SECMarket Types & SectorsOctober 27, 20267 min read

The Power Advantage: Why Tighter Capital Could Strengthen the NeoCloud Model

AI infrastructure bears assume tighter financing automatically crushes CoreWeave, Nebius, and other NeoClouds. The scarce asset may be powered data-center capacity, not capital alone.

Download:PDFWordText

The standard bear case for CoreWeave, Nebius, and other NeoCloud providers is easy to repeat: AI spending slows, capital gets more expensive, data-center projects are canceled, and the companies that need the most money get crushed. That chain of reasoning may be too simple.

In This Article
  1. The contrarian thesis
  2. Power is harder to finance than to buy
  3. Why leasing can get more attractive
  4. What CoreWeave and Nebius disclose
  5. What could still break the thesis
  6. The filing signals to watch
  7. Bottom line

The contrarian thesis

When financing is cheap, a hyperscaler or GPU cloud provider can justify owning more of the stack: land, substations, buildings, cooling systems, networking, and GPUs. The company spends heavily up front, but it captures more of the economics if demand arrives on schedule.

When money gets more expensive, the calculation changes. A customer may still need compute capacity, but it becomes less willing to commit billions of dollars before utilization is visible. Leasing an already-powered facility, signing a build-to-suit agreement, or taking capacity from a specialist operator can preserve flexibility and move some construction risk to the provider.

That is the part of the bear case I question. Tighter capital is not automatically bearish for every infrastructure owner. It can be bearish for projects with no secured customer, no interconnection, and no path to cash flow. It may be supportive for operators that already control scarce power and can turn it into billable capacity.

The key distinction

Capital scarcity can reduce the number of viable new projects while increasing the strategic value of capacity that is already powered, permitted, and deployable. The outcome depends less on how much a company promises to build than on what it already controls and how quickly it can monetize it.

Power is harder to finance than to buy

The real bottleneck may not be money. It may be the physical sequence between a site and a live AI cluster: land, zoning, utility studies, transmission upgrades, substations, transformers, cooling, permits, and construction. Each link can take longer than the financing round that is supposed to fund it.

Berkeley Lab’s 2025 interconnection-queue analysis found that the median timeline from an interconnection request to a signed agreement was well over three years, while the path to commercial operation exceeded five years in the regions with available data.[1] That is not an overnight supply response. A customer can raise capital next quarter; it cannot guarantee that a 100 MW site will have grid access next quarter.

The queue is also a poor proxy for usable capacity. Berkeley Lab reported thousands of active projects seeking interconnection, but only a minority of projects that entered queues from 2000 through 2019 had reached commercial operation by the end of 2024.[2] A power reservation with a real utility path is not equivalent to a speculative site plan.

Why leasing can get more attractive

Leasing does not eliminate capital intensity. It changes who funds the asset, how risk is shared, and when the customer pays. A hyperscaler can replace part of its construction budget with a long-term capacity commitment. A NeoCloud can finance the facility against contracted revenue. A data-center landlord can build or convert the site while the customer supplies the demand signal.

This structure becomes more valuable when customers become selective. A company that wants to preserve cash may prefer a five- or ten-year capacity contract over owning every substation and building. It still takes on a fixed obligation, but it gains speed and avoids making a decade of infrastructure decisions before the workload mix is known.

The trade-off is concentration and take-or-pay risk. A leased site can produce attractive recurring revenue if the customer deploys on schedule. It can also become an expensive liability if the customer delays, cancels, or renegotiates. The bullish thesis therefore requires both scarce capacity and credible counterparties—not just a large megawatt number.

What CoreWeave and Nebius disclose

CoreWeave’s 2025 annual report said the company had more than 850 MW of active power across 43 data centers and approximately 3.1 GW of total contracted power expected to roll out over time.[3] The filing also states that CoreWeave works with data-center partners and may incur costs if it leases capacity that customers do not use. That is the NeoCloud model in plain language: secure capacity, deploy GPUs, and turn infrastructure access into a service rather than requiring every customer to build its own facility.

