Skiba’s AI Debt Warning: Hedge Funds Unwind Buildout Bets

Andrzej Skiba says hedge funds have been aggressively selling debt tied to artificial intelligence infrastructure, exposing tension between construction commitments and investor patience. Skiba’s AI debt warning concerns the financing behind the buildout, rather than a new judgment about model capabilities.

Speaking on Bloomberg Television on November 5, RBC Global Asset Management’s head of BlueBay US fixed income described investors closing positions. “We have seen a massive hedge fund capitulation within the space,” he said, according to Bloomberg’s coverage of AI debt selling.

The immediate conflict pits borrowers’ long construction schedules against investors’ changing willingness to finance them. Microsoft, Amazon, Alphabet, Meta, and Oracle operate within that financing landscape, although Skiba’s accessible remarks do not identify individual sellers or affected issuers. The distinction matters: selling debt can reflect portfolio pressure without establishing that the underlying borrower cannot repay.

What Skiba’s AI Debt Warning Actually Establishes

Skiba described an unwind in investor positions, which is evidence of selling pressure rather than proof of widespread borrower distress.

Bloomberg reported that the selling occurred over the preceding few weeks. Skiba’s remarks concern debt associated with the AI buildout, a category encompassing obligations connected to infrastructure investment.

That description leaves important boundaries unresolved. The accessible account does not provide a comprehensive list of funds, securities, transaction sizes, or realized losses. It therefore cannot establish how broadly the activity represents the market.

The word “capitulation” conveys the intensity of Skiba’s observation. It does not supply a standardized measure of investor withdrawals or demonstrate that every category of AI-linked borrowing experienced the same pressure.

A hedge fund, an investment partnership using strategies that can include borrowing and hedging, can sell for reasons beyond deteriorating business prospects. Position limits, investor withdrawals, financing terms, and portfolio adjustments can all influence trading.

Those are possible mechanisms, not verified explanations for the transactions Skiba described. His remarks alone do not establish which mechanism dominated.

That distinction becomes important when interpreting the headline. An investor selling an existing bond transfers a claim to another buyer. The transaction does not automatically withdraw money already delivered to the issuer.

A separate question concerns future borrowing. If buyers require better compensation, borrowers seeking fresh funding face different terms. Infrastructure plans become sensitive to that change when they depend on continued access to financing.

The first conclusion is therefore specific. A senior credit investor says hedge funds have been reducing exposure to AI-related debt. The next question concerns whether that selling changes financing conditions.

Readers should also separate the observation from the institution making it. Skiba’s position provides relevant market experience, but his comments remain an attributed assessment. They are not a marketwide transaction database.

For the RBC AI debt outlook, that attribution should remain attached to the claim. An observation from a portfolio manager carries different evidentiary weight from audited borrower disclosures or a completed financing announcement.

The practical consequence is a narrower, more useful reading. The remarks identify a market development worth investigating. They do not establish a complete explanation for the development.

This is where the story moves beyond a trading headline. The financing behind AI infrastructure requires investors willing to hold obligations across construction, equipment installation, and commercial operation.

A change in that willingness deserves attention even when no borrower has missed a payment.

The Pressure Falls on Projects That Still Need Funding

The most exposed borrowers are those whose remaining commitments depend on financing they have yet to secure.

The AI expansion includes hyperscalers, meaning technology companies operating computing platforms at very large scale. Their financing needs connect them with developers, lenders, equipment suppliers, and institutional investors.

The OECD’s 2026 analysis documents the shift toward borrowing. It found that major hyperscalers issued 122 billion dollars in bonds during 2025, more than three times their historical annual average since 2000.

Its corporate borrowing analysis also distinguishes direct company borrowing from financing conducted through separate project entities. Those structures complicate comparisons based only on headline corporate debt.

The implication is that funding pressure must be assessed project by project. A completed facility with committed financing faces a different situation from a development still assembling its funding package.

Consider a hypothetical data center developer with construction underway and a customer contract signed. If its next financing round remains unfinished, less favorable lending terms can change the project’s expected economics.

The customer contract still matters. Its value depends on who signed it, when payments begin, and which conditions permit delays or termination. A recognizable customer name cannot answer those questions.

This creates pressure on sponsors, the companies or investors backing a project. They might need to contribute additional capital, revise construction schedules, or accept financing conditions they previously rejected.

None of those responses follows automatically from Skiba’s comments. They describe the channels through which sustained selling pressure would become relevant to the buildout.

The distinction between existing and future obligations also matters for large technology companies. A company can possess substantial resources while deciding that a particular expansion no longer justifies its financing cost.

That is a capital allocation decision, meaning a choice about where to commit money. It does not require the company to face an immediate repayment problem.

For enterprise customers, the corresponding concern is delivery reliability. A promised computing expansion has commercial value only when the necessary facilities become available.

Developers evaluating future capacity therefore need more than a construction announcement. They need credible milestones linking financing, electrical connections, equipment delivery, and operational readiness.

The pressure is especially relevant when commercial promises precede physical availability. A supplier can announce future capacity before every component needed to deliver it is secured.

