How do you invest in a future whose direction you know but whose shape you don’t?
The debate about data centers is usually conducted as a technology debate. In fact, it reveals something more fundamental: how differently institutions deal with capital allocation under uncertainty – and why the biggest challenge is often not recognizing the future, but allocating capital to a future whose economic contours are still blurred.
Same future, different risk models
The Financial Times recently reported on a warning from the European Central Bank about risks surrounding the financing of data centers and AI infrastructure.
The starting point is not the technology itself. Rather, the ECB is examining the question of what consequences could arise if the expected AI-related cash flows fall short of today’s expectations. In a stress scenario, the central bank points to possible valuation losses of four to six percent of insurers and pension funds’ assets. The real concern is not so much immediate loan defaults as possible second-round effects via private credit, leveraged loans, high-yield bonds and equity markets.
Central banks usually deal with banks, credit markets, inflation or financial stability. Data centers belong more to the world of hyperscalers, infrastructure investors and private credit funds.
The ECB’s concern is not about the technology itself. It focuses on the robustness of a chain of assumptions whose economic success depends on demand, capacity utilization, financing and technological progress at the same time.
While the debate is often conducted along technological possibilities, the central bank focuses on the complexity of the prerequisites that lie between today’s investments and the expected future returns.
This is exactly where capital allocation under uncertainty begins.
Large parts of Wall Street, on the other hand, come to an almost mirror-image conclusion. Morgan Stanley estimates the need for investment in data centers and AI infrastructure to be nearly $3 trillion by 2028 and points to a funding gap of around $1.5 trillion. Private credit houses such as Apollo, KKR or other large investors position themselves along a similar basic assumption: the greater risk lies not in overinvestment, but in an undersupply of digital infrastructure.
Unlike the ECB, this view does not primarily discuss the robustness of the actual investment case. Rather, the starting point is the expected capacity requirement and the question of whether sufficient infrastructure will be created to meet the forecasted demand.
The two positions do not necessarily contradict each other. They answer different questions.
When technology cycles overtake infrastructure cycles
At first glance, data centers seem like another infrastructure investment. In fact, they elude a clear assignment.
They are real estate.
They are infrastructure.
They are technology platforms.
And they are capital market objects.
The ECB, investment banks, infrastructure investors, private credit funds, institutional investors and technology companies are therefore looking at the same asset. However, not through the same glasses.
Technology cycles are beginning to overtake infrastructure cycles.
The current debate therefore extends far beyond data centers. For a long time, infrastructure investments have been dominated by demand, regulation and financing. Increasingly, investors also need to assess how the underlying technology is evolving – and what the economic consequences will be.
The discussion about data centers is therefore not limited to a technology debate. At its core, it is a capital market question: How do institutional investors allocate capital when the economic viability of an investment case is influenced not only by markets, regulation and financing, but also by technological development paths?
This is precisely why actors who normally hardly talk to each other suddenly get into the same discussion. Central banks, infrastructure investors, private credit funds, investment banks, institutional investors such as insurers and pension funds, and technology companies are all looking at the same asset class. However, they look at it through different risk models.
History does not repeat itself. The question is.
The question associated with this is by no means new.
Whenever infrastructure investments meet profound technological changes, investors face the same dilemma: the direction of development seems increasingly plausible. It is much more difficult to be able to identify with any degree of certainty which investments will actually benefit from it.
When the railroad swept Europe and North America in the 1840s, its economic importance was hardly disputed by many contemporaries. Trade, mobility and industrial development were to change fundamentally.
The real uncertainty lay elsewhere.
Which routes would be economically viable?
Which operators would survive?
And how much of the expected future had already been financed before it had even happened?
The railway did not become a lesson in capital market history because the technology failed. It became a lesson because investors had to distinguish between a correct future thesis and their possible overfinancing.
The future can be right and the investment can still be wrong.
The underlying future thesis proved to be correct. And yet the decision to allocate capital to the right projects of the time was challenging.
It is precisely this area of tension that is once again evident today in data centers.
Neither the European Central Bank nor Morgan Stanley fundamentally question the economic importance of artificial intelligence. However, they look at the same development through different risk management lenses.
📌 Conclusion:
Whether artificial intelligence will change the economy has long since ceased to be the focus of debate for most market participants.
Rather, the key question is how capital can be allocated when the economic viability of an investment case depends on a multitude of interrelated assumptions.
The European Central Bank is looking at the robustness of the assumptions . Large parts of Wall Street are focusing it on the expected capacity requirements.
Sources
