Soaring Capital Expenditures in the Tech Sector: Good, Bad, or Ugly?

Some capex background

Capex seems naively like it should be an unalloyed good – it’s a company’s investment in its own future.

Still, countless academic and practitioner studies point to a different conclusion. As I put it recently in a Wall Street Journal article about capital spending and asset‑light companies, we know from 100 years of data that capex is on the average bad.^[4] What is bad about it? How can an investor benefit from this insight, if it is valid?

The principle of diminishing returns

Let’s go back to Economics 101, which teaches us about diminishing returns, and Corporate Finance 101, which asks us to compare the expected return on an investment to its cost of capital.

Diminishing returns means that the first dollar of investment in a business project will be very productive. Picture a would-be dairy farmer with no cows. The first cow turns the farm from a barren piece of land into a working micro-dairy farm. It’s the first step to having a real business.

Now, let’s skip ahead a few years and suppose that the farmer has kept adding cows until the land can bear no more, there are not enough customers to buy all that milk, and there are not enough workers to milk the cows. (Cow-milking robots are in their infancy – watch this space.) Clearly adding yet one more cow would be a foolish decision, with costs and no benefits.

Yet that is what some companies appear to be doing in the tech sector as they engage in a capital spending spree that has few historical parallels. The world’s biggest data center is 0.36 square miles. This is more than twice as big as the world’s largest factory, the Boeing Everett aircraft factory, at 0.15 square miles. The estimated capex spending of the world’s top nine cloud service providers (CSPs) in 2026 – that’s just one year – will top $830 billion, representing a one-year growth rate of 79%.^[5]

If an annual growth rate of 79% has been sustainable beyond a very short time period in any business at any time in history, it’s news to me. If 2026 data-center capex grew at 79% from this point forward, it would be larger than the world economy by 2035. So it won’t happen.

But this is not a rant against overspending on data centers specifically. It is an introduction to the more general idea that capex is a “priced factor” in the market. (For this purpose, we measure capex by its 5-year past growth rate.) A priced factor is finance jargon for saying that you can make money, on average over time and all other things being equal, by ranking stocks according to their recent past capex growth rates and buying those that rank in the lowest one or two quintiles. If you’re a long-short investor, you can make even more money by also shorting those in the highest one or two quintiles, as long as you pay close attention to risk control as all long-short investors must do.

Firms that currently show high capex growth rates are not limited to the Mag 7. They include (in decreasing cap size order): Eli Lilly, Advanced Micro Devices, Oracle, Applied Materials, GE Vernova, Amphenol, Boeing, H&R Block, Welltower, and Vertiv. These companies, most of which are familiar names, are all in the top quintile of stocks ranked by capex growth.

Business projects must earn their cost of capital, and then some

At this point Corporate Finance 101, in which we compare return on investment to the cost of capital, comes in. The criterion for whether to pursue a “project” (such as building a data center) is the project’s expected net present value (NPV), after taking into consideration the project’s cost of capital on an equal basis to all of the project’s other costs. One can rank all contemplated projects from highest to lowest by their NPV, and a rational business manager will pursue all those with an NPV greater than zero, while rejecting the rest.

It’s easy to see from this analysis that the projects ranking highest will have expected profits before capital costs far larger than their cost of capital, while those close to the cutoff point for being pursued will barely outearn their cost of capital. Below that, they will lose money. No matter how attractive a given business proposition may seem, there will be some projects pursuant to that proposition that are barely profitable enough to be worth doing, and many more that should be rejected. Competition ensures that this will be the case – a particularly attractive “hypergrowth” business will still have many more unprofitable potential projects than profitable ones.

Separating the wheat from the chaff in such circumstances is one of the reasons why top executives are worth their seemingly exorbitant paychecks.

Enough theory, let’s see the data

In 2003, my friend, advisor and collaborator Sheridan Titman, with two co-authors, studied the returns to firms with different asset growth rates.^[6] (“Asset,” in this context, means corporate assets carried on the balance sheet, reflecting cumulative past capex minus depreciation.) In this pioneering and influential study, they found a large “capex effect”:

Firms that substantially increase capital investments subsequently achieve negative benchmark-adjusted returns… These observations are consistent with the hypothesis that investors tend to underreact to the empire building implications of increased investment expenditures.

