Global Market Strategist
The AI buildout has absorbed a cumulative $1.6 trillion of capital spending since 2023, and investors anticipate upwards of $1 trillion in annual spending for the foreseeable future. What began as a largely cash-funded expansion increasingly depends on debt markets, and recent volatility in AI infrastructure debt and equities has reignited concerns about whether these investments can generate attractive returns.
How much does recent stress among AI infrastructure providers tell us about the sustainability of the broader buildout? Financing risks are growing, but so is demand for compute, supported by an ecosystem with substantial financial resources. The diversity of business models and balance sheets, however, means not everyone will benefit equally.
Why are neoclouds under pressure?
Neoclouds are specialized AI cloud providers that have emerged as important sources of computing capacity for hyperscalers and frontier AI labs. They have helped accelerate infrastructure deployment in a capacity-constrained market, and Gartner expects them to capture 20% of the AI cloud market by 2030.
Their business models are extraordinarily capital-intensive. Infrastructure investments must be financed upfront against customer revenues earned over several years, leaving returns vulnerable to higher borrowing costs, contract renewals and changes in compute pricing. Concentrated customer bases and rapidly depreciating GPUs can compound those risks.
These vulnerabilities are increasingly reflected in credit markets. Credit default swap spreads for CoreWeave recently approached 1,200 basis points, reflecting substantial concerns about creditworthiness. But pressure is not uniform. Customer contracts, leverage and financing structures vary considerably, and while hyperscaler spreads have widened too, investors are demanding far greater compensation from more leveraged infrastructure providers.
A stronger foundation for the broader buildout
The hyperscalers underpinning much of the expansion have strong balance sheets, diversified revenue streams and substantial internally generated cash flow. Cloud revenues grew over 50% year-over-year in Q2 across leading providers, while large contracted backlogs provide visibility into future demand. These companies have considerable capacity to absorb higher borrowing costs and take on additional debt.1
The ecosystem's interconnectedness may also provide support. Chip suppliers, hyperscalers and AI developers have strong incentives to keep critical infrastructure projects moving, encouraging strategic investments, long-term capacity commitments and alternative financing arrangements. These relationships can help sustain investment, although they also introduce dependencies and potential channels for contagion.
Most importantly, demand for compute remains exceptionally strong. OpenRouter data suggest agentic AI requests consume roughly 15 times as many tokens as human-initiated requests.2 As these more compute-intensive applications scale, falling inference costs could expand the market further. Meanwhile, persistent bottlenecks in power, equipment and construction should keep compute capacity constrained, supporting pricing power for infrastructure providers.
What should investors watch?
The cost of capital is becoming an increasingly important determinant of the pace and profitability of the AI buildout. Persistently higher borrowing costs could weaken project economics and delay capacity additions. There is also a risk that negative sentiment becomes self-reinforcing, pushing credit spreads wider and making financing more difficult still.
So far, however, sustained spillovers into higher-quality credits appear limited. Hyperscalers retain substantial borrowing capacity, and investor demand for their debt issuance has generally remained robust. A sudden demand shock, perhaps from a broader economic downturn, or a significant shift in competitive dynamics could change that outlook. Weakening compute utilization, pricing or customer commitments would be important warning signs.
For now, the expanding market for AI compute continues to support the broader investment cycle, even as rising capital costs and competition make profitability more challenging for individual operators.