CPD: AI financing – bubble, breakthrough or the next fixed income supercycle?

Key takeaways

• AI represents a major investment cycle with distinctive financial foundations. Unlike earlier infrastructure booms, much of the AI build-out is supported by substantial hyperscaler operating cash flows rather than primarily by debt. Semiconductor supply, electricity grid access and project level execution may constrain speculative expansion. Nevertheless, accelerating capital expenditure and the relatively short ‘useful’ lives of computing equipment warrant scrutiny.
• Credit risks differ sharply across borrowers and structures. Senior hyperscaler bonds, data centre project finance, equipment-backed transactions and leveraged cloud providers offer different exposures to construction, tenant, refinancing and demand risks. Assessing repayment sources and contractual protections is essential; additional yield may not adequately compensate for weaker structures.
• The next two to three years will test the financing case. The critical indicators are enterprise adoption, AI-related revenues, hyperscaler cash flows, infrastructure availability and debt issuance. These should provide clearer evidence of financial sustainability, even as the broader effects on productivity, businesses and employment take longer to materialise.

The AI infrastructure expansion is creating opportunities across investment grade, high-yield and structured credit markets.

For bond investors, the central question is whether borrowers and their financing arrangements can generate dependable cash flows to service debt through the investment cycle.

How does the AI infrastructure boom compare with previous investment cycles?

The financing mix is a key distinction. To take one historical example, the expansion of the railroads in the US during the 19th century was more than 75%¹ funded through debt. The current AI build-out remains substantially supported by the operating cash flows of major hyperscalers.

Although accelerating capital expenditure is compressing free cash flow and could push some companies into negative territory, underlying operating cash generation remains substantial.

Asset lives also differ. Railroad infrastructure may operate for decades, whereas graphics processing units (GPUs) are typically depreciated over a much shorter, single-digit number of years, although assumptions vary. Semiconductor availability, power access and regulatory constraints also limit the pace of expansion.

Previous cycles nevertheless demonstrate the risks of rapidly expanding debt funded capacity, including the 2015–16 correction in energy credit. The expansion of new financing structures also warrants scrutiny as underwriting practices and rating methodologies evolve alongside the market.

AI-related borrowing could become a major component of high-yield issuance within two to three years. However, building capacity is more difficult than in some previous cycles, and credit investors are assessing links to hyperscaler balance sheets and contractual support at an early stage. The banking system is also less leveraged than before the global financial crisis.

Where could AI disruption create the greatest credit risks?

Software, business process outsourcing and IT services face potential disruption, given AI could reduce barriers to software development and challenge business models based on charging for labour.

However, providers of mission-critical systems may be more resilient because switching is costly, while simpler services could face greater competition.

Against this backdrop, credit analysis must consider whether management teams are adapting their products and operating models.

Long-dated investment-grade bonds introduce an additional challenge: assessing competitive positions and creditworthiness over several decades becomes more uncertain during rapid technological change.

How should investors assess hyperscaler commitments and off-balance-sheet financing?

The nature and duration of each obligation are important. Future equipment purchases are more flexible than long-term data centre leases.

A data centre equipped with NVIDIA GPUs represents an investment of tens of billions of US dollars. The powered building and related infrastructure account for a substantial minority of the cost, but the majority is typically attributable to computing equipment.

If demand moderates, hyperscalers may be able to reduce GPU purchases more quickly than lease obligations. Lower capital expenditure could therefore support free cash flow, even where lease commitments remain.

Where are the main opportunities and risks across credit markets?

Senior unsecured hyperscaler bonds provide exposure to diversified businesses and established revenue streams rather than a single project.

However, special-purpose vehicles financing data centres may depend on a hyperscaler tenant or other contractual support, while equipment-backed transactions introduce different asset and maturity risks. The source and enforceability of repayment cash flows are therefore central to security selection.

AI-related issuance is also expanding in high yield, particularly through data centre construction projects intended to be refinanced once facilities are operational, powered and leased.

Given these considerations, underwriting should assess construction execution, electricity availability, tenant strength and refinancing assumptions.

Highly leveraged neocloud providers, represent a different risk profile from projects supported by investment-grade tenants. Their relatively high financing costs reflect concerns around leverage, future demand and the sustainability of their business models.

What could trigger a deterioration in AI-related credit?