As shown in Exhibit 3, the authors sorted stocks into quintiles by their capex growth rates^[7] and tracked subsequent five-year performance. The quintile returns are adjusted for the market factor and three style factors, to avoid reporting results that are simply the size effect, value effect, or some other factor in disguise. The authors then formed a market-neutral “spread portfolio” consisting of, in each year, a long position in the bottom two quintiles of stocks sorted by capex growth and a short position in the top two quintiles:

Exhibit 3: Market- and Style-Adjusted Annual Returns on Portfolios of U.S. Stocks Ranked by Capex Growth, July 1973 to June 1996^[8]

A 16.8% per year “purified” spread between low and high capex-growth companies is very large, even in the exaggerated context of factor studies conducted with the benefit of hindsight. Such a result cannot be ignored.

Admittedly, because the period studied is July 1973 to June 1996, this is old research. Markets can change a lot over time, and an effect that “worked” between 30 and 50 years ago may have disappeared or flipped the other way, so confirming evidence from more recent times is needed before we’d invest based on it.

In addition, the capex effect observed over 1973-1996 had one long run of bad performance, 1980-1986, comprising almost one-third of the period studied. This seesaw effect is typical of stock price factors, including the well-known small-cap and value effects.

So, newer evidence is required. Exhibit 4 shows the results of a study we did covering territory similar to Exhibit 3, but with data from June 2006 through April 2026. For this more recent sample, we use the Barra Investment Quality factor, a composite measure based on asset growth, issuance growth, and capital expenditure growth^[9]. This factor is designed to capture the same underlying idea as our earlier capex-only analysis: firms that aggressively expand their asset base or issue large amounts of new equity tend, on average, to deliver weaker subsequent stock returns.

The effect is directionally the same as in the earlier study – high capex-growth stocks underperform – but is less dramatic. The bottom-to-top quintile spread portfolio would have earned 2.65% compounded annually, and the more diversified bottom-two-quintiles to top-two quintiles spread portfolio would have earned 1.73% annually.^[10]

Exhibit 4 ^[11]

How does the concern about capex apply to today’s markets?

We are in a period of extreme concentration in the U.S. equity markets. Exhibit 1 provided an impression of that, but we can use a more formal measure, the Herfindahl-Hirschman index (HHI), to ascertain how top-heavy the market is. The current level of the index, about 200 (mid-2025)^[12], is nearly at a record for modern times and is more than double the average over 1980-2019. You have to go back to the “Smokestack America” era of the 1950s through the 1970s to find comparable levels of concentration, and at that time, the top firms were not mostly in the same industry as they are today!

In 1970, the year M*A*S*H hit movie theaters and a baby named Gregg Fisher entered the world, the “Mag 7” were IBM, AT&T, General Motors, Eastman Kodak, Exxon, Sears Roebuck, and Texaco, heavy on manufacturing, but with big weights in telephony, oil, and retailing.

Exhibit 5: The “Mag 7” of the 1970s

In our view, markets are increasingly shaped not just by the usual suspects – fundamentals and sentiment – but by how capital is structurally allocated. Mega-cap companies such as Microsoft, Amazon, and Alphabet (Google) are collectively committing hundreds of billions of dollars to data centers, GPUs, and tech infrastructure.

This spending is not entirely motivated by the NPV (value added to the bottom line) calculation described earlier – it is defensive. The risk of underinvesting in AI is perceived to be higher than the risk of overinvesting, a situation that creates a one-way bias toward aggressive capital expenditure. “We have to make these investments,” the thinking goes, “because if we don't and our competitors do, we will face risks from which we may not recover.”

That way lies the path to overinvestment and low profitability or losses. Every bust in every industry in the past began with overinvestment – the dot-com bubble is an iconic example, but other instances go back to the South Sea Bubble in 1720. Human nature, including greed and fear, do not change much over time, if at all.

Behavioral considerations and their consequences

All asset “mispricings” (if the capex bubble is one) are in some sense behavioral, and having welcomed greed and fear into the conversation, let’s add some detail along those lines. Looking at both the period studied by Titman et al. and the more recent study covering the last 20 years, we can identify some reasons why high capex growth rates have been bad for subsequent stock returns:

• The capital allocation choices of leaders after huge stock price increases have not been great — The leaders tend to repeat what worked in the recent past and expect the repeated behavior to generate the kinds of profits it generated the first time.
• Empire building — Executives are rewarded for managing a lot of people and assets, not profits that might be earned far in the future.
• Industry-wide high capex leads to higher production, more supply, depressed prices and thus lower profits for everyone — This is just Econ 101.
• High capex means that firms distribute less cash to investors, which might depress valuations — Investors might have a more profitable use for the cash than the company does – including buying stock in other companies.
• If the AI capex cycle turns out to be like what happened to electric vehicle auto firms in China, it is problematic — Firms spent freely at the beginning of the cycle, where prospects look good, and then were mired in competition that caused everyone's profits to decline.

