The AI infrastructure boom may lose its financing before it loses its customers.
This is the core judgment put forward by market commentator Bill Cara on October 30. Many data center projects must invest billions of dollars upfront before generating stable cash flow years later. The truly vulnerable link lies between when spending occurs and when revenue arrives.
A project can have customers, equipment, and signed contracts, yet still be unable to repay its immediate debts. If there is insufficient funding to cover the waiting period between construction, power connection, and delivery, demand growth cannot prevent a break in the funding chain.
Therefore, to judge whether AI investment is sustainable, beyond observing demand, one must also ask: can these projects continue to borrow at affordable terms when they need capital?
Beyond Big Tech Credit, Risk Is Sinking into the Financing Chain
In the past, large tech companies mainly relied on their own cash flow to support expansion. Today, the scale of AI infrastructure is changing this model.
Estimates cited by Cara show that hyperscalers such as Amazon, Alphabet, Meta, Microsoft, and Oracle may spend up to $940 billion in capex in 2026; JPMorgan's forecast for AI spending through 2030 reaches $5.5 trillion. These figures correspond to different forecast methodologies and reflect construction scale—they cannot be directly regarded as financing needs that have already been secured.
The financing structure is also changing. According to data cited by Cara, debt coverage of these companies' new expenditures rose from 9% in 2024 to 32% by mid-2026; their investment-grade bond issuance in 2026 is estimated at approximately $274 billion, more than double the full-year total for 2025.
Cara also cited UBS estimates that hyperscaler debt is approaching $800 billion; including leases, service contracts, and guarantees, total obligations are approximately $3.5 trillion. However, contractual obligations and interest-bearing debt are not the same concept, nor does it mean these payments will come due in a concentrated manner anytime soon.
More noteworthy is that infrastructure expansion has extended beyond the balance sheets of the giants. Microsoft's credit rating cannot tell investors whether a data center developer or equipment financing vehicle can survive a delay.
Participants in this chain bear different risks.
Hyperscalers have the strongest financial strength, but increased capital expenditures are also eroding some companies' free cash flow—the funds remaining after operating cash flow minus capital expenditures. New cloud service providers (Neoclouds) such as CoreWeave, Nebius, and Iren are expanding by providing GPU computing power services, with some relying on equipment, lease arrangements, or future contracts to obtain financing, and their financing terms differ markedly from those of the giants.
Developers need to bear construction overruns and delay costs before rent payments arrive; banks and private credit funds provide capital, but who ultimately bears the losses depends on specific contracts. Utility companies face the same question: if the expected data center electricity demand fails to materialize, who pays for the grid construction costs already invested?
Debt on company financial statements is only part of this financing network. Estimates cited by Cara show that the AI data center-related asset-backed securities (ABS) market is approximately $34 billion, with analysts expecting approximately $50 billion in new issuance that year. Such securities are backed by assets such as data center-related cash flows.
Another arrangement is Significant Risk Transfer (SRT) transactions: banks retain the loans while transferring part of the credit loss risk to external investors, thereby freeing up capacity to continue lending. Such transactions change who holds the risk; they do not make the risk disappear.
Cara listed institutions including Blue Owl, Brookfield, PIMCO, Blackstone, KKR, and BlackRock as participants in this financing network. But assessing risk still requires drilling down to specific funds, financing vehicles, and contracts—one cannot infer loss exposure merely from an institution's name.
No Customer Churn, Yet Financing Can Still Fail
Data center projects are simultaneously bound by multiple timelines: construction, power interconnection, customer acceptance, rent commencement, and debt maturity—these may not align.
If cash flow is delayed while debt service timelines do not shift accordingly, someone must fund the gap in between.
A hypothetical case can illustrate the impact of interest. A project carries $1 billion in debt; if the borrowing cost rises from 7% to 11%, annual interest increases from $70 million to $110 million.
Assuming the project generates $120 million in annual cash flow available for debt service, its interest coverage ratio would drop from approximately 1.71x to 1.09x—and this excludes principal repayment. With no customer attrition, merely rising financing costs nearly exhaust the buffer.
Refinancing may also hit capacity constraints. Assets that previously supported $1 billion in borrowing, if now only able to secure $700 million in loans, would require the borrower to inject $300 million in equity, or sell assets and restructure debt.
These are hypothetical calculations, not performance forecasts for any specific company, but they reveal three pressures that can reinforce each other: rising interest weakens debt service capacity, collateral depreciation compresses borrowing capacity, and project delays postpone cash receipt.
Even if revenue continues to grow, companies may still encounter funding gaps. The earliest signals may be a loan requiring additional equity, a financing commitment that fails to renew, or a delivery date that exceeds what the debt arrangement can withstand.
October market data cited by Cara noted that CoreWeave credit default swaps (CDS) quotes at one point exceeded 1,140 basis points, and its $2.75 billion bond maturing in 2031 yielded over 15.7%. CDS is a contract used to transfer default risk; 100 basis points equals 1 percentage point. If these quotes remain elevated, it means investors are demanding higher compensation for credit risk, but this alone does not constitute evidence of default.
A financing retreat also does not require creditors to coordinate with each other. As long as institutions simultaneously mark down collateral valuations, push back expected delivery dates, or reassess tenant credit, they may collectively reduce lending, tighten credit lines, and refuse rollovers.
