The danger in a boom is not always that it bursts. Sometimes it merely feeds. That is the more civilized disaster: money does not disappear, it is rerouted, and the rest of the economy slowly discovers it has become a tributary. The latest warning on artificial intelligence is not about a crash in prices. It is about crowding out — the old, unfashionable problem of too much promise in one place and too little oxygen everywhere else.
A Financial Times column by Soumaya Keynes argues that AI infrastructure spending is beginning to crowd out capital investment across the broader economy. That phrasing matters. It suggests a system that is still functioning, but not evenly. The machine keeps running while its gears are stripped from the rest of the workshop. The evidence offered in the column is not a sensational bubble chart. It is a more mundane and therefore more troubling picture: capital expenditure in non-tech sectors has contracted while AI infrastructure spending has surged.
This is how distortions usually begin. Not with a siren, but with a spreadsheet. Investors tell themselves they are funding the future, and perhaps they are. Yet the future rarely arrives alone. It arrives by taking labor, metal, chips, power, debt capacity, and patience from elsewhere. Classical economics has always known this. A dollar cannot build two bridges at once. Capital is finite, even when optimism is not.
The FT column cites Goldman Sachs data showing that roughly 25% of US gross investment-grade debt issuance in 2026 originated from AI-related companies. That is not a minor footnote. It implies that a quarter of a major financing channel has been pulled into one theme. Markets love concentration when the story is upward. They hate it when the bill arrives. The same capital that cheers scale can become brittle under its own weight.
Debt markets are especially revealing because they expose intent. Equity can indulge fantasy longer. Debt insists on repayment. When a large share of issuance comes from one sector, the market is not merely expressing conviction; it is underwriting a shared assumption that the promised infrastructure will earn its keep. History suggests caution here. Railroads, utilities, telecom, shale — every age finds a cathedral to finance. Some leave behind durable networks. Others leave an archaeology of overbuilt hopes.
The useful question is not whether AI is real. It plainly is. The question is whether the current allocation of capital is producing an economy that is stronger overall, or merely more dependent on one expensive line of expansion. A forest can grow faster when sunlight is abundant. It can also become so dense that only a few species survive beneath the canopy.
Investors are drawn to visible winners because they are simpler than invisible risks. That is a psychological flaw as old as markets. People prefer a straight road, even when the road ends at a cliff. If one sector appears to be absorbing the highest returns, the crowd naturally assumes the answer is to buy more of it, finance more of it, and explain away everything left behind. But the economy is not a scoreboard. It is an ecosystem.
When AI investment is projected to rise from 1.8% of GDP to 2.8% by 2028, as the column reports, the number is less important than the direction. Nearly one full percentage point of GDP moving toward a single investment theme is a large swing in a large economy. That does not automatically mean misallocation. It does mean concentration. And concentration is where fragility hides. In probability terms, systems that look efficient in the middle often become unstable at the tails. The average is a poor bodyguard against the extreme.
The deeper problem is that capital chases narratives with the discipline of a river seeking the lowest path. Once a theme becomes legible, lenders and investors begin to believe they are being prudent by joining it. In reality, they may be compounding the same exposure from different angles. The result is not diversification but synchronized confidence. That is a dangerous form of consensus, because it feels like due diligence.
Crowding out sounds abstract, but it behaves like a tax. Not a tax voted on in parliament, but one collected by scarcity. If AI infrastructure demands more financing, more power, more equipment, and more managerial attention, something else gets less. The cost is not always visible in a quarterly report. It may appear as delayed factory upgrades, smaller research budgets outside tech, or a cautious tone in businesses that decide to wait rather than invest. The absence is harder to photograph than the boom.
This is what makes the current debate harder than a simple bubble narrative. Bubble stories comfort the crowd because they promise a single dramatic reckoning. Crowd-out stories are less theatrical. They suggest a long transfer of resources from many ordinary sectors into one extraordinary one. That kind of shift can persist for years before anyone calls it a problem. By then, the economy may have adjusted to lower breadth, as trees adapt to shade by growing taller and thinner. The forest still stands, but it is more exposed to wind.
The FT column’s framing is useful because it avoids the lazy binary. AI can be transformative and still distort capital formation. Indeed, transformative technologies are often the most distorting, because they attract not only their own cash flows but the imagination of everyone who fears missing history. The danger is not enthusiasm alone. The danger is enthusiasm that borrows against the future of everything else.
The correct response is not to sneer at AI or pretend the spending wave is imaginary. It is to remember that every boom has a shadow price. In game theory, once the payoffs of one strategy become obvious, everyone piles into it and changes the game itself. The winning move in the first round becomes the crowded mistake in the next. Markets do this constantly. They reward speed, then punish density.
A more disciplined approach would ask how much of the current AI build-out is creating durable productive capacity, and how much is simply absorbing capital because capital has nowhere else that feels as exciting. That distinction matters because the latter can leave the broader economy thinner even if the headline growth looks strong. If investment rises in one area while spending elsewhere softens, the aggregate may still look healthy for a while. But the base of the pyramid narrows.
There is a Roman lesson in this. Empires often mistake monumental construction for strength. Roads, aqueducts, and temples can all be signs of power. They can also be signs that the state has concentrated its resources where the eye can see them. Invisible maintenance is less glamorous but more valuable. So it is with capital today. The most important allocation may be the one that does not produce a dazzling headline.
The slow sucking sound is more alarming than a crash because it is easier to ignore. It does not demand panic. It demands judgment. And judgment is in short supply whenever a market begins to believe that one idea can justify almost any price, any debt, and any sacrifice elsewhere. That belief is the real vulnerability. AI may change the economy. The harder question is whether the economy can remain broad enough to survive the cost of being changed.