Summing up, the risk is not whether AI itself fails as a concept or product, but whether returns on invested capital fall short of expectations during the buildout phase. We cannot invest directly in the social and economic improvements that are linked to an innovation – we buy companies “stocks”, so we need to look at sales, earnings, capex, labor costs, sentiment, the impact of competition, and the prices of the stocks, not just the remarkable achievements of AI entrepreneurs and engineers.

Overspending on the average ultimately underperforms

Today’s corporate leaders – in the Mag 7 and elsewhere – have risen to the top during a long period when intangible assets such as patents, copyrights, and labor contracts ruled. Now they’re investing trillions in buildings, like the industrial titans of old. Do they know how? Do they know when to stop? Are their support teams, tasked with forecasting profits and estimating the cost of capital, up to speed on this new (old) type of investment?

We are concerned that, in enough cases to make a difference, they aren’t. History bears out this fear. Every business boom from barge canals and railroads in the nineteenth century to the internet and AI today has resulted in overinvestment and, eventually, poor market returns. Projecting past returns forward indefinitely is consistent with human nature, or at least the nature of those who have succeeded in business; understanding the limits to growth and responding in anticipation of them is not.

Invoking the marginal thinking that is at the heart of all economics, we can, therefore, expect returns in the real economy below the cost of capital for the last dollar (or hundreds of billions of dollars?) of capex in the tech sector. Whether this means negative or below-market returns in the stock market is for analysts to determine, taking into account sentiment and many other factors, but for stock prices eventually to follow the economics is usually the way to bet.

Modern capex: Are companies overpaying for people?

The possible bubble in tech-related capex may eventually be overshadowed by a bubble in people. By this, we mean very high salaries and bonuses paid to employees perceived as having unique talents. (This is how Shohei Ohtani got a $700 million contract to play baseball – Sure he is good …but that’s a lot of money.)

While capex goes on the balance sheet as an asset, human capital does not; the whole paycheck is expensed in the year that it’s paid. This makes allocation between physical capex and human capital investments more difficult. Executives who are having trouble finding the point of diminishing returns with capex may have even more difficulty finding it with high-priced hires. After all, one data center looks like another (all of them are ugly), but prestige employees serve as ambassadors for the firm as well as being highly productive internally. And they are usually not ugly.

In future work, we will expand on the idea that companies inclined to overinvest because of past success may be doing so with hiring as well as building. For now, it suffices to express the concern and save detailed commentary about it for later.

Some ideas for portfolio management

How should we, as investors, react to these realities?

One possible implication is to evaluate exposure to companies taking the most capital spending risk, including concentration within capitalization-weighted indexes such as the S&P 500. Small companies, growth at a reasonable price, international investing. But these are ground-level, simple ideas. Portfolio management is complicated. Looking carefully at each company, we need to be aware that excessive capital spending is taking place at many companies – but not all.

Because of the incredible concentration in today’s market, your portfolio is probably concentrated in the top 10 or 15 or 20 names by market cap. This has happened to everybody. Looking back at 100 or more years of history, this situation doesn't end well for companies at the vanguard of a mammoth capital spending boom. This may support holding larger weights in companies that are more careful with spending (on both physical and human capital), and that are not seduced by competitive pressures to engage in ferocious bidding wars for stuff and people. We know how bidding wars end: the bidder with the biggest overestimate of the value of the item being auctioned gets the item. It’s called the “winner’s curse.”

Conclusion

The AI revolution is real. You can’t put the genie back in the bottle.

However, that does not mean you should buy large positions in AI-related stocks and hold them forever. Historically, periods of extraordinary investment enthusiasm have produced poor subsequent returns when too much capital chases the same opportunity.

The question facing us as investors is not whether AI is good or bad, or how it will change the world, but which firms will earn returns above their cost of capital as the reward for each unit of capital spending inevitably declines. Our job as portfolio managers is to answer that question and diversify our holdings accordingly, away from companies spending too much on capex and towards those with better prospects.

