Every quarter, when the technology giants report, attention drifts to a single item — capital expenditures.
Microsoft, Alphabet, Meta, and Amazon are spending hundreds of billions of dollars on data centers, custom chips, and the power to run them. Sums so large they are genuinely hard to picture. Over the last three years the cumulative figure, just from the Mag7, is over one trillion dollars. That's about 17% of US GDP in 1959, the year the silicon chip was invented (in 2026 dollars). And with each new figure comes the same question, asked with increasing unease. Are these companies quietly spending their way out of the very economics that made them worth owning?
It's a fair worry. Returns on capital represent the North Star of how we judge a business, and the mechanics of a spending spree work against it until it is proven out whether the spending was worthwhile. A dollar of capex hits the books the moment it's committed, while the profit it's meant to produce arrives some time later, if at all. In between, returns on capital should sag.
So rather than take this worry on faith, we did the rest of the arithmetic. For each of the seven, we measured what the business actually earns on every dollar of capital it puts to work (per our adjusted ROIC methodology), quarter after quarter, and set it against that company's own weighted average cost of capital (WACC) — the return its investors could earn elsewhere for taking a similar risk. The gap between those two numbers is the plainest measure there is of whether all this spending is creating value or quietly burning it. You make more than your WACC, and you get to keep the excess. You make less, and you are shrinking. The time spent in either of these zones matters, naturally, but that is the line.
And in the case of the Mag7, the answer, it turns out, isn't one story. It's three.
Where the seven stand today
| Company | Return on capital | Cost of capital | Economic-profit spread |
|---|---|---|---|
| NVIDIA | 83.9% | 12.3% | +71.6 pts |
| Apple | 47.1% | 9.2% | +38.0 pts |
| Microsoft | 26.9% | 9.0% | +17.8 pts |
| Alphabet | 24.6% | 9.3% | +15.3 pts |
| Meta Platforms | 24.0% | 9.5% | +14.5 pts |
| Amazon | 11.8% | 9.9% | +2.0 pts |
| Tesla | 8.5% | 12.3% | −3.7 pts |
Five of the seven clear their hurdle with room to spare. One sits just above it. One has slipped below. And every company on this list has expanded its invested-capital base enormously over four years, most by well over half, several by more than double. What separates them is whether the earnings followed the capital.
The question is never how much they spend nominally, that number doesn't matter, it's what they earn on it.
The Good — The capital is compounding
The two most-watched names in the group aren't a capex-drag story at all, but for opposite reasons. One sells the picks and shovels. The other barely digs.
+18.6 → +71.6 pts over four years
+33.9 → +38.0 pts over four years
Nvidia is the buildout's supplier, not its payer. Its capital base has grown several times over, but the profit it earns on that base has grown faster still. So its return on capital has multiplied rather than thinned. However, the path wasn't a straight line. The spread nearly closed in early 2023, in the depths of the gaming and crypto downturn, before the AI cycle sent it vertical. When the rest of the world is buying your chips to build their data centers, the capex wave becomes your revenue line, not your cost. Notably, even before AI, Nvidia has always ranked high in our approach, clearing the high quality bar as far back as 2001. Whether that continues past the initial AI wave is the real question to ask.
Apple is the mirror image. It has chosen to rent the AI era rather than build it — leaning on partners and the cloud instead of capitalizing its own campuses. Its invested capital has barely moved in four years, and the return on that capital has quietly climbed, with a distinct step up in the past several quarters. Neither company's economics are being squeezed by AI spending. If anything, both are on the receiving end of it.
The Deteriorating — Compressing, but a long way above the line
Here the market's instinct is right, as far as it goes. These are the companies writing the checks, and you can see it in the numbers — just not the way the bears mean it.
+36.0 → +17.8 pts over four years
+18.5 → +14.5 pts over four years
+18.2 → +15.3 pts over four years
Microsoft, Meta, and Alphabet are the archetypal builders, and their returns on capital have compressed. Microsoft's invested capital has grown more than two-and-a-half times in four years, and its return on that capital has fallen from the low forties to the high twenties. Meta's arc is more dramatic than a single number suggests. Its spread dropped hard through 2023's "year of efficiency," then recovered while Alphabet's has been the steadiest in the entire group. In each case the spending is running ahead of the profit it has so far produced, and the cushion above the cost of capital has narrowed. On that, the skeptics are correct.
But narrowed from what? Even after the compression, all three still earn roughly two-and-a-half to three times their cost of capital. A business whose return falls from 42% to 27% against a cost of capital near 9% has not stopped creating value — it has gone from extraordinary to merely excellent. The economic profit these firms throw off on every dollar of capital is still among the widest in the market. The distance still separating them from their hurdle is the whole point. We cannot say these are suddenly bad businesses. But this, of course, says nothing about the price worth paying.
A return that falls from 42% to 27% against a 9% cost of capital hasn't stopped creating value. It has gone from extraordinary to merely excellent.
The Bad — Where the capital hasn't paid off
The honest test of any framework is whether it flatters the whole group or actually discriminates within it. This one discriminates.