Nebius describes a similarly flexible stack. Its 2025 annual report said that, as of February 2026, it had contracted more than 2 GW of power, with approximately 170 MW active at December 31, 2025.[4] The company says its footprint combines owned sites, build-to-suit facilities where a partner has secured land and power, and co-location agreements that let it deploy compute in existing data centers. It also disclosed a five-year commercial agreement giving Microsoft access to dedicated GPU infrastructure in New Jersey.[5]

These disclosures do not prove that either stock will outperform. They do show why “capital gets tighter” does not map cleanly to “all NeoClouds lose.” A company with contracted power, multiple deployment paths, and customers willing to sign for capacity has more options than a company with only a proposed campus and a financing presentation.

QuestionWeak infrastructure storyStronger power-advantaged story
PowerConceptual site, uncertain interconnectionUtility path, contracted or connected capacity
CapitalMust raise before demand is contractedCan finance against leases, prepayments, or committed customers
DeploymentMulti-year greenfield buildExisting, converted, build-to-suit, or co-located capacity
Customer behaviorAssumes customers buy everything up frontOffers customers a way to preserve flexibility
Main riskFunding and execution before revenueConcentration, utilization, refinancing, and contract quality

What could still break the thesis

The bullish counterargument is not a free pass. If AI demand falls sharply, scarce power does not rescue an operator that has overleased capacity or bought GPUs at the wrong price. CoreWeave itself warns that underutilized leased capacity can create costs, and customer concentration can make cash flow dependent on a small number of counterparties.[6]

There is also a financing trap. A company can control power and still fail if it financed the build with short-term debt, needs constant equity issuance, or cannot get GPUs and cooling equipment delivered. Rising rates can make a good site uneconomic if the lease rate does not reset with the cost of capital.

Finally, “contracted power” is not the same as active revenue-producing capacity. Investors should separate power that is merely reserved from power that is connected, powered, commissioned, and being billed. The lag between those categories is where optimistic presentations can become disappointing financial statements.

The filing signals to watch

  1. Contracted versus active power: read the definitions and track the conversion rate from secured capacity to billable capacity.
  2. Customer concentration: identify whether one hyperscaler or AI customer represents most revenue, leases, or future commitments.
  3. Lease obligations: inspect undiscounted lease payments, take-or-pay terms, renewal rights, and commencement dates.
  4. Interconnection evidence: look for utility agreements, substations, permits, and construction milestones—not only megawatts in a presentation.
  5. Funding mix: compare customer prepayments, asset-backed debt, project finance, corporate debt, and equity issuance.
  6. GPU economics: monitor depreciation, useful lives, utilization, power costs, and the risk that a new chip generation makes current hardware less valuable.
  7. Dilution capacity: watch S-3 shelves, 424B5 supplements, ATM programs, convertible notes, and warrant repricing when the buildout outruns cash flow.

Bottom line

A tighter capital environment can be bearish for speculative AI infrastructure. It can also make existing power and deployable data-center capacity more valuable by discouraging customers from funding every project themselves. If AI demand remains strong, the scarce asset may be the powered site with a credible grid path—not the ability to announce another campus.

That is why I am not convinced that tighter financing automatically crushes the NeoClouds. The better question is which companies have already secured the physical bottleneck, which can convert it into revenue, and which are merely financing a story about future capacity. CoreWeave and Nebius still carry real leverage, execution, concentration, and utilization risks. But in a world where customers want compute and want to preserve capital, the lease-and-capacity model may become more—not less—important.

Sources

  1. Lawrence Berkeley National Laboratory — 2025 interconnection queue findings
  2. Lawrence Berkeley National Laboratory — Queued Up: 2025 Edition
  3. CoreWeave — 2025 Form 10-K
  4. Nebius Group — 2025 Form 20-F
  5. Nebius — Q2 2026 investor presentation
  6. CoreWeave — lease, utilization, and capacity risk disclosures

Track AI Infrastructure Dilution Risk

Follow shelves, ATM programs, PIPEs, convertibles, warrants, and other financing events across capital-intensive companies.

Start Monitoring →

Related Articles

Not investment advice. Research and analysis for information only; verify against the primary filings linked above.

More in SEC