Physical Address
304 North Cardinal St.
Dorchester Center, MA 02124
For more than a decade, one of the most persistent complaints about modern capitalism went like this. Large companies were earning record profits and, instead of putting that money to work in factories, research and people, they were handing it back to shareholders through share buybacks. Buybacks lifted earnings per share, flattered executive stock options and starved the real economy of investment. Fix the incentives, the critics argued, and companies would reinvest again.
In 2026 the largest companies in the world did exactly that. They stopped buying back their own stock and poured almost everything into building. By the logic of the critics, this should be the moment shareholders and economists alike applaud.
Instead, the market has spent the autumn punishing them for it, and on 5 November the selling turned into a full correction in technology stocks. That contradiction is worth examining closely, because it gets at a question investors rarely ask directly: when a company reinvests, how do you know it is creating value rather than burning it?
The Decade the Critics Hated
The scale of the buyback era is hard to overstate. In the twelve months to September 2025, S&P 500 companies repurchased a record $1.02 trillion of their own shares, only the second time the total had crossed the trillion dollar mark. Alphabet alone spent about $55.8 billion over that period and Meta about $44.2 billion, placing both among the largest buyers in the index.
The intellectual case against this behaviour had been building for years. The economist William Lazonick framed it as a shift from a “retain and reinvest” model of corporate management to one of “downsize and distribute”, an argument that won the McKinsey Award for his Harvard Business Review essay “Profits Without Prosperity”. Other economists, including Mariana Mazzucato, went further and asked whether markets had lost the ability to distinguish value creation from value extraction at all, since a rising share price was increasingly treated as proof of value in itself.
Whatever you think of those arguments, they had a clear policy implication. Companies should spend less on their own stock and more on productive investment.
2026: Retain and Reinvest, at Full Throttle
Now look at what happened. Alphabet’s second quarter cash flow statement shows share repurchases of zero, compared with about $13.2 billion a year earlier. In the same quarter it raised about $49.6 billion in net proceeds from new common and mandatory convertible preferred stock, set up a separate $40 billion at-the-market programme, and issued a further $20.3 billion of senior notes. Quarterly capital spending reached about $44.9 billion.
Meta followed the same path. It bought back no stock at all in the first nine months of 2026, after $26.2 billion in the same period of 2025, then launched about $35 billion of new stock sales in late October with a further programme of up to $60 billion planned. Third quarter capital spending rose 80% from a year earlier and free cash flow turned slightly negative. Amazon, which has not repurchased stock in years, is funding its own buildout increasingly with debt.
Goldman Sachs analysts, as reported by MarketWatch in May, estimated that the largest AI spenders were directing about 20% of their spending to buybacks and dividends this year, compared with an average of 34% between 2017 and 2022, and that their capex growth was on course to absorb essentially all of their operating cash flow. Across the group of Amazon, Alphabet, Meta, Microsoft and SpaceX, total 2026 capital spending is now expected to land at between $803 billion and $828 billion, with executives signalling more in 2027.
This is retain and reinvest on a scale no critic of buybacks ever imagined. The companies are not merely retaining profits. They are raising outside capital, through debt and new shares, in order to invest more than they earn.
Why the Market Isn’t Applauding
The reaction has been brutal. Alphabet suffered its worst trading day since listing after its third quarter results, and Meta’s share sales were met with a falling stock price and a negative outlook from S&P Global Ratings, which questioned whether the company could monetise its AI spending as effectively as its peers.
Then came 5 November. The Nasdaq Composite fell 5.5% in a single session, its worst day since April 2025, pushing the index into correction territory, after the US midterm results raised the prospect that rising electricity prices and local opposition would make the data centre buildout slower and more expensive. The heaviest losses landed on the suppliers. The Philadelphia Semiconductor Index dropped more than 12%, Nvidia lost over 9% and several data centre operators fell by more than 14%. The big spenders fared far better, with Meta down about 4% and Alphabet and Amazon about 2%. More telling still, all three traded higher before the opening bell, because a slower buildout was read as good news for them. For most of this cycle, every increase in capital spending was rewarded. On this occasion, the prospect of having to spend less was briefly treated as a gift to shareholders.
The reason is simple once you separate two ideas the buyback debate often blurred together. Investment is not the same thing as value creation. Value is only created when the return on the money invested exceeds what that money costs. A company that spends a dollar to build an asset that earns eight cents a year, when its investors could have earned ten cents elsewhere, has destroyed value, however impressive the asset looks.
