The artificial-intelligence boom is entering a more demanding phase. After two years of extraordinary spending on data centres, chips and cloud capacity, investors are no longer asking only how quickly AI can grow. They are asking whether the revenue generated by AI services will be large and dependable enough to support the debt financing the build-out requires.
That shift does not mean the technology has failed. It means the market is moving from an early infrastructure race to a test of commercial durability. The distinction matters: a company can be strategically important and technologically impressive while still producing returns too slowly to justify its valuation or borrowing costs.
From capital rush to financial pressure
Recent market commentary has highlighted rising concern about AI-related credit. Deloitte reported that credit-default-swap prices and bond yields for companies tied to AI infrastructure had increased, reflecting investor anxiety that future revenues may not service the debts accumulated during the expansion.[1] The same analysis cited Gartner’s forecast that 80% of “neocloud” providers could cease operating independently by 2030, largely because capital expenditure may outrun returns.[1]
The underlying model is capital-intensive. Data centres require land, power connections, cooling systems and specialised processors before customers generate meaningful recurring income. Their owners must then keep investing as chip generations change and computing demand evolves. This creates a timing problem: costs arrive immediately, while the hoped-for productivity gains and enterprise subscriptions may take years to materialise.
Higher interest rates magnify that problem. When financing becomes more expensive, investors place greater emphasis on cash flow rather than distant growth forecasts. A business dependent on constant refinancing is more vulnerable than one able to fund expansion from operating income.
Why the optimism has not disappeared
There are strong reasons for continued confidence. AI demand is already supporting unprecedented orders for advanced semiconductors and data-centre capacity. Major technology companies are competing to secure computing power because falling behind in model development could weaken their search, cloud and software businesses. For these firms, infrastructure spending is not simply a bet on selling AI access; it is also defensive spending intended to protect existing markets.
AI may likewise deliver benefits outside the technology sector. Faster software development, automated customer support, scientific modelling and industrial optimisation could raise productivity. If those gains spread widely, current investment could look excessive in the short term but necessary in retrospect.
Market prices have also shown that enthusiasm remains resilient. Morningstar reported that AI stocks rallied globally in late October after strong results from chipmaker SK Hynix eased fears that the spending cycle was already weakening.[2] That reaction suggests investors still see real demand, even as they become more selective about which companies will capture it.
The sceptical case
The bearish argument is not that AI has no uses. It is that the industry may be building too much capacity ahead of proven demand. If customers experiment with AI but do not convert those trials into profitable, long-term contracts, cloud providers could face underused facilities and falling prices. That would make it harder to recover the cost of construction and equipment.
Concentration adds another risk. A small number of chipmakers, cloud platforms and model developers dominate the ecosystem. This can create powerful economies of scale, but it also means that a change in spending by one major customer can affect suppliers, lenders and landlords throughout the chain.
There is a further uncertainty around energy. Data centres require large and reliable electricity supplies, potentially increasing pressure on grids and local communities. Delays in transmission upgrades or public resistance to new facilities could raise costs and slow deployment, undermining aggressive growth assumptions.
What comes next
The next stage is likely to separate infrastructure winners from highly leveraged challengers. Investors will focus more closely on utilisation rates, contract duration, operating cash flow and the share of revenue coming from repeat customers. Announcements of new capacity will matter less than evidence that existing capacity is producing sustainable returns.
Policymakers and lenders also face a choice. Easy financing can accelerate innovation, but excessive debt could turn a technology correction into a broader financial shock if projects are built on unrealistic demand projections. Stronger disclosure about power use, customer commitments and debt exposure would help markets distinguish productive investment from speculative construction.
The AI economy therefore faces neither an inevitable collapse nor a guaranteed transformation. Its central question is financial: can productivity gains arrive quickly enough, and broadly enough, to justify the infrastructure now being financed? The answer will determine whether today’s boom becomes the foundation of a durable technology cycle or a costly lesson in building ahead of demand.