Buyers should distinguish operational resources from resources dependent on unfinished projects. That distinction helps explain why a financing story belongs in technology coverage.

AI infrastructure financing determines which plans move from announcements to usable computing capacity. Investors’ willingness to fund the remaining work becomes part of the delivery schedule.

The immediate risk is therefore uneven execution across projects, rather than an automatic halt to the entire buildout.

Long Construction Schedules Meet Investors Who Can Sell

Infrastructure commitments persist even when the investors holding their debt decide to leave.

Research from the Bank for International Settlements describes how AI infrastructure can receive funding through separate entities. These entities develop assets and raise debt, supported by customer leases or capacity commitments.

A special purpose vehicle, a legally separate entity created for a defined transaction, can place project financing outside a technology company’s direct borrowing structure.

The BIS explains that these arrangements link technology companies with private credit investors, insurers, and banks. Its infrastructure borrowing research identifies refinancing pressure and guarantees as potential channels for financial stress.

The analytical issue is the difference between the physical asset and the investor’s position. Construction proceeds through contracts and milestones. A security can change ownership while that work continues.

This mismatch does not make the structure inherently unsound. Long-term investors can hold debt through temporary market weakness. Problems arise when financing assumptions depend on investors behaving differently from their actual constraints.

An investor using borrowed money faces obligations to its own lenders. A project developer faces obligations to contractors and customers. Those obligations operate on different schedules.

If investors become less willing to hold a security, its market value can fall without any immediate change in project operations. That decline becomes operationally relevant when someone needs additional funding.

The same distinction applies to refinancing, which replaces an existing borrowing arrangement with a new one. A project might function as intended yet encounter difficulty obtaining replacement financing on acceptable terms.

That possibility explains why repayment dates deserve attention alongside construction dates. A facility opening on schedule does not eliminate every financial obligation attached to its development.

Guarantees add another layer. A guarantee is a promise to cover specified obligations if stated conditions occur. Its usefulness depends on its contractual scope and the guarantor’s ability to perform.

Readers should avoid treating every guarantee as interchangeable. Support for a facility’s residual value differs from an unconditional promise to repay all outstanding debt.

These distinctions also explain why a large technology customer cannot substitute for examining the borrower. A customer purchasing computing services and an entity owing lenders are not necessarily the same legal party.

For the RBC AI debt outlook, the central question is where obligations ultimately land. The trading headline matters most when it reveals pressure at a financing link supporting unfinished investment.

This is the mechanism behind the tension. Long-term physical commitments depend on financial arrangements whose market value and availability can change much sooner.

The infrastructure remains in place when a position changes hands. The terms available for the next stage of investment can nevertheless become less accommodating.

A Strong Cloud Business Does Not Protect Every Borrower

AI-linked debt combines borrowers with different resources, contracts, and exposure to disappointing demand.

Microsoft and Alphabet operate established cloud businesses alongside their AI investments. Oracle also supplies enterprise computing services, while Meta approaches infrastructure from a business centered on its own applications.

Those business models provide useful comparisons, but they do not establish equivalent credit risk. Credit risk means the possibility that a borrower fails to meet its financial obligations.

A project company serving one customer differs from a diversified corporate borrower. Their securities can share an AI label while giving investors claims on very different sources of repayment.

The International Monetary Fund’s April 2026 assessment provides a counterweight to blanket pessimism. It described major hyperscalers’ credit quality as supported by strong balance sheets and substantial cash generation.

The same assessment identified potential future pressure from investment requirements and equipment becoming outdated. Its AI balance-sheet assessment treated those concerns as vulnerabilities rather than evidence of an immediate financial stability crisis.

That balanced assessment matters. It allows financing conditions to become less favorable without assuming that every major technology company faces the same threat.

For a diversified borrower, lenders can consider several businesses when assessing repayment capacity. For a project entity, the relevant cash might depend heavily on one facility and its contracted customer.

The analysis changes again when a borrower depends on specialized equipment. A building, electrical infrastructure, and computing hardware have different replacement requirements and alternative uses.

The issue is not whether equipment eventually becomes outdated. The question concerns whether the financing arrangement allows enough room for replacement while meeting existing obligations.

A useful comparison therefore begins with the legal borrower. It then examines the assets, customer contracts, repayment schedule, and financial support available to that borrower.

Company size alone cannot complete that analysis. A large sponsor might provide extensive support, limited support, or support triggered only by particular events.

Conversely, a smaller borrower might possess conservative financing and reliable contracts. The AI label does not establish that its obligations are unsuitable for long-term investors.

This is the skeptical angle within Skiba’s AI debt warning. Heavy selling deserves scrutiny, but an industrywide label can obscure the differences that determine actual losses.

A broad market decline can also create opportunities for buyers with different constraints. An investor able to hold a security longer might assess it differently from a seller needing immediate liquidity.

Liquidity means the ability to transact without a substantial effect on price. A borrower’s capacity to repay and a security’s liquidity are related considerations, but they are distinct.

The defensible judgment is that financing structures require closer differentiation. Declaring every AI-linked borrower safe or distressed would outrun the evidence.

