Microsoft, Amazon Web Services and Google Accelerate AI Capital Cycle

Key Takeaways

• AI infrastructure spending increasingly circulates through cloud contracts, equity investments and debt financing
• Foreign capital is concentrating in data centers and semiconductor projects as demand for computing capacity rises
• The cycle supports rapid expansion, but exposes investors and operators to utilization, energy and refinancing risks

The AI buildout is becoming more than a technology investment cycle. It is also a financing system in which hyperscalers, chip suppliers, data-center operators, lenders and overseas investors increasingly depend on one another.

Microsoft, Amazon Web Services, Google Cloud and Oracle are committing capital to chips, cloud regions and computing capacity. Infrastructure vendors use those commitments to support additional construction or financing. GPU-cloud providers such as CoreWeave can then offer more capacity to customers, including businesses whose spending ultimately flows back toward the same technology ecosystem.

That circularity helps explain how infrastructure can expand so quickly. It also makes the underlying demand picture harder to read.

Bloomberg has characterized the structure as a money loop in which technology companies invest in one another while foreign capital finances US data-center and AI development. The comparison with the pre-2008 period centers on interconnected funding and global imbalances, rather than a claim that AI infrastructure resembles mortgage-backed securities in every respect.

There are important differences. Data centers, chips and power infrastructure are productive assets, and cloud services generate revenue from enterprise customers. Many of the companies driving spending also have large existing businesses and substantial cash flows. Still, productive assets can be overbuilt if projected demand arrives slowly, financing costs rise or customers struggle to turn AI experimentation into durable revenue.

The scale is already striking. UNCTAD reported that data centers attracted more than $270 billion in announced foreign direct investment during 2025, representing over one-fifth of global greenfield investment. Announced semiconductor project values rose 35% as AI demand intensified.

Meanwhile, global foreign direct investment increased 14% to an estimated $1.6 trillion in 2025. Inflows to developed economies rose 43% to $728 billion. Those figures suggest that AI infrastructure is not merely redirecting technology budgets. It is reshaping cross-border capital allocation.

Announcements do not automatically become completed facilities, and reserved computing capacity does not automatically become sustained usage. A long-term cloud contract can provide confidence for lenders and developers, but its value depends on the creditworthiness of the customer and the commercial need behind the reservation. If several layers of the market rely on the same optimistic demand assumptions, apparent diversification can mask concentrated exposure.

Spending momentum remains powerful. Intersect360 Research estimated that global accelerated AI and high-performance data-center infrastructure spending exceeded $300 billion in 2025, an increase of 60.1% year over year. It projects the market will surpass $520 billion by 2030.

Hyperscaler concentration adds another layer. Microsoft, Amazon Web Services and Google controlled 58% of global hyperscale data-center capacity, according to Synergy Research Group. Hyperscaler infrastructure spending reached $142 billion in the third quarter of 2025, nearly triple the year-earlier level.

That concentration can improve execution because major cloud operators have procurement expertise, established customers and access to capital. It can also make suppliers more sensitive to changes in a small number of corporate budgets. What happens if even one hyperscaler slows its expansion plans?

Power is another constraint. New facilities require grid connections, generation capacity, cooling systems and, in many regions, lengthy permitting. Capital may be available before electricity is. Delays can leave expensive equipment or land waiting for supporting infrastructure, weakening expected returns and complicating debt schedules.

For enterprise buyers, the central issue is not whether the AI boom will continue, but how contract structures distribute risk. Procurement leaders should examine minimum-spend commitments, pricing escalators, capacity portability and termination provisions. Investors may want clearer disclosure around customer concentration, related equity stakes, debt maturities and expected utilization.

The money loop can keep supporting innovation while demand expands. Yet its resilience will depend on whether AI workloads produce enough recurring economic value to support the chips, facilities, power projects and financing built around them. The closer those relationships become, the more closely boards, lenders and regulators are likely to examine who ultimately carries the risk.