Artificial intelligence is at the forefront of the economic debate, promising to reshape how businesses operate and how productivity is generated. Its investment cycle is already supporting activity, well before the full benefits of adoption can be measured. Expectations are considerable, although the pace of commercialisation and the eventual returns remain uncertain. What is tangible is the commitment of capital to the infrastructure required to make those ambitions possible. For credit investors, this opens a substantial opportunity, alongside a growing responsibility to distinguish financial resilience from technological promise. The challenge is to finance the future with sufficient protection against the uncertainties of the present.
AI enters credit markets at scale
Since September 2025, AI financing has become an increasingly prominent feature of credit markets. The surprise has been the scale and breadth of borrowing by the hyperscalers, the large technology groups behind the expansion of cloud and AI capacity. Their high-grade bond issuance reached approximately $182 billion by August 2026, compared with $93 billion during the whole of 2025.[1] This marks a change for a sector long associated with the ability to finance growth from its own cash generation. The scale of infrastructure spending is bringing even the strongest balance sheets back to the bond market. Funding now extends across currencies and maturities, combining conventional bonds with secured debt, leases and project structures across the wider ecosystem. These issuers are also becoming harder for investors to overlook: major hyperscalers represented approximately 4.8% of the Bloomberg US Corporate Investment Grade Index in July 2026.[2] AI is consequently influencing portfolio exposure even where investors have made no explicit thematic allocation.
The financing requirement is also reaching high yield, as specialist developers and operators undertake part of the data-centre construction that hyperscalers choose to procure externally. Much of this borrowing finances buildings, power connections and cooling systems against future contracted revenues. The distinction is important: investors are often financing bricks and mortar, with repayment dependent on the delivery of a functioning asset, rather than lending against an established software business. The hyperscaler may be the ultimate customer without being the entity responsible for servicing the construction debt. This type of financing therefore brings construction and project-finance risk into portfolios alongside technology exposure. Direct data-centre exposure reached roughly 3.4% of the US corporate high-yield index by August 2026.[3] Strong customers can support these structures, but their credit quality does not automatically transfer to the borrower. We expect substantial further supply into 2027 as investment commitments are translated into operating capacity. The opportunity set should broaden, while the market’s ability to absorb that supply will increasingly depend on pricing and structure.
A broader opportunity for credit
There are sound reasons to approach this development constructively. AI investment is already supporting economic activity, with demand extending into utilities, electrical equipment, engineering and construction. These industries provide channels through which spending can support employment and growth beyond the technology sector. This gives the investment cycle a broader economic footprint and helps explain why its significance extends beyond the fortunes of individual AI providers. Debt financing can support productive assets whose useful contribution outlasts the initial wave of enthusiasm. The eventual productivity dividend remains uncertain, but the current investment cycle has tangible economic value. For creditors, the prospect of durable revenues is more relevant than the most ambitious projections of technological change. Established franchises, substantial operating cash flows and contracted demand provide a credible foundation for parts of this expansion.
Credit markets also gain greater diversity and additional sources of performance. Investment-grade investors can access more technology debt, while high-yield investors gain a growing infrastructure segment. Long-term leases with strong counterparties can improve revenue visibility; explicit guarantees or financial support may strengthen a project’s credit profile further. Completion and the commencement of rental payments can materially reduce risk, although an investment-grade tenant does not guarantee an investment-grade borrower.
There is also the attraction of carry. Greater financing needs can create opportunities to secure additional spread from otherwise resilient issuers. Where the yield adequately compensates for leverage, execution risk and the bond’s terms, this can be a valuable addition to portfolios. We do not need every investment to generate exceptional equity returns for a bond to perform. We need cash flows capable of meeting interest payments and repaying principal, with enough flexibility to withstand a less favourable outcome. The benefit lies in the quality and price of the debt being offered, rather than issuance growth alone.
The risks begin with physical delivery
Execution is central to the credit assessment. Much of the planned expansion in AI depends on data centres that must first be built, connected and commissioned. Across major North American data-centre markets, capacity under construction in the first half of 2026 was equivalent to almost 70% of existing inventory.[4] That pipeline illustrates the distance between announced ambition and revenue-producing assets. The financial timetable can begin well before the operating timetable: capital is committed and debt drawn while the asset is still unable to earn the revenues on which its financing depends. Pre-leasing provides comfort on demand, but cannot remove the risk that a facility is delivered late or above budget.
