The AI Investment Paradox: Great Technology Does Not Guarantee Great Capital Allocation
Updated: 6 days ago
Can great technology still produce poor investment returns?
Yes. A technology can transform the economy while investors still earn disappointing returns if companies spend too much capital, competition compresses economics, or expectations in the share price become too demanding.
The key mechanism is capital allocation: technological success and investment success are not the same thing. What matters is whether the capital committed to growth can earn attractive returns relative to its cost and to what investors already expect.
Artificial intelligence may become one of the most important technologies of this century.
That does not automatically mean that every dollar invested in building AI infrastructure will earn an attractive return.
These two statements can both be true:
AI can transform the economy.
and
Companies can still invest too much capital trying to win the AI race.
That distinction is essential.
Much of today's debate asks whether artificial intelligence is “real” or a “bubble.”
That may be the wrong question.
The more important question is:
Can a transformative technology generate a poor return on capital because too many competitors spend too much, too quickly?
More than $1 trillion
The Bank for International Settlements estimates that the five largest hyperscalers are set to spend more than $1 trillion on AI-related capital expenditure across 2025 and 2026. The BIS notes that these commitments are outpacing earnings and free cash flow, leading some companies to raise additional debt financing.
The scale is remarkable.
But scale alone tells us nothing about whether the capital is being allocated efficiently.
Imagine five competitors who all believe that only two will dominate a future market.
Each has an incentive to spend aggressively.
For an individual company, refusing to invest could be fatal if competitors build superior infrastructure first.
But when every competitor responds rationally to the same incentive, the sector as a whole can build more capacity than future economic returns justify.
This is a classic strategic problem:
What is rational for each competitor can become irrational for the industry collectively.
The BIS investment-race model
A July 2026 BIS working paper attempted to model precisely this problem.
Its calibrated baseline suggests AI investment could exceed its modelled socially efficient level by roughly 50%, with greater overinvestment under less favourable demand assumptions. The model also highlights potential fragility arising from specialised hardware, debt financing and financial links between companies.
This statistic requires an important qualification.
It is a model result, not proof that today's AI industry has empirically overinvested by 50%.
That distinction matters.
The paper tests what could happen under specified competitive and financial assumptions. It should not be converted into the simplistic claim that “half of AI investment is wasted.”
Doing so would destroy the very nuance the research is trying to illuminate.
The relevant insight is different:
Competitive pressure can cause companies to invest beyond what expected industry economics alone would justify.
The technology can succeed while investors miscalculate
History repeatedly demonstrates that technological success and investment success are not identical.
Railways transformed economies.
Telecommunications transformed communication.
The internet transformed commerce.
Yet infrastructure races around transformative technologies have also produced periods in which too much capital chased the same opportunity.
The assets created during those booms often remained useful long after investors who financed some of them suffered losses.
That is the subtle distinction.
An industry can create enormous social value while some companies destroy shareholder capital.
Infrastructure can be useful without being sufficiently profitable.
Demand can grow without growing fast enough to justify the price paid for capacity.
The technology can win.
The capital allocation can lose.
But we must challenge the thesis
There is a strong opposing case.
Major technological transitions often look wasteful before their full economic value becomes visible.
Demand can expand in response to capacity.
Lower computing costs can create applications that were previously impossible.
Infrastructure built “too early” can later become essential.
And there is already evidence suggesting that AI is associated with productivity improvements. The BIS notes that U.S. sectors with greater AI exposure have experienced stronger productivity gains, although partly alongside weaker employment growth.
This means aggressive AI spending could ultimately prove rational if AI creates enough productivity, demand and new revenue.
The counterargument is therefore substantial:
Perhaps the market is not overbuilding for today's AI economy. Perhaps it is building infrastructure for an economy that does not fully exist yet.
That is exactly why declaring the boom either rational or irrational today would be premature.
The better investment question
The real analytical work begins after accepting both possibilities.
Instead of asking:
Is AI transformative?
ask:
What return must this investment generate to justify the capital committed?
Instead of:
Will AI demand grow?
ask:
Will demand grow fast enough relative to installed capacity?
Instead of:
Which company has the best technology?
ask:
Can technological leadership translate into durable economics after accounting for competition and capital intensity?
And instead of:
How much is being invested?
ask:
What assumptions about future revenues, margins and market structure are required for that investment to earn its cost of capital?
That is an Asset Intelligence question.
The PerCapita decision
Technology quality is only one dimension of an investment thesis.
A great technology can coexist with:
high competition, excessive capital intensity, declining returns on incremental investment, optimistic valuation assumptions and disappointing shareholder economics.
Conversely, apparent overinvestment today could become rational if productivity and demand significantly exceed current expectations.
The thesis therefore should not be:
AI investment is excessive.
It should be:
The more extraordinary the investment boom becomes, the more important it is to distinguish technological importance from economic return.
That is the paradox.
CTA: ANALYZE YOUR ASSET →https://www.percapita.be/asset-intelligence-pro
Technology can be extraordinary while the investment case remains demanding. PerCapita Asset Intelligence separates business quality, competitive advantage, expectations and valuation. Explore the NVIDIA analysis →
Don’t just read the analysis. Test your own thesis.
Use PerCapita Asset Intelligence to structure your asset across six dimensions: Purpose, Quality, Moat, Future, Economics and Valuation. Start free, then upgrade to PRO for €19.99/month to unlock unlimited analyses, Challenge, Compare, Watch, Reassess and Remember.
Asset Intelligence PRO: €19.99/month. Cancel anytime.
Comments