A message from Advisor Perspectives and VettaFi: Discover something new! Click here to register for our upcoming webcasts.

Footnotes & Sources:

^[1] Expressed as a percentage of GDP in the peak year for each of these projects.

^[2] Source: The Kobeissi Letter; underlying sources include Manhattan District History, BEA, The Planetary Society, Eno Center for Transportation, San Francisco Fed, Hoover Archives, Baruch, GoldenGate.org, The New York Times, and J.P. Morgan Asset Management, 2025.

^[3] Source: Quent Capital Research; Bloomberg

^[4] Spencer Jakab, “Leading Stocks Are Losing Their Low-Asset Edge,” The Wall Street Journal, May 11, 2026.

^[5] TrendForce, "North American AI Data Center Expansion Drives 2026 CapEx," Yahoo Finance, May 6, 2026.

^[6] Titman, Sheridan, K.C. John Wei, and Feixue Xie. 2003. “Capital Investments and Stock Returns.” NBER Working Paper 9951, http://www.nber.org/papers/w9951.

^[7] In this case, not the 5-year growth rate, but a more complex measure of capex growth that compares capex in the year prior to the year of portfolio formation to capex in the previous three years, where each year’s capex has been scale by the company’s sales in that year.

^[8] Source: Titman, Wei, and Xie (2003), table 1, panel A.

^[9] MSCI Barra’s Investment Quality factor measures company investment activity using asset growth, issuance growth, and capital expenditure growth. Here, those inputs are combined as 0.4 × Total Assets Growth + 0.4 × Issuance Growth + 0.2 × Capital Expenditure Growth and used as a proxy for investment growth.

^[10] Other studies have corroborated these findings. See, for example, Cooper, Gulen and Schill (2008), who showed that asset growth rates that exceeded an industry’s norm often led to subnormal equity returns.

^[11] These three metrics are combined into one composite factor: 40% Total Assets Growth Rate, 40% Issuance Growth, and 20% Capital Expenditure Growth.

^[12] The Herfindahl-Hirschman Index, developed independently by Orris Herfindahl (1950) and Albert Hirschman (1945), measures industry concentration by summing the squared market shares of all firms in a market.

The intuition behind the index, which is expressed on a scale from 0 to 10,000, is that if all 500 stocks in the S&P had the same capitalization, the HHI would be (10,000/500)=20, and if the entire market were concentrated in one stock, the HHI would be (10,000/1)=10,000. On such a scale, 200 doesn’t sound very concentrated but the number should be compared to its own history, not to its hypothetical maximum and minimum values.

Disclosures

This material is for informational and educational purposes only and does not constitute an offer to sell or a solicitation of an offer to buy any security, investment product, or investment advisory service. The views expressed reflect Quent Capital’s opinions as of the date of publication and are subject to change without notice.This material should not be construed as personalized investment advice or as a recommendation to buy, sell, hold, or avoid any security, sector, index, strategy, or investment product. References to specific companies, securities, indexes, sectors, or market segments are for illustrative purposes only and should not be interpreted as a recommendation or endorsement.Past performance is not indicative of future results. Investing involves risk, including the possible loss of principal. There can be no assurance that any investment strategy, factor, methodology, or investment program discussed herein will be profitable or achieve its objectives.Certain statements in this material are forward-looking and are based on current views, assumptions, estimates, and expectations. Actual results, performance, market conditions, or events may differ materially from those expressed or implied. No representation is made that any forecast, estimate, projection, or opinion will be realized.Certain analyses in this material include academic research, historical factor analysis, and/or hypothetical or backtested results. Hypothetical and backtested results have inherent limitations, are prepared with the benefit of hindsight, do not reflect actual trading or actual client performance, and may not reflect the impact of advisory fees, transaction costs, shorting costs, borrowing costs, taxes, market impact, liquidity constraints, or other expenses that would reduce returns. Actual results may differ materially.Index references, including references to the S&P 500, are provided for informational purposes only. Indexes are unmanaged, do not reflect the deduction of fees or expenses, and cannot be invested in directly.Quent Capital, LLC is an SEC-registered investment adviser. Registration does not imply any level of skill or training. For additional information regarding Quent Capital’s services, fees, and conflicts of interest, please review Quent Capital’s Form ADV Part 2A and related disclosures available on the firm’s website.

© Quent Capital

Read more commentaries by Quent Capital