Power is a particularly difficult constraint. A building may be completed before the grid connection, transmission capacity or generation needed to operate it is ready. These dependencies are not easily compressed to match the pace of technological development. An ambitious construction schedule offers limited reassurance if the infrastructure outside the site cannot be delivered alongside it. Permits and specialist equipment introduce further dependencies, while inflation and supply-chain pressures can raise the cost of copper-intensive electrical systems and other construction inputs. Delays then become a credit issue: interest accrues while lease commencement and cash generation may be deferred. A developer can require additional funding precisely when its financial flexibility is diminishing. For the hyperscaler, unavailable capacity can delay the revenues expected to justify its investment. As these exposures grow, a common construction bottleneck could affect credit portfolios across rating categories.
Oracle’s September force-majeure notice relating to Project Jupiter in New Mexico[5] illustrates how physical constraints can enter contractual relationships. The notice followed permitting problems affecting a gas-pipeline component. Oracle maintained that the wider project remained on schedule; the notice was neither a cancellation nor a credit event. The lesson is nevertheless relevant: investors need to understand who bears the cost when an essential part of a project cannot proceed as planned.
Public acceptance creates another hurdle. Concerns over electricity bills, water use and noise have drawn opposition across political divides, with New York’s moratorium on new hyperscale facilities illustrating the potential for restrictions to interrupt development. The concern for creditors is not only whether a project eventually receives approval, but how long that process takes and whether additional conditions change its economics. Local opposition can therefore matter to a borrower even when the wider economic case for AI investment remains compelling. Commercial demand alone is insufficient to ensure delivery. A project must also secure the infrastructure and local acceptance required to operate. For bondholders, these considerations belong in the assessment of completion dates, liquidity needs and the reliability of future revenues.
Financing costs compound these pressures. Elevated sovereign yields increase the underlying cost of borrowing, while heavy supply can require issuers to offer larger new-issue premiums. Even strong hyperscalers face greater investor scrutiny as their funding needs expand. New-issue concessions reward investors for committing fresh capital, but they also demonstrate that the market is becoming more selective. Strong ratings can preserve access to funding without insulating an issuer from a higher cost of that funding. For borrowers, higher coupons increase the importance of converting investment into cash flow on time. For investors, the additional carry is attractive only if it provides sufficient compensation for the risks assumed. Abundant demand for AI services cannot by itself resolve an unfavourable combination of construction delays, rising costs and expensive debt.
The financing structure also matters. Leases, joint ventures and project vehicles can distribute obligations across several entities without removing the underlying economic dependence. Reported leverage may consequently provide an incomplete view of financial commitments. We need to assess who ultimately supports the asset, where creditors rank and how much refinancing is required before operations become self-funding. A long-term commercial relationship is not equivalent to an unconditional guarantee. The contractual allocation of cost overruns, completion obligations and delayed payments needs to be understood before the strength of the customer can be treated as protection for the creditor. Strong headline ratings deserve attention, but the terms of an individual security can be just as decisive.
Financial resilience will determine the outcome
AI offers a meaningful long-term opportunity, but the path from investment to dependable cash generation is unlikely to be linear. Fundamental, bottom-up research remains our most useful guide. At issuer level, this means assessing balance-sheet resilience, total financial commitments and the capacity to moderate spending if revenues disappoint. Ambition must be assessed against the resources available to fund it, rather than against the potential size of the market alone. At project level, it means testing construction progress, access to power and the credibility of completion schedules. At bond level, covenants, guarantees, seniority and refinancing requirements determine how much protection creditors actually receive.
These distinctions should create dispersion and opportunities for active management. A substantially completed facility with secured power and a strong tenant presents a different proposition from a development still awaiting essential approvals. Equally, a hyperscaler able to finance expansion largely from operating cash flow has more flexibility than one increasingly reliant on external capital. The same discipline applies within an issuer’s capital structure, where maturity and contractual protection can produce materially different risk profiles. A convincing sector outlook is a starting point for that analysis, not a substitute for it. Issuer and issue selection will be essential to capture the carry on offer while retaining resilience if assumptions prove too optimistic.
Earlier technological revolutions remind us that lasting innovation can coexist with financial excess. Businesses and financing structures may fail even as the technology becomes embedded in everyday life. Railways, telecommunications and the internet each experienced periods when capital ran ahead of monetisation. Their subsequent development did not reward every original investor equally. For bondholders, recognising the importance of a technology has therefore never been enough on its own. The more useful discipline is to ask whether the borrower can remain solvent and retain access to liquidity while the commercial model matures.
AI may follow its own course, but creditors have good reason to focus on the capacity to survive the gap between expectations and execution. Amid the debate over whether AI is viable, beneficial or overvalued, financial resilience remains fundamental.