Capital Formation for the AI Infrastructure Cycle

The artificial-intelligence investment cycle is moving beyond software budgets. Compute clusters, data centers, power generation, grid interconnection, cooling, networking and specialized semiconductors are turning AI into a capital-formation story with implications across public equity, investment-grade credit, private credit, infrastructure funds and private equity. For institutional investors, the opportunity is no longer simply exposure to AI demand. The question is how that demand moves through physical assets, financing structures and contractual cash flows.

The financing challenge is unusually broad because the AI stack combines short technology cycles with long-lived infrastructure. A server generation can change quickly while a substation, data-center shell or power contract may have a life measured in decades. That mismatch places more weight on tenant quality, contract design, residual values, power certainty and refinancing assumptions.

01 / The scale of the buildout

The International Energy Agency reported in 2026 that capital expenditure by five of the largest technology companies exceeded US$400 billion in 2025 and was expected to rise materially in 2026. The same analysis describes rapid expansion of AI-focused data-center capacity. At that scale, the financing question changes: technology companies may be the demand anchor, but the physical assets supporting that demand sit across utilities, developers, equipment suppliers, real-estate structures and multiple layers of capital.

Goldman Sachs Research has separately argued that private markets are likely to play a larger role in data-center financing as expected AI and data-center spending extends into the trillions of dollars through the end of the decade. Individual forecasts will change, but the direction is strategically important. The AI stack increasingly requires financing capacity beyond retained corporate cash flow alone.

This also means the market should not be treated as one homogeneous “AI infrastructure” category. Hyperscale facilities, colocation capacity, powered land, utility upgrades, merchant generation, contracted generation, fiber, networking equipment and semiconductor supply each carry different risk, duration and return characteristics.

02 / Four financing layers

Corporate balance sheets

Large technology platforms remain the most visible source of AI infrastructure spending. Their credit quality, free-cash-flow generation and public-market access provide an important anchor. Yet concentration can become a constraint for lenders, developers and investors with issuer, tenant or sector limits. Counterparty quality must therefore be evaluated alongside counterparty concentration.

Infrastructure and project capital

Data centers combine long-lived physical assets with technology-cycle risk. Underwriting requires attention to tenancy, contract duration, power availability, development rights, construction risk, equipment timing, residual asset value and the credit profile of counterparties. Project economics can change materially if energization is delayed or if equipment costs rise before contracted revenue begins.

Private credit

Private lenders can address bespoke construction, acquisition and transitional financing needs. The opportunity is paired with risks around leverage, refinancing, utilization assumptions, collateral quality and the value of specialized assets if technical or demand conditions change. Covenant design and information rights become particularly important where utilization ramps over time.

Private equity and strategic capital

Equity capital absorbs operating execution and terminal-value risk. The most defensible theses connect asset scarcity or operating advantage to measurable cash-flow outcomes rather than treating AI demand as a universal valuation premium. Strategic investors may accept different return profiles when an asset also secures supply, distribution or infrastructure access.

03 / The capital stack is part of the competitive strategy

In capital-intensive technology markets, financing capability can become an operating advantage. A developer with reliable access to low-cost capital may be able to secure land, equipment or power before a smaller competitor. A borrower able to match debt duration to contracted cash flows may preserve more equity value. A platform with a diversified customer base can potentially access a broader set of financing providers than one dependent on a single tenant.

Investors should therefore examine the capital stack as part of market structure. Questions include who bears construction risk, when leases begin, what happens if power delivery slips, whether debt amortization matches contract duration, how much refinancing is assumed, and what residual value remains if compute density or cooling architecture changes.

04 / Bottlenecks become underwriting variables

Power is increasingly central. A data-center project can have land, capital and customer demand but still face years of interconnection or generation constraints. Equipment lead times, permitting, construction labor, transformer availability and cooling requirements can also change project economics. In an infrastructure cycle, the constraint can be more valuable than the headline technology.

A useful underwriting model separates controllable execution from system-level constraints. Construction scheduling, procurement and lease negotiation may be controllable at the project level. Grid queues, regional transmission constraints and policy changes may not be. The investment case should identify which assumptions depend on third parties and how much downside protection exists if those assumptions move.

05 / Unit economics below the headline capex

Large aggregate capital-expenditure numbers do not by themselves establish attractive project returns. Investors need a unit view: cost per delivered megawatt, cost per rack or compute unit, contracted revenue, escalation terms, operating expense, maintenance capex, power pass-throughs, financing costs and utilization.

The economics can look very different across a fully pre-leased facility, a speculative development, an expansion attached to an existing campus, or infrastructure built around a single strategic customer. A durable research framework should make those differences explicit rather than relying on total market growth.

06 / Scenario analysis for lenders and equity investors

New York AI™ uses scenario analysis to examine three broad cases. In a base case, contracted demand ramps on schedule and financing remains available at expected terms. In a constrained case, power or construction delays push revenue outward while interest and carrying costs continue. In a technology-reset case, customer architecture changes faster than expected, reducing the value of some specialized capacity or requiring additional capex.

The purpose is not to assign false precision to probability estimates. It is to identify which assumptions drive debt-service coverage, equity returns and refinancing needs. If a project only works under perfect utilization and lower future financing costs, the risk is fundamentally different from one supported by contracted cash flow and conservative leverage.

07 / What investors should measure

New York AI™ frames the AI-infrastructure cycle around seven variables:

• Contracted demand — committed revenue, customer concentration, termination rights and ramp schedule.
• Delivered power — energization timing, redundancy, power price and dependency on grid or generation projects.
• All-in cost of capital — debt pricing, fees, hedging, equity requirement and refinancing assumptions.
• Construction execution — equipment availability, labor, permitting, contingencies and schedule discipline.
• Asset adaptability — the ability to support future density, cooling and hardware requirements.
• Operating conversion — how contracted capacity translates into cash flow after power, maintenance and operating costs.
• Residual economics — what remains valuable after the initial technology cycle or anchor contract ends.

08 / Research conclusion

The AI economy is creating a new intersection between technology underwriting and infrastructure finance. The central research question is no longer simply how quickly AI use grows. It is how the cost, ownership, duration and financing of the physical stack affect returns across the capital structure.

For institutional capital, the strongest opportunities are likely to be those where technology demand is translated into contractual economics, scarce infrastructure access and financing structures that remain resilient under slower utilization or higher capital costs. Growth matters, but the quality of the capital structure determines how much of that growth reaches investors.

09 / Selected references

• International Energy Agency — Key Questions on Energy and AI (2026)
• Goldman Sachs Research — Private Markets Are Expected to Have a Growing Role in Data Center Financing (June 2026)
• Goldman Sachs Research — Global AI Investment Is Forecast to Exceed $1 Trillion in 2026 (August 2026)