AI Debt Risks: CDS Surge and Data Centre Financing

The AI investment boom is facing closer scrutiny in credit markets. Investors are examining how quickly spending on chips, data centres and electricity can turn into dependable cash flow—and who bears the losses if projects run late or demand disappoints.

The Financial Times reported on 20 October that weekly CDS trading across seven technology borrowers averaged more than $9 billion in September, versus less than $2 billion in July, citing Morgan Stanley Research. The group comprises Oracle, SpaceX, Alphabet, Amazon, Meta, Broadcom and Nvidia. The increase highlights hedging activity, rather than establishing that defaults are imminent.

CDS trading: more activity does not mean more defaults

A credit default swap is a derivative that transfers specified credit risk. The protection buyer makes contractual payments; the seller compensates the buyer if a defined credit event occurs, subject to the contract’s terms. Investors can use CDS to hedge debt exposure or express a view on creditworthiness.

Three measures answer different questions:

How to interpret credit-market signals
Measure	What it shows	Key limitation
Trading volume	How much contract notional changes hands during a period	Repeated trading is not the same as a net increase in protection outstanding
CDS spread	The quoted cost of credit protection	Reflects liquidity and market assumptions as well as credit risk
Bond yield	The return implied by a bond’s market price and promised payments	Combines interest-rate, credit and liquidity effects

The reported dollar totals describe trading activity, not premiums paid, losses incurred or the probability of default. A more active derivatives market can improve risk transfer while also revealing greater concern about concentrated exposure.

The AI spending challenge: cash leaves before returns arrive

Reuters reported on 19 October that S&P Global expects Amazon, Microsoft, Alphabet, Oracle, Meta and SpaceX to spend over $7 trillion on AI infrastructure during 2025–2030, with negative free cash flow through at least 2027. S&P also warned of incompletely visible AI liabilities, while seeing no immediate reason to change ratings. These are forecasts and credit assessments, not realised spending totals or default findings.

S&P’s August research announcement separately forecast negative free operating cash flow for all six companies in 2026 and 2027. It identified debt, leases, guarantees and special-purpose structures as increasingly important financing channels.

Negative free cash flow means operating cash generation does not cover the relevant capital expenditure under the measure being used. It does not necessarily mean an accounting loss or an inability to pay debts. A company can fund an investment phase from cash reserves, borrowing, asset sales or equity. The credit question is whether those resources remain sufficient until the investment produces sustainable returns.

Where the risk lies: follow the contracts as well as the debt

An infrastructure supplier may borrow to buy processors, acquire land and construct a facility, supported by a long-term customer contract. The supplier’s borrowing is its own liability. The customer may nevertheless have substantial future payment obligations, guarantees or other contractual exposure.

It would be misleading to count every supplier loan as additional debt owed by the technology customer. Equally, looking only at a customer’s outstanding bonds can miss economically significant commitments. Some obligations are recognised on balance sheets; others appear in disclosures or become relevant under specified conditions. The accounting and contractual details matter.

Credit analysis therefore needs to establish who owes what, when payments begin, what happens after delivery delays, and whether commitments can be cancelled. A large contracted order book is valuable, but it is not cash already received.

CoreWeave: bond yields expose the financing challenge

Neocloud providers rent specialised computing infrastructure to AI customers. Their business models can combine concentrated customers, expensive equipment and substantial financing needs.

Reuters reported on 16 October that the yield on CoreWeave’s October 2031 dollar bond briefly exceeded 16% the previous day. That is a historical market observation, not a current executable quote.

A higher secondary-market yield signals a lower bond price and a greater return demanded by buyers. It does not automatically change the coupon on existing fixed-rate debt. The pressure becomes particularly important when a borrower needs fresh financing or must refinance maturing obligations.

There is a counterweight to contagion fears. Reuters reported on 20 October that J.P. Morgan considered most banks’ CoreWeave exposure limited and well secured, although its estimates relied on simplified assumptions and incomplete disclosures. The bank identified the mismatch between upfront investment and later revenue as a central risk.

Microsoft and Nscale: a power delay can become a credit problem

Reuters reported on 19 October that Microsoft ended its agreement for Nscale’s Loughton, Essex, facility after electricity availability slipped into the 2030s. The planned site involved roughly 23,000 Nvidia GPUs. Both companies said their broader relationship continued. This was a site-specific withdrawal, not a general Microsoft exit from AI infrastructure. See the Reuters report on the Nscale power delay.

The financial mechanism is straightforward: debt and construction costs can accumulate before a facility earns revenue. If a delayed connection also affects a customer’s obligation to pay, a physical bottleneck becomes a financing problem. Alternative generation may help in some cases, but introduces its own costs, permits and delivery risks.

Market implications: watch cash conversion and credit conditions

Our assessment is that AI infrastructure risk should be monitored as a chain linking customers, developers, equipment suppliers and lenders. A problem at one project does not establish a sector-wide collapse. Persistent financing stress across several borrowers would be a more consequential warning.

Credit pricing: watch sustained spread widening and new-issue financing terms, alongside trading volumes.
Cash conversion: compare capital expenditure with operating cash flow, available liquidity and debt maturities.
Project execution: track energised capacity and paying workloads rather than announced construction alone.
Contract quality: examine customer concentration, payment triggers, guarantees and cancellation rights.
Transmission: look for evidence of tighter bank lending, private-credit losses or reduced infrastructure orders.

For Nasdaq-linked equities and NQ futures, a sustained rise in financing costs could undermine confidence in AI earnings expectations. Conversely, successful project delivery and stronger cash generation could ease those concerns. These are potential transmission channels, not directional trading forecasts.

Energy adds a separate constraint. Higher power costs can weaken project economics, while wider energy inflation can complicate monetary policy. Our analysis of changing energy routes and inflation risks explores that broader backdrop. Neither a single grid delay nor higher CDS turnover establishes that the overall economy is slowing.

Preparation improves resilience; it cannot eliminate risk. Diversification, appropriate position sizing, liquidity planning and clearly defined exposure limits can reduce vulnerability. Hedges also carry costs, counterparty exposure and the possibility that protection does not match the underlying loss. The practical objective is to withstand adverse outcomes while assessing whether AI investment is producing the cash flows needed to support its financing.

AI Debt Risks: CDS Trading Surges as Infrastructure Costs Mount