A material weakening in hyperscaler financial strength or a reassessment of long-term AI demand would have broader implications than the failure of an individual project. Many high-yield financing structures ultimately rely on hyperscaler leases or commitments.

Refinancing risk also depends on transaction design. A project with a long-term investment-grade lease may be capable of repaying debt from contractual cash flows without relying on a favourable refinancing market.

Diversified hyperscalers are not necessarily dependent on any single AI model succeeding. Their infrastructure can support multiple customers and applications.

Consensus estimates indicate that the five largest hyperscalers could generate approximately US$6 trillion in aggregate operating cash flow over the five years to 2030. A slowdown in capital expenditure could also improve free cash flow and reduce borrowing requirements, although outcomes will differ across issuers.

Which indicators will be most important over the next two to three years?

Looking ahead, a central indicator is whether greater computing capacity and model training continue to improve AI capabilities and translate into commercial applications, enterprise demand and revenue.

Investors should also assess the balance between supply and demand. Excess capacity could place pressure on pricing, while electricity shortages and lengthy grid connection queues may constrain growth.

Further, enterprise monetisation and the financial health of the AI ecosystem will also be critical indicators of long-term viability.

The ability of major AI developers to access public equity and investment-grade credit markets could also provide useful evidence of the financial health and maturity of the wider AI ecosystem.

Ultimately, the next two to three years should clarify the sustainability of infrastructure financing, although the wider effects on employment and productivity may take a decade or more to emerge.

Could AI debt issuance become too concentrated in investment-grade markets?

Increased bond issuance by hyperscalers could raise their weighting in investment-grade indices, although this is not currently considered a significant concern for market access or financing capacity.

The potential returns from AI investment may justify higher borrowing costs. Even a 100-basis point increase in funding costs may not materially deter major hyperscalers from raising capital to support infrastructure expansion.

Stronger-than-expected operating cash flows could also moderate future issuance requirements, particularly if AI investments begin generating returns sooner than anticipated.

What are the implications for structured credit and private markets?

The repayment profile is decisive. Some investment-grade data centre structures are designed to amortise through contractual lease payments, while certain asset-backed transactions retain exposure to residual asset values at maturity.

Location, power availability, construction status, alternative uses and tenant quality all affect recoveries. Analysis should consider the borrower’s full obligations across parent bonds, subsidiaries and equipment-backed financing.

When it comes to private AI infrastructure financing, this should be distinguished from conventional direct lending to leveraged software businesses, which may be more vulnerable to disruption and refinancing pressure. Some infrastructure transactions resemble investment-grade credit and may suit long-term investors.

Across public and private markets, the essential tests remain the same: the identity and financial strength of the repayment counterparty, contractual protection and whether expected returns adequately compensate for risk.

Take the FAAA accredited quiz to earn 0.5 CPD hour:

CPD Quiz

The following CPD quiz is accredited by the FAAA at 0.5 hour.

Legislated CPD Area: Technical Competence (0.25 hrs) and General (0.25 hrs)

ASIC Knowledge Requirements: Economic Environment (0.25 hrs) and Fixed Interest (0.25 hrs)

please log in to start this quiz

———–

References:
[1] Source: Liaquat Ahamed, https://fortune.com/2026/06/02/ai-bubble-history-railroad-crash-1873-pulitzer-liaquat-ahamed/

Statements attributed to an individual represent the opinions of that individual as of the date published and may not necessarily reflect the view of Capital Group or its affiliates. This communication is intended for the internal and confidential use of the recipient and not for onward transmission to any other third party. This communication is of a general nature, and not intended to provide investment, tax or other advice, or to be a solicitation to buy or sell any securities. All information is as at the date indicated and attributed to Capital Group unless otherwise stated. While Capital Group uses reasonable efforts to obtain information from third-party sources that it believes to be accurate, this cannot be guaranteed. In Australia, this communication is issued by Capital Group Investment Management Limited (ACN 164 174 501 AFSL No. 443 118), a member of Capital Group, located at Suite 4201, Level 42 Gateway, 1 Macquarie Place, Sydney, NSW 2000 Australia. All Capital Group trademarks are owned by The Capital Group Companies, Inc. or an affiliated company. All other company names mentioned are the property of their respective companies. © 2026 Capital Group. All rights reserved.