Selling Pressure Is Not Yet Evidence of a Solvency Crisis

Skiba’s remarks establish a trading concern; a claim about widespread repayment failure requires additional evidence.

Solvency means the ability to meet financial obligations over time. A decline in a bond’s market value does not independently demonstrate insolvency.

Bond prices respond to several influences, including benchmark interest rates and the compensation investors demand for holding corporate obligations. Selling pressure can therefore coexist with borrowers continuing to meet payments.

The Federal Reserve Bank of Kansas City examined the wider financing landscape in research published October 5. Its authors emphasized structural risks extending beyond short-term asset valuations.

Their AI financing study found that project vehicles and specialized computing providers carried higher leverage than major hyperscalers. Leverage means using debt relative to the capital or assets supporting a business.

That work supports distinguishing vulnerable structures from stronger corporate borrowers. It does not establish that the hedge funds Skiba discussed sold because those structures were failing.

The International Monetary Fund has separately warned about circular financing, arrangements in which connected AI companies act as one another’s investors, customers, and financiers.

Its October discussion of AI financial risks identifies a transmission problem: stress at one participant can undermine expectations elsewhere in the network.

The implication is that multiple contracts do not necessarily represent independent sources of demand. A customer’s purchasing capacity might depend partly on financing supplied within the same commercial network.

That is a reason to examine relationships, not an accusation that the underlying transactions lack economic value. Connected companies can finance useful investment while creating concentrated exposure.

The unresolved question is how much of today’s repayment expectations relies on assumptions that weaken together. Public announcements often reveal individual commitments more clearly than the full network supporting them.

Several evidentiary gaps remain in the immediate story. The accessible remarks do not quantify total hedge fund exposure or establish the share that has been sold.

They also do not provide a complete account of the buyers replacing those sellers. A transfer toward investors willing to hold debt longer would carry different implications from repeated difficulty finding buyers.

Nor do the remarks demonstrate that construction has stopped, customers have canceled contracts, or borrowers have missed payments. Those developments would require separate reporting.

For technology readers, maintaining these distinctions prevents a market observation from becoming an unsupported operational forecast. A selloff can be consequential without establishing an industrywide crisis.

The strongest current conclusion is therefore conditional. Persistent financing pressure would matter most where projects need fresh capital, contractual support remains uncertain, or repayment depends on concentrated demand.

Evidence of functioning markets and reliable project execution would weaken the more pessimistic interpretation. Evidence of unfinished financing and deteriorating contracts would strengthen it.

Watch New Financing, Cash Generation, and Project Delivery

The next test is whether selling pressure changes completed financing, borrower resources, or the delivery of usable infrastructure.

Over the coming months, the most useful evidence should arrive through financing announcements, company disclosures, and project updates. These signals connect investor behavior with consequences that readers can evaluate.

Start with completed financing. An announced borrowing plan shows intent, while a completed transaction demonstrates that investors supplied funding under agreed conditions.

Watch whether borrowers secure the funding they sought and whether they disclose additional protections for lenders. More restrictive conditions would strengthen the interpretation that investors demand greater compensation or contractual support.

Successful financing without substantial changes would weaken the argument that hedge fund selling has become a broad obstacle. It would suggest that other investors remain willing to supply capital.

The relevant comparison should involve similar borrowers and obligations. A financing completed by a diversified technology company cannot establish that every independent developer has equivalent access.

Next, examine cash generation and investment commitments in company disclosures. Free cash flow, cash remaining after operating expenses and capital investment, helps assess the resources available alongside external borrowing.

The useful question is whether operating performance supports committed spending. Stronger cash generation would reduce dependence on new funding, while persistent shortfalls would increase its importance.

Read lease obligations and guarantees alongside direct borrowing. A repayment assessment becomes incomplete when it considers only debt shown under a familiar corporate heading.

Also distinguish contracted business from cash already collected. A future customer commitment can support planning while still leaving construction and payment timing unresolved.

Finally, track project delivery. Operational milestones provide a check on whether financing commitments are producing facilities customers can use.

An opening announcement should identify what has actually become available. A partially operating campus differs from a fully completed development, and energized infrastructure differs from installed computing capacity.

Repeated delays, reduced construction scope, or renegotiated customer obligations would strengthen concerns about the buildout’s financial assumptions. Delivery supported by functioning customer demand would weaken them.

These observations should remain in order: financing first, borrower resources second, and project execution third. Together, they test whether the selling represents investor repositioning or pressure reaching the physical buildout.

For enterprise buyers, the immediate action is to connect promised capacity with evidence of delivery. Ask which resources operate today and which depend on unfinished construction or financing. For developers, maintain a realistic alternative when a service depends on future infrastructure. For researchers tracking Skiba’s AI debt warning, compare subsequent disclosures with the original claim rather than treating every weak trading session as confirmation. The useful question is whether borrowers continue funding and delivering their commitments under changing market conditions. Follow completed transactions, cash generation, and operational milestones to see whether the warning remains a portfolio story or becomes a constraint on available computing capacity.