Maintenance vs Growth CapEx: Buffett's Real FCF Number

When you read "$3.2 billion in capex," you have no idea whether the company is spending to stay alive or to expand. Until you split that number into maintenance and growth capex, free cash flow is partly fiction.

What is maintenance capex versus growth capex?

Maintenance capex is the cash a company must spend each year just to keep its existing assets producing what they produced last year — replacing worn-out forklifts, refreshing store interiors, swapping out servers at the end of their useful life. Growth capex is the cash it spends to expand capacity beyond that baseline — opening a new factory, building a new fulfillment center, adding fabs.

Standard GAAP financial statements lump both into a single line: total capital expenditures. That is convenient for accountants and disastrous for investors — because the two categories produce wildly different returns on capital. Maintenance capex earns roughly zero incremental return (the asset just keeps running). Growth capex either creates real value or destroys it, depending on the project's ROIC versus the cost of capital.

The split matters because true free cash flow — what an owner could pull out of the business without shrinking it — is operating cash flow minus only maintenance capex, not total capex. Reported FCF deducts everything and tends to understate the cash an established business actually throws off.

Why did Buffett invent "owner earnings" around this idea?

Because reported FCF can be off by roughly 30-50% on capital-intensive companies. In his 1986 letter to Berkshire shareholders, Buffett defined owner earnings as net income plus depreciation and amortization, minus the average annual maintenance capex needed to maintain unit volume and competitive position.

The implication: a company growing roughly 12% per year with $1 billion of total capex, of which $400 million is maintenance and $600 million is growth, is wildly different from a company with $1 billion of capex that is all maintenance. The first is reinvesting two-thirds of its capex at incremental returns; the second is just running to stand still.

Buffett uses this framework most aggressively in his published work on Coca-Cola (KO), where reported FCF and owner earnings track closely because maintenance is a small fraction of total capex. For comparison, look at the math on a company like Intel (INTC), where fab maintenance alone runs in the range of roughly $10-12 billion per year — owner earnings tells a very different story than reported FCF.

How do you estimate maintenance capex when companies don't disclose it?

Three methods, in order of how much you should trust them.

Method 1: depreciation expense as a proxy. Over a full economic cycle, replacement capex should roughly equal depreciation — because depreciation is the accountant's estimate of how much of your asset base wears out each year. This is the quickest method and works for stable, mature companies. It breaks when depreciable lives are wrong (fab equipment is depreciated over five years but lasts ten) or when inflation runs ahead of the depreciation schedule.

Method 2: management disclosure or guidance. Some companies break out maintenance versus growth in 10-K supplements or investor day decks. Costco (COST) does this informally on calls — refresh capex versus new-warehouse capex. Home Depot (HD) breaks out new-store capex separately. When management gives you the number, use it.

Method 3: bottom-up by asset class. Multiply the gross PP&E balance by the inverse of the average useful life, weighted by category. This is the most accurate but the most work. Most investors do not need this level of precision. The 80/20 rule here is: use depreciation as your maintenance proxy unless you have a specific reason to override it.

Where does this framework matter most?

In capital-intensive cyclicals and in growth companies whose capex is genuinely transformational. Two examples.

ExxonMobil (XOM) reports total capex around $25 billion per year. Roughly $15-18 billion of that is maintenance — keeping wells producing, refining capacity online, and downstream chemicals running. The remaining roughly $7-10 billion is growth capex in Permian basin expansion, LNG capacity, and low-carbon initiatives. Owner earnings for XOM is closer to reported earnings plus that ~$15-18 billion replacement gap, which makes the dividend coverage look very different than the headline payout ratio suggests.

Microsoft (MSFT) is the other extreme. Total capex has surged to a range of roughly $90-100 billion as AI data centers ramp. The growth-capex portion is almost the entire envelope — historical maintenance ran closer to ~$15-20 billion in the pre-AI era. Investors who deduct all $90+ billion as if it were maintenance dramatically understate the company's owner earnings — they are valuing MSFT on a model the business is no longer running.

A side-by-side comparison

Estimates are rough — they vary year to year and depend on which method you use. The relative ranking matters more than the exact percentage.

Common mistakes investors make

The first is double-counting growth in your DCF. If you forecast roughly 12% revenue growth and then use total capex (including growth capex) as a deduction, you have already paid for the growth — counting it again as a cash outflow understates the value of the business.

The second is mistaking transitions. When a low-capex business pivots to a capex-heavy strategy (META building AI infrastructure, GOOGL adding TPU capacity), the old maintenance ratio breaks immediately — and applying the historical ratio gives you a wildly inflated owner-earnings figure.

The third is ignoring obsolescence in fast-moving categories. Semiconductor fabs are technically replaceable over five years on the schedule, but the technology cycle is faster than the depreciation schedule. Real maintenance capex for leading-edge foundries is structurally higher than its depreciation expense suggests, because each replacement is buying more capability than the prior asset had.

When does the framework break?

When growth capex creates obsolescence risk faster than depreciation captures it. AI hardware is the current example: the H100-to-Blackwell transition shortened the useful life of cloud GPUs well below their five-year depreciation schedule. In that regime, depreciation underestimates true maintenance capex, and owner earnings calculated the standard way is overstated.

Critics also argue the maintenance-versus-growth split is impossible to do consistently across companies, which makes it a tool for narrative more than analysis. There is real force to that critique — if you cannot estimate the number with discipline, do not use it. For stable mature businesses (consumer staples, regulated utilities, established retail), the split is reasonable enough to add meaningful information to a valuation model. For deep cyclicals at inflection points, treat any maintenance-capex figure as a range, not a point estimate.

For a deeper walk-through of how to build a DCF that respects this split, see our fundamental analysis guide. For more on how Buffett uses owner earnings across his portfolio, see the super investors section.

Bottom line

Total capex is one number, but it contains two very different stories. Splitting it into maintenance and growth changes your free cash flow estimate, your ROIC estimate, and ultimately your fair value estimate — sometimes by roughly 30% or more. Use the split where it matters; do not pretend it works everywhere.

Ready to analyze these stocks yourself? Search any ticker on MainRatios to see valuations from 6 legendary investors - free.