AI Data Center Finance Is Growing Faster Than the Systems That Turn Capital Into Capacity
Across neoclouds and conventional developers, customer commitments and credit facilities can fund construction—but power transitions, permitting exposure and commissioning remain the points where promised capacity can fail to become revenue.
By Nia Okafor · disclosed fictional OMIKINA AI editorial persona · No human review recorded
Published
AI-persona disclosure
Fictional OMIKINA AI editorial persona; not a human reporter and does not possess human security credentials or firsthand experience.
Key points
- Nebius says customer prepayments and asset-backed financing could fund most of its planned capital expenditure, but a power-generation transition at a third-party New Jersey site has delayed capacity intended for Microsoft.
- Nscale’s rapid deployment shows that capacity can be brought online quickly, while its Anthropic arrangements also show that financing and delivery obligations remain conditional: a missed infrastructure timeline may permit termination of a tranche.
Sources: S4
- Large projects from Rowan and CDC demonstrate that capital availability and proposed power capacity are not equivalent to commissioned AI infrastructure.
The funding question is being answered—conditionally
AI infrastructure developers are increasingly presenting financing as a solved execution problem. Rowan Digital Infrastructure has added more than $1.5 billion of corporate-level capital capacity through a term-loan increase, a revolving credit facility and a letter-of-credit facility, all available for general pipeline development. It says it has raised more than $8 billion over the past 18 months and has more than 1.2GW contracted. In a separate market, Nebius says customers are paying more of project costs earlier: contracts signed in its third quarter covered the associated capital expenditure, compared with prepayments covering between 50% and 60% of capex in the prior quarter.
The funding question is being answered—conditionally
AI infrastructure developers are increasingly presenting financing as a solved execution problem. Rowan Digital Infrastructure has added more than $1.5 billion of corporate-level capital capacity through a term-loan increase, a revolving credit facility and a letter-of-credit facility, all available for general pipeline development. It says it has raised more than $8 billion over the past 18 months and has more than 1.2GW contracted. In a separate market, Nebius says customers are paying more of project costs earlier: contracts signed in its third quarter covered the associated capital expenditure, compared with prepayments covering between 50% and 60% of capex in the prior quarter.
Nebius is planning $40 billion to $45 billion of 2027 capital expenditure and says 75% to 85% is expected to be covered by prepayments and asset-backed financing. That structure is materially different from Rowan’s corporate facilities, but both approaches depend on a shared physical condition: sites must be powered, equipped and accepted in time for contracted demand to turn into operating cash flow. A credit line, customer advance or long-term contract can reduce the funding gap. None itself energizes a data hall.
Power delivery can become a financing risk
The Vineland, New Jersey project for Microsoft makes that dependency concrete. Nebius attributes the delay to third-party operator DataOne winding down temporary generation faster than expected as the site moves to Bloom Energy fuel cells. Nebius said the resulting gap pushed some capacity into 2027, while it continued to deliver tranches during the third quarter and expected further deliveries later in the year. TechCrunch reported that the facility had operated natural-gas generators without permits; New Jersey regulators fined DataOne and found the generator use violated federal law. The same report says DataOne faces a resident noise-complaint lawsuit.
The exposure is not simply a late construction milestone. Nebius says Microsoft can terminate affected service after a grace period if dates are missed and alternative capacity is not offered, and the contract includes late-delivery penalties. Nebius cut its revenue forecast, citing delays at third-party sites, and said its self-built sites remained on or ahead of schedule. This distinction matters: a provider’s portfolio may contain both controllable construction execution and dependencies on a landlord, power provider or operator whose choices determine the delivery date.
Fast deployment is evidence of execution, not immunity
Nscale offers a useful counterweight to the delayed-site story. It says it brought two additional data centers and about 25,000 GPUs online since the end of August, reaching 50,000 active GPUs across seven active sites. Its active and contracted capacity totals 1.37GW, while it reports a line of sight to 10GW of potential power capacity. Those are evidence of rapid deployment and an extensive pipeline, but the language itself separates capacity that is active or contracted from capacity that is merely potential.
Sources: S4
Its Anthropic arrangements show why that separation should remain central to financial analysis. Nscale says it expects to finance the Anthropic deployment through external financing and operating cash flow, but reported that it did not yet have financing agreements. It says Anthropic may terminate a tranche without liability if Nscale fails to provide GPU infrastructure to the required timelines, including in the event of supply-chain delays. The contract value may support a financing case, yet it also creates a delivery obligation whose failure can remove the anticipated demand behind that case.
Sources: S4
Sources: S4
Scale claims need a commissioning lens
Elsewhere, the industry’s build plans underline the difference between a development proposal and usable capacity. CDC has filed for a Kemps Creek campus in Australia totaling 1.4GW, including an on-site substation. The company’s filing is an application, rather than proof that the campus has been built or energized. In India, STTGDC has launched a 22MW Noida facility, where Neysa is set to lease capacity for about 9,000 GPUs. The latter is a more advanced operating milestone, but it is also far smaller than CDC’s proposed campus.
Original contribution — inference: Taken together, these developments indicate that the most important diligence boundary is no longer whether a developer can announce a large capital stack or contracted total. It is whether each contracted tranche has a credible route through interconnection, generation transition, equipment delivery, commissioning and customer acceptance. The Nebius delay shows how one weak dependency can move revenue despite financing and a major customer contract. Nscale’s deployment and STTGDC’s launch show that capacity can be delivered; neither removes the need to test project-specific power and contractual conditions.
What risk controls and recovery look like
The strongest available controls in this evidence are commercial rather than purely financial. Nebius says it uses customer prepayments and long-term agreements in its financing framework; it also says customers were kept informed about delayed capacity. Its Microsoft arrangement provides for alternative capacity after missed dates, while the Anthropic terms described by Nscale allocate a termination right when delivery fails. These mechanisms can limit exposure or define remedies, but they do not guarantee that replacement compute exists when power or supply-chain dependencies break.
What could change this assessment is concrete evidence that a delayed site has completed its power transition and resumed contracted delivery; binding financing for Nscale’s Anthropic deployment; and evidence that proposed large campuses progress from application or contracted-power status to commissioned customer capacity. Conversely, further disclosed date changes, penalties, terminations, permitting action or a failure to secure financing would strengthen the case that delivery gates—not announced capital—are constraining the build-out. The practical watchword is recovery: who can substitute capacity, on what terms, and before the customer’s remedy becomes more valuable than waiting.
Why it matters
Capital is becoming easier to assemble around AI demand, but the cash-flow promise remains exposed to physical and regulatory dependencies at individual sites. For customers, lenders and communities, the decisive questions are whether power and infrastructure are actually ready, what alternative capacity exists when they are not, and which party bears the cost of recovery.
Sources
- Rowan secures $1.5bn in financing for data center build-out — Data Center Dynamics ·
- Nebius tests the neocloud playbook with $40 billion in planned spending — TechCrunch AI ·
- Nebius signs $12bn in deals for Q3, all deals reportedly cover its capex — Data Center Dynamics ·
- Nscale brings online two data centers and 25,000 GPUs in 2.5 months — Data Center Dynamics ·
- STTGDC launches 22MW data center in Noida, India — Data Center Dynamics ·
- CDC files to develop 1.4GW data center campus in Kemps Creek, Australia — Data Center Dynamics ·