TL:DR

Moody’s Ratings has released its 2026 outlook for global data center infrastructure, projecting at least $3 trillion in investments over the next five years as capacity growth continues at double-digit rates.

The firm outlined capacity projections, estimating global data center electricity consumption to reach approximately 600 terawatt-hours (TWh) in 2026, a 14% increase from 525 TWh in 2025. The six largest hyperscalers in the US market are on track for $500 billion in capital expenditures in 2026, rising to $600 billion in 2027.

Most new capacity is pre-leased to large tech companies, limiting vacancy risk while increasing counterparty concentration. However, power limitations and regulatory opposition continue to constrain development, with local resistance mounting in some markets amid concerns about electricity and water consumption.

High demand for skilled labor, key commodities, and essential equipment is increasing construction costs while raising operating expenses for existing facilities.

The outlook also flags growing concerns about an AI investment bubble, as capital spending on computing infrastructure far outpaces revenue generation from AI applications.

Capital markets are adapting to finance rapid growth, with banks and institutional investors increasingly lending alongside one another during construction.

“I feel like every day there’s another announcement for power capacity,” John Medina, senior vice president at Moody’s Ratings, told Data Center Knowledge.

Power Infrastructure Emerges as Critical ConstraintPower Infrastructure Emerges as Critical Constraint

Power availability is now the key bottleneck limiting data center development in major markets. According to Moody’s analysis, the US Department of Energy projects AI workloads will account for 9% of total US electricity demand by 2030. Interconnection queue delays now routinely exceed the time required to construct the data center facilities themselves.

The industry is responding across three distinct time horizons. For immediate needs, developers are deploying on-site behind-the-meter power generation and relocating facilities to areas with available grid capacity.

Medium-term strategies focus on contracting for new generation assets and securing transmission agreements. Long-term investments target emerging technologies, including nuclear-powered small modular reactors.

Medina noted that power is not just a short-term consideration for data centers but a decades-long investment in secure, reliable energy.

Remote Locations Banking on Connectivity EvolutionRemote Locations Banking on Connectivity Evolution

Large-scale AI campuses exceeding 500 MW or 1 GW are being built in remote areas, including Southern Texas, Wisconsin, and rural Louisiana. These facilities support AI model training where low-cost, reliable power matters more than proximity to users.

Medina said these remote facilities raise questions about future reuse capabilities. The value proposition depends on whether technological advances reduce latency issues despite the physical distance from population centers.

Connectivity infrastructure improvements could change the calculus as inference workloads grow alongside AI product deployment.

“Technology is evolving, the value of that data center is going to change as well,” Medina said.

Adaptable Design Addresses Technology UncertaintyAdaptable Design Addresses Technology Uncertainty

The rapid evolution of AI has raised questions about whether facilities built today will support future requirements that today are not yet knowable.

Medina said developers are building data centers of the future that are designed to be adaptable. Regardless of whatever the specific technology direction might go with AI or any other application for that matter, the core infrastructure requirements remain constant, even as computing equipment changes.

“Whatever that internal need may be, from a compute perspective, you still need power, you still need the physical infrastructure, you still need access to cooling,” Medina said.

Future uncertainty is reflected in the risk allocation of lease structures. Increasingly, the standard is for triple net lease (NNN), where tenants have more responsibility for maintenance and operating costs.

“Whatever you want in the future, great, we'll do it for you, but we have no idea, and you have no idea, so let’s just upfront agree that nobody knows, and the tenant is going to pay for that,” Medina said, describing how contracts handle the uncertainty.

Diversification Strategy Addresses Multiple Risk VectorsDiversification Strategy Addresses Multiple Risk Vectors

Raj Joshi, senior vice president at Moody’s Ratings, said sophisticated operators are diversifying to manage technology risk across five dimensions: between customers, between workloads, between technologies, within technologies and between use cases.

The within-technology dimension addresses how capacity gets allocated between serving inference requests versus training, and how much to tie to any single vendor.

This diversification began to emerge in early 2024 as major spenders recognized their exposure to rapid technological shifts. Joshi said the approach provides some protection but doesn't eliminate risk entirely.

Capital Spending Sustainability Remains an Open QuestionCapital Spending Sustainability Remains an Open Question

The time lag between infrastructure investment and revenue generation spans two to three years. Companies must spend first before generating returns.

Joshi said no single metric will indicate whether current spending levels are sustainable. Key indicators include whether spending translates into growth and whether profitability characteristics improve.

Equity market reactions could trigger rapid adjustments.Equity market reactions could trigger rapid adjustments.

“If they start doubting something, then they’ll freeze,” Joshi said. “And the pushback from investors is generally so big that that causes these companies to kind of, you know, slow down spending.”

The trajectory of AI improvement and cost reduction remains critical. If progress plateaus or costs stop declining, the industry could face a reckoning over whether investment levels were justified.

“AI needs to continue to improve, and the costs need to come down,” Joshi said. “If that trajectory kind of flattens out or stops, then you have a problem in the industry.”

My Thoughts 💭My Thoughts 💭

Man this AI trade has to work out. If it fails the bust will be massive. $3T on power and data centers. But will the cost of AI ever be sustainable? I think these AI companies maybe have to switch to a price per prompt model if not what’s stopping heavy users spamming the prompt cause costs to increase?