Buybacks, for all their flaws, have one virtue: they return cash to owners who can redeploy it. When a company with no attractive projects buys back stock, it is at least admitting that shareholders might find better uses for the money. When a company issues shares to fund spending, it is doing the reverse. It is asking existing owners to accept dilution today in exchange for a promise about returns tomorrow. A share issue is, quite literally, a buyback run backwards.
So the market is not rejecting reinvestment as such. It is questioning the expected return on this particular reinvestment, at this particular price, funded in this particular way.
What History Says About Spending Booms
Investors have good reason to be cautious, because the historical record on aggregate spending surges is unkind to shareholders.
The idea is known as the capital cycle. When an industry is highly profitable, capital floods in, capacity expands faster than demand, returns fall and weaker players fail, until the survivors eventually recover. A CFA Institute review of the investment firm Marathon’s work on the subject describes a persistent link between heavy investment and weak subsequent returns, across companies, sectors and even whole countries. Its examples include Global Crossing, which overbuilt fibre networks in the late 1990s and was sold for a fraction of their cost, and the radio, automobile and aircraft booms of the 1920s, which transformed society while losing money for many investors.
Academic research points the same way. A widely cited study by Michael Cooper, Huseyin Gulen and Michael Schill sorted US companies by how fast they grew their total assets. Over 1968 to 2003, the slowest growing tenth of firms delivered value-weighted returns of about 18% a year, against roughly 5% for the fastest growing tenth, and the effect persisted for up to five years. Debt and equity financing were among the main drivers of the gap.
None of this proves that AI spending will disappoint. It does mean that a company aggressively expanding its asset base, financed with new debt and new shares, starts with the historical odds against its shareholders rather than in their favour. The same logic explains why the suppliers fell hardest on 5 November. Their revenue depends on the spending boom continuing, so any hint of a slowdown hits them first.
The Case for the Builders
A genuine contrarian should argue the other side too, and there is a serious case.
The fibre built during the telecom bubble did not disappear. It was bought cheaply by survivors and went on to carry the streaming, cloud and mobile traffic of the following two decades. The investment created enormous value for society. It simply did not accrue to the shareholders who paid for it at the peak.
The AI builders also differ from the telecom upstarts in important ways. Alphabet, Meta, Amazon and Microsoft entered this cycle with dominant, highly profitable core businesses, and they are building partly to serve demand that already exists, including large contracted cloud backlogs. If any companies can survive a capital cycle shakeout and emerge as the dominant survivors, it is these. In the classic capital cycle, the eventual high returns go to the firms still standing once the overbuilding ends.
The honest conclusion is that two things can be true at once. AI infrastructure may well prove transformative for the economy, while still delivering poor returns on the capital invested at today’s prices. The buyback critics were right that reinvestment matters to society. Investors have to ask a narrower question about whether it pays them.
A Practical Test for Productive Reinvestment
For investors trying to tell the difference, a handful of measures do most of the work.
First, track incremental returns on invested capital, not total profits. The question is what each additional dollar of spending earns, and whether that figure is rising or falling as the spending grows.
Second, watch free cash flow conversion over time. Heavy investment can depress cash flow temporarily, but there should be a credible path to the point where operating cash flow comfortably covers both maintenance spending and growth.
Third, look at how the spending is funded. Investment paid for out of internally generated cash is a bet with the company’s own money. Investment funded by dilution, long dated leases or new debt shifts risk onto shareholders and creditors, and should require a higher expected return.
Fourth, check who the customers are. Revenue from a broad base of independent businesses is worth more than revenue from a small group of counterparties who are themselves funded by the companies doing the building.
Fifth, compare asset lives with payback periods. Data centre shells may last decades, but the chips inside them depreciate far faster. Spending only creates value if it pays back before the equipment becomes obsolete.
Value Is Not the Size of the Cheque
The buyback debate was always framed as a choice between returning cash and investing it. 2026 has shown that the framing was incomplete. The companies that stopped buying back their shares have not answered the critics’ deeper question. They have simply moved it.
The question was never whether a company spends its money on its own shares or on new assets. It is whether the money ends up earning more than it costs. A rising share price did not prove value during the buyback era, and a record capital budget does not prove it now. For investors, the discipline is the same in both cases: ignore the size of the cheque and follow the return.
