AI Debt: Data Centre Financing Faces Growing Pressure

On October 16, 2026, CoreWeave shares closed at $61.39 on the Nasdaq, down approximately 25% over the week. In the bond market, some securities issued by the US AI data centre operator were yielding 15%–16%.

The sell-off raises a question extending far beyond technology valuations: will future AI revenues be sufficient to fund infrastructure, replace processors and repay the capital committed to expansion?

CoreWeave's accounts show why this question matters. Revenue is growing rapidly and operating cash flow has turned positive. Yet infrastructure spending and new financial commitments remain substantial.

CoreWeave: credit markets demand higher returns

According to Reuters, certain CoreWeave bonds issued in June 2026 at an initial yield of 9.675% declined enough in price to offer approximately 15% in the secondary market. Other securities approached 16%.

The issue yield is different from the contractual coupon. The relevant 2032 notes carry an annual coupon of 9.625%. Falling prices do not change the interest due on existing fixed-rate securities, but they reveal the returns now demanded by investors. If these conditions persist, new borrowing and refinancing could become substantially more expensive.

Credit default swap prices also reflect the concern. Reuters reported spreads near 1,200 basis points during the week. These prices measure market-perceived credit risk, not an actual payment default.

Financial professionals disagree over the significance. Some expect a sustained tightening of AI infrastructure credit, while others believe selling pressure exaggerated the correction. Current prices cannot establish which interpretation will prevail.

Revenue is growing, but investment is growing faster

In the first half of 2026, CoreWeave generated $4.653 billion in revenue, compared with $2.194 billion a year earlier, an increase of approximately 112%. Yet it reported a net loss of $1.366 billion and net interest expense of $1.176 billion.

Accounting earnings alone do not describe the company's financing needs. Its cash flow statement shows $3.663 billion generated by operating activities against $14.874 billion in net investing outflows.

Those outflows included $14.117 billion in purchases of property, equipment and capitalised internal-use software, largely associated with computing hardware and infrastructure.

Under a simplified definition, free cash flow equals operating cash flow less these purchases. This produces an estimate of −$10.454 billion for the half-year. The figure is calculated from the financial statements and is not an official free cash flow metric reported by CoreWeave.

Including all investing activities, the pre-financing cash balance was −$11.211 billion. The $757 million difference between the two measures comes from other investing transactions. Both figures show that operating cash generation was insufficient to fund the expansion undertaken during the period.

$9.796 billion in equipment-related liabilities without immediate payment

Cash expenditure does not capture every financial commitment created by investment.

In the supplemental disclosures accompanying its cash flow statement, CoreWeave reported $9.796 billion in liabilities related to property and equipment additions during the first half of 2026. These included equipment financed by original equipment manufacturers, or OEMs. They represent non-cash investing and financing activities.

The company also disclosed $1.469 billion in reclassifications of equipment-related liabilities into debt following the execution of certain OEM financing arrangements.

These amounts must not simply be added to the $14.117 billion in cash purchases. Their accounting treatment and payment timing differ, while some figures represent reclassifications. Their significance is that equipment acquisitions can create financial obligations before payments appear in the cash flow statement.

Financing risk therefore extends beyond cash already spent. It also encompasses obligations generated by asset acquisitions, which will eventually need to be covered through operating revenue, existing liquidity or new financing.

Not all $6.904 billion in cash is unrestricted

CoreWeave funded its expansion through $13.985 billion in net financing inflows during the first half, including debt issuance and equity placements. Its combined cash, cash equivalents and restricted cash increased from $4.130 billion to $6.904 billion.

The total requires an important qualification. At June 30, CoreWeave held $5.524 billion in cash and cash equivalents not classified as restricted, alongside $873 million in current restricted cash and $507 million in non-current restricted cash. The restricted portion therefore totalled $1.380 billion.

The full $6.904 billion should not be presented as freely deployable cash. The company also reported $15 million in marketable securities and approximately $10.014 billion in undrawn availability under existing financing facilities, subject to contractual conditions.

