 billion and kept $60.9 billion of it as cash. Amazon reported $77.7 billion of net profit, and after paying for the warehouses, chips and data centres that produced it, $7.7 billion was left. That gap is not a flaw in anybody's accounting and it is not evidence of anything going wrong. It is the single most important fact about how each of these businesses should be valued, and it is available free, from the filings, before you have looked at a single multiple.

When is P/E enough?

When the gap is small and the balance sheet is quiet, which is more often than the professional consensus likes to admit.

NVIDIA spent $3.2 billion of capex in FY2025 to produce $130.5 billion of revenue, and carried $8.5 billion of long-term debt against $8.6 billion of cash — which is to say, none. With capital spending that light and no leverage to adjust for, earnings, EBITDA and free cash flow all move together, so the elaborate methods are correcting for problems the business does not have. A DCF adds forecasting apparatus and buys precision you have not earned. EV/EBITDA adjusts for a debt load that isn't there.

This is the unfashionable conclusion of the whole comparison, and it is worth stating plainly: for a capital-light, low-debt, profitable business, the crude method is the right one. P/E is fast, hard to fudge and comparable across a sector. The reason to reach for something more elaborate is never sophistication for its own sake. It is that the simple number has stopped describing the business.

When does P/E start lying to you?

When capital expenditure runs well ahead of depreciation — which is precisely what a company building capacity looks like from the outside.

Amazon's 2025 is the cleanest case. Net income of $77.7 billion, up from $59.2 billion the year before: on an earnings screen, a story visibly accelerating. Free cash flow over the same two years went the other way, from $32.9 billion to $7.7 billion, because capex went from $83.0 billion to $131.8 billion. Both facts are true and a P/E ratio can only see one of them. Oracle is the same mechanism at a smaller scale and a sharper angle, because it crossed zero — capex tripled in a single year and free cash flow went negative while reported earnings rose. If your entire read on Oracle that year came from an earnings multiple, you would not have known the company had stopped generating cash, and you would not have known that you did not know.

None of which proves either company is overvalued, or that the spending is wasteful. It proves that the earnings number on its own can no longer tell you.

Why EV/EBITDA is the worst method for a company like this

Here is where the conventional ranking — P/E is for beginners, EV/EBITDA is what professionals use — falls over completely.

EV/EBITDA deliberately strips depreciation out of the denominator, and the stated reason is a fair one. Depreciation is a non-cash accounting estimate, and different depreciation policies can make otherwise identical companies look different. But depreciation, for all its imprecision, is the only line in the income statement that gestures at the fact that assets wear out and have to be replaced. Take it out during a build-out and you have removed the one term standing in for the $131.8 billion Amazon actually spent. EV/EBITDA does not merely fail to capture capital intensity; it is designed not to see it, and it is at its most flattering exactly when the flattery is least deserved. Wherever capex is running at a large multiple of depreciation, it is the least informative of the three rather than the most.

So when does EV/EBITDA earn its keep?

When leverage is doing the work and the asset base is stable — which is a real and common situation, just not the one above.

Duke Energy ended 2025 with roughly $87 billion of debt against $51.8 billion of equity and $0.2 billion of cash. At that level of borrowing, net income is a residual left over after a large and variable interest bill, so P/E becomes a poor instrument for comparing Duke with a less-indebted peer: the multiple moves with the financing decision rather than with the business. EV/EBITDA puts the debt back into the price you are notionally paying and compares the operating asset directly, which is the job it was built for. Duke's free cash flow was also negative in 2025, at minus $1.7 billion, which for a regulated utility funding a rate-base expansion is a structural feature rather than a warning — and that is the point. The same negative number means something different at Duke than it does at Oracle, so the method should follow the reason for the gap rather than the sign of it.

A five-minute decision rule

From the last two annual filings, write down net income and free cash flow, then total debt and cash. Then choose.

That last row matters most for a beginner. A negative free cash flow figure is a question rather than a verdict, and the answer is almost never in the financial statements — it is in what management said about the spending on the call. Asked in 2026 about the returns on Amazon's capital programme, chief executive Andy Jassy said: "Of the AWS CapEx we intend to spend in 2026, much of which will be installed in future years, we have high confidence this will be monetized well as we already have customer commitments for a substantial portion of it." He is, of course, paid to say that; don't ask the barber whether you need a haircut. But the claim has a checkable edge, because "customer commitments for a substantial portion" either does or does not show up in later disclosure, and you can hold him to it. That is the useful form of an answer — a sentence a future quarter can contradict. Commitments like that one, and what happens to them, are what Promise Tracker keeps; the DCF calculator will let you test what growth and margins a price is already assuming.

How Inve helps you close the gap

The earnings-to-cash test tells you which question to ask. It does not answer it — for that you need to know what management said the spending was for, and whether that story has held.

That is where Inve earns its place in this workflow. Capital-expenditure guidance is extracted from every US call with the figure, the period, the executive who gave it and their exact words, and it is kept quarter after quarter — so the difference between Alphabet widening a range for an acquisition it had just closed and widening it because demand outran capacity is visible as two different sentences, not one indistinguishable raise. When a company's free cash flow goes negative, that record is the fastest way to find out whether it is a build-out being explained consistently or a story that keeps changing.

The DCF calculator closes the loop from the other end: instead of forecasting, solve for the growth the current price already assumes, and judge that against what the company has actually delivered.

Where this approach can mislead you

The strongest case against all of it is that a DCF is not "the honest method." It is the most forecast-dependent of the three, and in practice its answer is set almost entirely by two inputs nobody can know: the terminal growth rate and the discount rate. A model built on a real free-cash-flow gap and a fictional terminal value is not more rigorous than a P/E — it is a P/E with more decimal places and a worse audit trail. Where the gap is large, the honest move is usually a reverse DCF: ask what the current price already assumes, and judge whether that assumption is reasonable.

Nor is a wide earnings-to-cash gap evidence of a bad business. The entire argument for Amazon and Oracle is that today's capex is tomorrow's revenue, and sometimes it is exactly that. The gap tells you which questions to ask; it does not tell you the answer.

Two limits on the figures themselves. These are the consolidated annual filings as extracted for the US record on Inve, which does not carry a separate depreciation line for US companies or Oracle's long-term debt — so the EBITDA arithmetic above is described rather than computed, and Duke's leverage is the only one quoted in full. Operating leases and stock-based compensation are also unadjusted here, and both move free cash flow for precisely these companies. Anyone building a real model needs the 10-K, not a blog table.

Frequently asked questions

The owner's question

If you owned the whole business and could never sell it, the multiple would be irrelevant. You would care about one thing only: how much cash it handed you, and when. That is the question a DCF is trying to answer and the reason the method exists, even though its arithmetic is so often untrustworthy. The practical compromise is to let the earnings-to-cash gap decide the method, and then to keep asking the question a five-year owner would ask about the spending itself: is this capex buying an asset that will still be earning in 2031, or is it simply the cost of standing still? Nothing in the ratio will tell you. The call transcript sometimes will.

Related reading

• EV/EBITDA vs P/E: Which to Trust
• Reverse DCF: What the Market Already Expects
• How to Read an Earnings Call Transcript