Fig 2. Average size of newly built hyperscale/wholesale data centers in the US

Fig 3. Year-on-year % growth rate of estimated global data center revenue

Several signals stand out:

• Utilization has remained strong, with average utilization rising over time, showing that demand has kept pace with supply.
• Operational scale has increased, with consistent growth in average data center size as measured by MW of IT power.
• Revenue growth has shown strength, thanks to strong data center leasing by hyperscale IT firms and GPU-as-a-service providers to serve AI demand.

In addition, power—not physical space—has become the dominant constraint on growth, although local resistance to data center installations is increasingly a factor as well. AI workloads amplify this effect. Compared with traditional enterprise or cloud applications, AI training requires significantly higher power density, making it harder to obtain the power needed within required timelines.

Power, Capital, and Buildability as Credit Variables

As data centers scale, physical infrastructure constraints have moved from the periphery to the center of the investment case. Power availability, grid interconnection timelines, equipment lead times, access to labor and permitting complexity¹ increasingly determine whether projects can be delivered on schedule and within budget.

These constraints introduce credit‑relevant risks typically assessed using project finance frameworks:

• Execution risk, as delays in power delivery or construction can defer revenues and strain liquidity.
• Cost risk, as energy pricing volatility, grid upgrades, and backup power requirements affect capital intensity and operating margins.
• Concentration risk, as development clusters around a limited number of power‑rich regions.
• Resilience risk, as exposure to grid disruptions or extreme events grow with scale.

As a result, AI‑driven data center expansion behaves less like traditional technology investment and more like industrial infrastructure development—capital intensive, long dated, and sensitive to external systems.

Rising Default Correlation in an AI‑Driven System

As data centers become larger, more power‑intensive, and more interconnected, default risk is becoming increasingly correlated across the industry. Operators are exposed to many of the same underlying drivers: power availability, energy costs, supply chain constraints, permitting regimes, capital market conditions, and the pace of AI adoption.²

A regional power constraint, regulatory shift, financing shock, or slowdown in AI‑related demand can now affect multiple operators simultaneously. This creates the conditions for systemic stress rather than isolated credit events.

Traditional fundamentals-driven risk analysis can underestimate these shared vulnerabilities. What is increasingly required is a framework that links physical infrastructure realities with financial performance and macro conditions.

Translating Infrastructure Reality into Credit Insight

Understanding credit risk in the AI era requires a multi‑dimensional view—one that connects operational scale, energy exposure, and financial structure into a coherent analytical framework.

This is where an integrated approach becomes essential. A credit‑relevant assessment of AI infrastructure must address questions such as:

• How efficiently is capital being converted into energized, utilized capacity?
• Where are operators most exposed to power constraints or rising energy costs?
• How resilient are balance sheets under continued capital spending and tighter financing conditions?
• Where do geographic, customer, or utility dependencies create correlated risk?

S&P Global can support this type of analysis by bringing together complementary capabilities across the organization:

• S&P Global Energy contributes power‑ and energy‑market insight, including electricity price benchmarks, forward power curves, fuel linkages, emissions exposure, and regional grid dynamics—inputs that directly shape data center operating costs, siting decisions, and long‑term economic viability in power‑constrained regions.
• Customizable Advisory dashboards within S&P Global Market Intelligence (MI) solutions can provide industry‑level visibility across capacity expansion, utilization dynamics, geographic concentration, demand absorption, and financial intensity across operators and markets.
  Together, these indicators support a forward‑looking view of supply–demand balance and emerging sector‑level stress as data centers scale into capital‑intensive, infrastructure‑like assets.
• MI financials allow consistent comparison of balance‑sheet resilience, leverage, and operating performance across data center operators and related infrastructure players.
• S&P Global Market Intelligence’s (MI) Project finance and Corporate Digital Infrastructure Scorecards provide structured lenses to assess buildability, construction and completion risk, downside resilience, liquidity and refinancing exposure, and structural protections - dimensions that are becoming increasingly relevant as large AI‑driven data centers resemble infrastructure assets rather than traditional technology investments.
• MI Credit Analytics combines quantitative models with robust data to help assess the credit risk of public and private companies that rely on data centers.

By connecting these dimensions, market participants can move beyond narrative‑driven growth stories toward a disciplined, comprehensive and robust view of credit risk.

Looking Ahead

AI is transforming data centers from a real estate‑adjacent asset class into a core layer of economic infrastructure. The shift is already visible in operational scale, power dependence, and financial structure and it is likely to intensify further.

As this transformation continues, credit outcomes will increasingly be shaped by shared constraints rather than isolated company‑specific factors. Early identification of these risks before stress becomes visible in defaults or downgrades—can be critical for investors, lenders, and risk managers alike.

The next phase of the data center industry will not be defined solely by how much capacity is built, but by how resilient that capacity proves to be in an AI‑driven, power‑constrained world.