The artificial-intelligence boom is entering a more difficult phase. Investment in data centres, chips and cloud infrastructure continues to support economic activity, but investors are increasingly questioning whether future AI revenues will justify the debt and capital spending required to build the industry.
The tension is visible in financial markets. Deloitte reported in October that credit-default-swap prices and yields on AI-related bonds had risen, suggesting that investors see greater risk in the companies financing the infrastructure build-out. The concern is not that demand for AI has disappeared. It is that the scale and speed of spending may be running ahead of dependable cash flow. Deloitte’s weekly economic update also cited Gartner’s forecast that 80% of neocloud providers could cease operating independently by 2030, largely because capital expenditure may overwhelm returns.
Why the spending keeps accelerating
The economic logic behind the boom remains powerful. Training and operating advanced models require vast quantities of specialised chips, electricity and computing capacity. Technology companies are therefore racing to secure data-centre space and long-term power supplies before competitors do. For semiconductor manufacturers, utilities, construction firms and equipment suppliers, that race has created a major source of demand.
AI investment is also functioning as a broader macroeconomic support. The International Monetary Fund raised its 2026 global growth forecast to 3.1%, while saying that AI investment was helping activity in the near term. The fund expects growth to reach 3.4% in 2027, although it warned that energy costs, public debt and stretched asset valuations could threaten the outlook. The Asian Banker’s report on the IMF outlook summarised those projections and the associated risks.
That spending has a self-reinforcing quality. Rising demand encourages more data-centre construction; new capacity makes it easier for businesses to experiment with AI; and successful applications can then be used to justify further investment. Governments also view computing infrastructure as strategically important, linking AI to productivity, defence and technological sovereignty.
The problem of uncertain returns
The difficulty is measuring how quickly those investments will produce durable profits. Some companies are already selling AI services, but the market is crowded, pricing is under pressure and many customers remain in the testing stage. A model may be technically impressive without generating enough revenue to cover the cost of the servers, electricity and financing used to operate it.
Neocloud providers face particular exposure. These firms rent specialised computing capacity to model developers and other businesses, often after borrowing heavily to buy or build hardware. If demand grows more slowly than expected, they may be forced to cut prices while still carrying fixed debt and infrastructure costs. Consolidation could follow, with larger technology companies acquiring assets or smaller providers shutting down.
Optimists counter that early infrastructure spending often precedes productivity gains. The internet, mobile networks and previous computing waves required substantial investment before their most valuable applications became clear. On this view, today’s financial pressure may remove weaker companies while leaving a more efficient industry behind. AI could eventually reduce costs, raise output and create new products in sectors that have barely begun to adopt it.
Energy and inflation complicate the picture
The AI build-out is arriving alongside a wider energy shock. Reports in October said disruptions to Middle Eastern fuel supplies had pushed energy prices sharply higher, with energy-price increases reaching 18.8% in September, compared with 14.3% a month earlier. Higher power costs directly affect data-centre economics and indirectly make inflation harder for central banks to contain. Le Monde’s world-economy coverage described the pressure from fuel disruptions and the strategic importance of the Strait of Hormuz.
For governments, that creates a trade-off. Subsidies, tax incentives and public investment can accelerate domestic AI capacity, but they may also transfer losses to taxpayers if projects fail. Restricting development could reduce financial risk and electricity demand, yet leave countries dependent on foreign providers and vulnerable in a strategic technology race.
What comes next
The next phase is likely to be defined less by headline investment totals than by utilisation, margins and balance sheets. Companies will need to show that customers are paying for AI at a scale sufficient to support infrastructure costs. Lenders will scrutinise contracts, power availability and the resale value of specialised hardware more closely.
A correction would not necessarily end the AI transformation. It could instead redirect capital from speculative providers toward companies with stronger customer demand, cheaper energy and clearer applications. The larger risk is a disorderly pullback: cancelled data centres, concentrated losses among lenders and a sudden reduction in technology spending while the global economy is already coping with inflation and geopolitical energy pressures.
The central question is therefore not whether AI matters. It is whether the industry can convert extraordinary technical ambition into ordinary, repeatable profits before its financing costs become the dominant story.