CoreWeave is not a business that has immediately run out of financial resources. The question is whether it can sustain access to liquidity, fund future investment and withstand higher borrowing costs.

$35.6 billion in debt against $103.7 billion in future contracts

At June 30, 2026, CoreWeave reported approximately $35.6 billion in total debt and $16.3 billion in operating lease liabilities. These are different categories of obligations and should not be combined into a single net debt figure.

Contractual principal repayments scheduled at that date amounted to $4.413 billion for the remainder of 2026 and $6.184 billion in 2027. These maturities can change through refinancing or early repayment.

Meanwhile, the company reported $103.7 billion in remaining performance obligations, representing future services under customer contracts. It expected to recognise approximately 41% of that revenue over the following 24 months and 39% over the subsequent 24 months.

The contracts provide substantial revenue visibility, but they represent neither cash already received nor guaranteed profit. Electricity, operating expenses, equipment replacement and future capital spending must be funded before determining the resources available to creditors.

Financial resilience therefore depends less on the headline value of the backlog than on timely delivery and conversion into cash.

Private credit: capital keeps flowing even as new lending slows

Data centre expansion also relies on private credit, including loans backed by computing equipment, infrastructure assets and commercial contracts.

According to Preqin data cited by Reuters on October 15, direct-lending funds raised $129.15 billion in the first nine months of 2026, up 35% year on year. Yet estimated third-quarter lending volume fell 28% to $49.77 billion, according to PitchBook/LCD.

The contrast concerns the wider direct-lending market, not AI infrastructure alone. It partly reflects competition from syndicated loan markets and a shortage of attractive transactions. Abundant capital can therefore coexist with caution towards certain borrowers, while also pressuring lenders to offer more accommodating terms elsewhere.

Alternative asset manager Carlyle has estimated that private credit could finance as much as $1 trillion in AI computing infrastructure. This is a forecast from a firm with a direct commercial interest in the sector, not a measure of loans already extended. The risk will depend in part on exposure concentration and collateral quality.

Electricity is becoming a condition of repayment

Financial markets are not the only factor shaping profitability. On October 18, Reuters reported, citing Britain's Telegraph newspaper, that Microsoft had withdrawn from an agreement involving an Nscale data centre in Loughton, Essex. The site faces grid connection problems, with some power capacity potentially unavailable until the 2030s.

Microsoft did not confirm the precise reason for its withdrawal. Grid constraints should therefore not be presented as the established and exclusive explanation. Nscale said it remained committed to the project and was exploring alternative power arrangements.

A funded data centre without sufficient electricity may have to delay revenue while property and financing expenses continue. For lenders, industrial delivery schedules are therefore as important as announced customer contracts.

Sovereign debt and AI: distinct financial pressures

The IMF's October 2026 Global Financial Stability Report identified two potential sources of market turbulence: a correction in AI-related assets and a disorderly rise in sovereign bond yields.

The mechanisms differ. Technology valuations depend on future earnings and the cost of capital. Government bond yields also reflect expected inflation, fiscal deficits, monetary policy and international capital flows.

On October 17, Christine Lagarde said rising long-term interest rates were tightening euro-area financing conditions. However, during the technology correction, some investors moved into benchmark sovereign bonds, temporarily pushing their yields lower.

A business can therefore face higher borrowing costs even as government yields decline: its own credit risk premium may widen by more than the benchmark rate falls. The IMF is also monitoring vulnerabilities among nonbank intermediaries, including concentrated or highly leveraged private credit structures.

Global AI investment extends beyond US operators

CoreWeave's funding problems do not describe the entire global market. In China, infrastructure growth involves industrial companies, telecommunications groups, private investors and public-sector-linked capital.