+0.2 → +2.0 pts over four years
+21.8 → −3.7 pts over four years
Amazon has run the same capital-heavy playbook, and for years it earned roughly its hurdle and no more. Its spread even dipped below zero in 2022. It climbed clearly above the line in late 2023 and looked to be building a real cushion. But the most recent quarters cut the other way. As its invested-capital base has surged past half a trillion dollars, the spread has thinned back toward its hurdle. Amazon is creating value again, but the current capex wave is pressing on that margin in real time, not relieving it. The verdict here is genuinely open.
Tesla is the cautionary tale the other six aren't. It ran an aggressive buildout while the return on that capital collapsed from the low thirties to single digits. Against a cost of capital its own volatility keeps elevated, it now sits below the line. For the moment, its capital spending is consuming value rather than creating it. Same capex-heavy pattern as the builders; the opposite answer. If it truly is a technology company, perhaps a reversal is in order. If it is in fact a car company, this is not out of the ordinary.
The four-year arc, at a glance
| Company | Return on capital (2022 → now) | Invested capital (2022 → now) |
|---|---|---|
| NVIDIA | 28% → 84% | $42B → $167B (+295%) |
| Apple | 41% → 47% | $268B → $281B (+5%) |
| Microsoft | 42% → 27% | $178B → $468B (+162%) |
| Alphabet | 25% → 25% | $299B → $487B (+63%) |
| Meta Platforms | 26% → 24% | $173B → $363B (+110%) |
| Amazon | 7% → 12% | $261B → $535B (+105%) |
| Tesla | 32% → 9% | $32B → $78B (+143%) |
The other side of the ledger
Everything up to this point has measured the business. But an owner doesn't hold a business — they hold a share of one, and those aren't quite the same thing. A company can earn a magnificent return on its capital and still shortchange its shareholders, if the value it creates is spread across an ever-larger pile of shares. Our return-on-capital work is deliberately silent on this; it weighs the enterprise, not the share count. So it's worth asking the second question out loud: over these same four years, did each of these companies concentrate its owners' claim on the business, or thin it out?
The answer lines up with the three stories almost too neatly. The five firms that clear their cost-of-capital hurdle with room to spare have, every one of them, shrunk their share counts — Apple by nearly a tenth, Alphabet and Meta by six to eight percent, Nvidia and Microsoft by less but still in the right direction. Whatever value those businesses create is now divided among fewer shares, not more. The two that struggle to clear the hurdle are the same two moving the other way: Amazon has let its share count drift up around six percent, buying back nothing while paying its people some twenty billion dollars a year in stock, and Tesla's count has swelled more than twenty percent, with a sharp jump in late 2025. The companies earning the least on their capital are also handing their existing owners the thinnest slice of it. Capital allocation is its own mark of quality, and here it runs along exactly the same line as the returns.
One caveat keeps this from being a victory lap. Even among the reducers, the discipline has softened lately. Alphabet's buybacks have roughly halved over the past year while its stock-based pay has climbed toward twenty-six billion dollars, and Meta's share count has quietly ticked back up over its last few quarters for the same reason. Net dilution is still negative for both — but the offset that made it so is thinning. A shrinking share count, it's worth remembering, is a choice a company renews every quarter, not a law of nature.
What it means for investors
The capex-panic framing asks the wrong question. Spending isn't the risk. Spending that fails to earn is.
On the measure that has always separated durable compounders from capital-destroyers — return on invested capital against the cost of that capital — most of the Magnificent Seven remain, even at the peak of the AI build cycle, among the most productive deployers of capital anywhere. But the group is no longer one thing. The dispersion inside it is wider than the label admits, and it is still widening as the spending continues.
For a long-term owner, that dispersion isn't a warning. It's the opportunity. The returns tell you which of these franchises is buying growth and which is only buying capacity; the share count tells you how much of that growth each owner actually gets to keep.
About this analysis
Return on invested capital and cost of capital are New West Capital estimates, derived from public company filings and our own research framework — measured on a trailing-twelve-month, operating basis, with capital adjusted for leases and other proprietary adjustments. The economic-profit spread is the difference between the two, in percentage points. Share counts, stock-based compensation, buybacks, and dividends are drawn from the companies' SEC filings, with shares stated across all classes and normalized to a current post-split basis; capital-return figures are trailing-twelve-month. Companies referenced: Nvidia, Apple, Microsoft, Meta Platforms, Alphabet, Amazon, and Tesla. Figures run through the first calendar quarter of 2026 (Nvidia through its April fiscal quarter).
Important disclosures
This material is provided for informational and educational purposes only. It does not constitute investment advice, an offer, or a solicitation to buy or sell any security, and should not be relied upon as the basis for any investment decision. Any securities mentioned are for illustration only and are not recommendations. Estimates and analyses reflect judgments as of the date shown and are subject to change without notice; forward-looking statements are inherently uncertain. Past performance and historical returns on capital are not indicative of future results. New West Capital and its clients may hold positions in the securities referenced. Recipients should consult their own advisers before acting on any information herein.