Each exit of capital makes the next gap harder to fill. A project may halt because the next construction payment cannot be made—before it even enters liquidation.
Power and Regulatory Delays Ultimately Become Financial Problems
Data center financing hinges on a public policy question: who ultimately pays for grid expansion?
On October 9, 2026, the U.S. Federal Energy Regulatory Commission (FERC) rejected most of PJM grid operator's large-load interconnection proposals while accepting the large-load definition and registration arrangements. The rejected proposals included allowing some new large users to connect to the grid without corresponding new supply capacity, with a commitment to accept curtailment under specific circumstances.
This decision does not mean all data center projects are frozen, nor does it mean cost-sharing mechanisms have been comprehensively rejected. It reflects that unresolved issues remain among interconnection conditions, cost allocation, and the boundary between federal and state regulatory authority.
For highly leveraged projects, this uncertainty affects how much capital lenders are willing to provide and what returns they demand. The longer the project timeline, the more lenders need clarity on which costs are borne by developers, tenants, utilities, or ordinary ratepayers.
Even with clear regulatory rules, grid interconnection faces real-world bottlenecks. The scale of applications in the queue does not equal ultimate demand, but once construction timing becomes difficult to determine, financing arrangements become harder to lock in.
Texas offers an example. In August 2026, Governor Greg Abbott ordered reviews of certain proposed interconnection projects and paused related approvals during the review period. Subsequent local disclosures showed the review covered approximately 250 to 300 projects in the first batch of cluster study processes, primarily data centers; this was not a permanent ban on all projects.
For developers, projects need not be permanently prohibited—delays alone can create sufficient financial pressure. Land, equipment, and construction costs have already been spent, yet rental income cannot begin until power is delivered. Grid investment and regulatory proceedings advance over years, while lenders may reassess risk on a quarterly basis.
The gap between these two timelines ultimately requires real capital to bridge.
The investment pace of cloud giants themselves also affects this chain. Microsoft's capital expenditure guidance for calendar year 2026, disclosed at the end of October, was lowered from approximately $175 billion to approximately $170 billion—a reduction of less than 3%. This change is worth tracking but cannot be directly interpreted as weakening AI demand or customers canceling orders; shifts in spending timing also change annual figures.
Cara cited UBS forecasts stating that hyperscaler capital expenditure growth could slow from 76% in 2026 to 25% in 2027 and 6% in 2028. These are forecasts, not results that have already occurred.
Slowing growth, project delays, and order cancellations must be viewed separately. Slower growth may still correspond to rising absolute spending; delays mainly shift the timing of cash collection; only order cancellations directly erode the future revenue base. Their impact on financing differs.
But if major tenants slow expansion, projects built on expectations of future leasing may find financing harder to secure. Developers must choose between selling at a discount and holding on, and equipment aging and financing costs further influence that choice.
From industry losses to market shock, there are still five gates in between
Losses on AI credit do not automatically mean the entire financial market will plunge into crisis. For a systemic shock to form, losses must exceed holders' capacity to absorb them, or trigger forced sales of other assets through financing arrangements.
Five conditions need to be tested:
· Concentration: Are the same lenders concentrated in multiple projects?
· Leverage: Are lenders borrowing against these assets as collateral?
· Liquidity: Are investors retrieving cash faster than assets can be liquidated?
· Contract terms: Do losses trigger margin calls or borrowing capacity contraction?
· Replacement funding: Can exiting capital be replaced by new capital at an affordable price?
Only when repayment deadlines precede asset liquidation timelines can industry losses evolve into a liquidity event.
One channel through which risk transmits to stock markets is credit spreads—the extra portion of corporate borrowing rates relative to the risk-free benchmark. If AI-related financing continues to deteriorate and prompts fund providers to broadly contract risk exposure, financing conditions for other borrowers may also be affected. But such spillover requires evidence and cannot be directly inferred from one company's high yield.
Utility companies may also be affected. If grid expansion depends on data center demand that fails to materialize, cost recovery and burden allocation become problematic, weakening the defensive characteristics of some companies.
Whether private credit funds face redemptions or margin calls due to falling asset prices also depends on specific structures. Closed-end funds, open-end products, and leveraged financing vehicles face different liquidity constraints. Only by confirming the specific fund, contract, and trigger conditions can one judge whether there is pressure to forced-sell other assets.
Therefore, observing whether AI infrastructure can be sustained cannot rely solely on total order value. More valuable questions are: When does the debt mature, are lenders still paying construction costs as committed, how long can interest reserves last, can customers exit before rent commencement, who bears cost overruns, and what collateral clauses are triggered by chip depreciation or delivery delays.
These details ultimately point to the same question: which payment obligations come due before the cash flow arrives?
Cara holds a distinctly pessimistic view of the AI financing chain and judges that a financing retreat could drag down global stock markets. This is a risk-scenario judgment and should not be treated as an inevitable outcome. But the mechanism she proposes deserves attention: even if AI demand still exists, not all ownership and financing structures built around that demand can necessarily withstand delivery costs and waiting times.
For construction projects that depend on continuous financing, expansion can quickly grind to a halt as soon as multiple funding providers simultaneously refuse to pay the next tranche.
Welcome to join the official BlockBeats community:
Telegram Subscription Group: https://t.me/theblockbeats
Telegram Discussion Group: https://t.me/BlockBeats_App
Official Twitter Account: https://twitter.com/BlockBeatsAsia
