Big Tech Is Spending $700 Billion on AI. Is That a Cost or an Investment?

The Magnificent 7 will spend more than $700 billion on AI this year — roughly 70% more than last year. That number is what's rattling the market, and I think the question is being framed wrong. What matters isn't how much they're spending. It's what kind of money they're spending.

Here's my answer up front: a rise in capital expenditures tells you nothing on its own. Money a company is forced to spend just to stay where it is (maintenance capex) is a cost. Money a company chooses to spend to earn bigger profits later (growth capex) is an investment. Most of Big Tech's AI spending is the second kind — but an investment isn't automatically a good investment.

The two kinds of capex, straight from Warren Buffett

This is one of the most important ideas in investing, and it doesn't get taught nearly enough.

Maintenance capex is money a company has no choice but to spend to keep the lights on and stay exactly where it is. Repairing machines, keeping the building operational, fixing the roof. It's the cost of not going backwards, and you cannot escape it.

Growth capex is money a company chooses to spend to expand and generate bigger profits down the road. Short term it looks like a cost. In substance it's a long-term investment.

Both land on the exact same line of the cash flow statement. Accounting will not separate them for you. That's the investor's job.

This is precisely the point Bill Ackman has been pounding the table on while buying Microsoft, Meta, and Amazon: you should not punish a great company for reinvesting in itself to earn larger profits later. I think he's right. That's the entire purpose of a business.

Why net income and free cash flow diverge — the Meta case

Meta makes the distortion visible.

Meta trades at a P/E of about 24, but a price-to-free-cash-flow multiple of 35. Why the gap? Over long stretches, net income and free cash flow are supposed to converge.

The answer is capex. Free cash flow is cash from operations minus capital expenditures. Look at Meta's capex path since 2020: roughly $15B → $19B → $31B → $27B → $37B → $70B → $75B. That's 5x in six years.

Net income, meanwhile, doesn't absorb that hit immediately. It gets recognized through depreciation over seven to ten years — sometimes five. So today's net income sits well above today's free cash flow.

Normally, net income running above free cash flow is a red flag for me. It can mean accounting profit without real cash behind it. But in Meta's case the cause is unambiguous: cash is going out the door to build data centers, not because earnings quality deteriorated.

The balance sheet supports the spending, too. Meta's market cap is about $1.69 trillion against an enterprise value roughly $70 billion higher — so about $70 billion of net debt. That sounds like a lot until you compare it to last year's $48 billion of free cash flow: under two years' worth. Gross margin sits at 82%, meaning 82 cents of every incremental revenue dollar survives to profit before overhead and taxes.

Amazon's free cash flow is outright negative

The more extreme case is Amazon.

Last year Amazon posted $91 billion in net income and negative $2.5 billion in free cash flow. That's how large the capex is. Amazon simply doesn't care about short-term cash flow. It optimizes for the future and for making life easier for customers, and that strategy has paid off enormously.

It's also clear where this approach hurts the numbers: return on invested capital. When capital takes years to convert, measured returns look depressed. Amazon's P/E is 30, its five-year average P/E is 61, its price-to-free-cash-flow is negative, and its five-year average price-to-free-cash-flow is a bizarre 437 — purely because free cash flow is so low.

I won't tell you to ignore those figures entirely. But with Amazon, the real story is the ability to reinvest. Gross margin is 51%, and net margin has climbed from a 6.5% ten-year average to 7% over five years to 12% most recently, driven by AWS — the largest cloud business in the world, with enormous margins. I walked through this structure in more detail in Amazon's negative free cash flow and capex.

Worth adding: this growth isn't bought. Revenue grew 21% annually over ten years, 12% over five, 12% over three, while five years of acquisitions totaled $27 billion — about 1% of market cap, roughly 0.2% per year. That's organic.

So what should you actually judge this on?

Whether capex is investment or waste comes down to one thing: is the money earning a high rate of return?

The metric I use is return on invested capital. Mid-teens and above, I consider that a high-quality business. Low double digits is still fine if the trend is improving.

This is where the seven separate. Meta clears the bar comfortably. Nvidia is in a league of its own at 40–45%. Amazon's number is currently suppressed, and whether you're willing to accept that is the entire investment decision.

The other thing I check is reversibility. When Meta collapsed in 2022, the market's fear was the metaverse. My view at the time was simple: if the metaverse fails, they save $10–12 billion a year in free cash flow and the company is fine. I bought during that period, and got lucky enough to pick up shares near the $88 low.

Does the same logic apply to AI? In theory, yes — if the returns don't show up, they can stop spending and revert to cash generation. But by then a lot of money would already be spent. That's the genuine risk, and I don't want to hand-wave it away.

Where I land

The market is currently punishing capex increases indiscriminately. That's lazy. But the opposite claim — "it's reinvestment, so it's all fine" — is equally lazy.

Practically, three checks are enough. One: can the balance sheet carry the spending (net debt versus free cash flow)? Two: is there a track record of converting invested capital into high returns (ROIC)? Three: does management retain the option to pull back if results don't arrive?

For a company that passes all three, shrinking free cash flow is a reason to buy, not sell. If even one check fails, the compressed valuation isn't a discount — it's a warning. The industry-wide picture is laid out in the mega-cap AI capex arms race.

FAQ

Q: Should I always avoid companies with falling free cash flow? A: No — look at the cause. If free cash flow is falling because operating cash flow itself is shrinking, the business is deteriorating. If operating cash flow is rising and free cash flow is falling because of capex, that's a completely different situation. Meta and Amazon are the second case.

Q: Can I separate maintenance from growth capex in the financials? A: Not directly — companies report capex as a single line. The practical workaround is to treat depreciation as a rough proxy for maintenance capex and everything above it as growth capex. It's imprecise, but it's enough to point you in the right direction.

Q: Is net income above free cash flow always a bad sign? A: Usually it deserves scrutiny. But when the cause is heavy capital investment, the picture changes. Capex hits cash immediately while flowing through the income statement over seven to ten years via depreciation, so during a build-out phase net income structurally runs ahead of free cash flow.

Q: What happens if that $700 billion never earns a return? A: Stop the spending and free cash flow recovers quickly. The real damage is the capital already sunk, plus whatever valuation you paid during the build-out. That's why I weigh reversibility and purchase price together.