On October 12, TechNode reported that Infinigence AI had raised 2.3 billion yuan in Series C funding, bringing cumulative fundraising to 4.3 billion yuan. Separately, the South China Morning Post documented an approximately $403 million expansion of a China Mobile computing centre in Inner Mongolia, with reported electrical capacity of 120 megawatts.

These projects show that investment continues and electricity access shapes industrial decisions. They do not support a direct comparison of Chinese and American profitability: the evidence concerns fundraising and technical capacity in one case, and bond pricing and detailed financial statements in the other.

Some Infinigence operating figures originate from company disclosures. The South China Morning Post is owned by Alibaba, itself active in cloud computing and AI. These are relevant considerations when assessing sources, not proof of inaccurate reporting.

Cross-financing and end-user demand

Financial relationships between suppliers and customers raise another question. In July, Al Jazeera examined a potential $250 billion guarantee involving Nvidia and OpenAI to secure lease obligations associated with an Ohio data centre. It was neither a loan already disbursed nor a fully executed arrangement.

Economist Aleksandar Tomic, quoted by Al Jazeera, argued that suppliers financing their customers can make independent demand harder to measure. Michael Monaghan, founder of Founder's ETF, noted that cross-investment and commercial transactions can also accompany genuine economic activity.

The decisive issue is not the existence of such arrangements but whether customers can sustain spending through revenue generated by their own users without remaining dependent on repeated capital injections.

What could this mean for the wider economy?

Tighter AI credit does not automatically spread across the economy. Losses concentrated among some lenders could nevertheless reduce financing for unrelated businesses. Construction delays can also postpone orders for engineers, builders and electrical equipment suppliers.

Governments, especially developing economies already burdened by debt service, must balance digital infrastructure, electricity networks, transport and public services. Reuters reported debt service averaging 14% of government revenue among the economies discussed at IMF and World Bank meetings. The figure is not a cost directly caused by AI.

For households, potential effects include employment, borrowing conditions and electricity network costs. Their scale depends on actual financial exposures and how energy infrastructure costs are allocated.

Three possible outcomes

Gradual profitability

Facilities achieve adequate utilisation and revenue covers operating costs, technology replacement and debt. Some projects may fail without undermining the entire sector.

Selective consolidation

Lenders become more demanding, leading weaker operators to postpone investment, sell assets or find better-capitalised partners.

A broader credit correction

Disappointing revenue, costly refinancing and industrial delays produce simultaneous defaults. Concentrated losses among funds or lenders could restrict credit beyond AI. This remains a hypothetical scenario.

The decisive question: which obligations can future contracts actually fund?

CoreWeave's financial statements reveal three separate dimensions: cash already invested, liabilities created through equipment purchases without immediate payment, and contracted revenue expected over several years.

Estimated free cash flow of −$10.454 billion illustrates the intensity of cash investment during the first half of 2026. The $9.796 billion in equipment-related liabilities shows, without being mechanically additive to that figure, that financial commitments also arise outside immediate cash outflows. Meanwhile, $5.524 billion in unrestricted-classified cash and equivalents must be distinguished from $1.380 billion in restricted balances.

Growing computing demand does not guarantee profitability for every operator. Future results will need to show whether customer contracts turn into sufficient cash quickly enough to finance new equipment and meet maturing obligations. That timing gap, rather than the share-price fall alone, is what markets are increasingly testing.

FAQ

Has CoreWeave defaulted on its debt?

The reviewed information does not establish a default. Rising bond yields and CDS costs indicate greater perceived credit risk, not an actual failure to meet payment obligations.

Why does CoreWeave's investment exceed the cash it has spent?

Some equipment is acquired through liabilities or supplier financing without immediate payment. The SEC report discloses $9.796 billion in such equipment-related transactions during the first half of 2026. These amounts cannot simply be added to cash spending.

Does CoreWeave have $6.904 billion in unrestricted cash?

No. At June 30, that total comprised $5.524 billion in cash and equivalents not classified as restricted and $1.380 billion in restricted balances. The company also had undrawn financing facilities subject to separate conditions.