2026 Outlook: Capacity growth remains robust as tenants prioritize speed to market

Demand for data center capacity to support AI, cloud computing, and internet services will continue to rise sharply in 2026. Most of this new capacity is pre-leased to large tech companies, or hyperscalers, limiting the risk of introducing a surplus of unoccupied capacity into the market, while increasing counterparty concentration risk.

• Double-digit capacity growth continues with new energized capacity quickly monetized by tenants: The race to build new data center capacity remains in its early stages with robust capacity growth poised to continue globally over the next 12 to 18 months. The International Energy Agency projects that global data center capacity as measured by electricity consumption will reach about 600TWh in 2026, up 14 percent from an estimated 525TWh in 2025, which is a 20 percent increase from 2024 actual consumption of 436TWh.
• Regulatory risk and power limitations continue to constrain development: Local opposition to new data centers has increased in some markets amid mounting public concerns about their power and water consumption. By contrast, regions with supportive laws are likely to continue to experience new development, while some are making changes to their regulatory frameworks to encourage new data center development for AI.
• Tenants take on more risk to facilitate expedited completion times amid rising construction risks: Some tenants are willing to share certain construction delivery risks that they did not take on historically to accelerate completion, like exempting power and essential utilities availability from completion requirements. High global demand for skilled labor, key commodities and essential equipment increases the cost of building new data centers, while also raising operating costs for existing data centers.
• Capital markets adapt to finance rapid hyperscale data center growth; creditworthy counterparties still key: As the number and size of data centers expand, the amount and diversity of development capital needed has increased. Banks will continue to play a major role, with institutional investors increasingly lending alongside them during construction. Long-term leases for new hyperscale data centers will continue to be supported by the credit of the large, big-tech companies, despite the capacity being primarily used by a start-up AI company like OpenAI or Anthropic.
• High GPU costs prompt developers to look for alternative financing options: Users and owners of GPUs are likely to seek more alternative financing sources. Access to sources of new capital is crucial because of the high cost of computing equipment.

Double-digit capacity growth continues with new capacity quickly monetized by tenants

The race to energize new data center capacity remains in its early stages with robust capacity growth poised to continue globally for at least the next 12 to 18 months.

The International Energy Agency projects that global data center capacity as measured by electricity consumption will reach about 600TWh in 2026, up 14 percent from an estimated 525TWh in 2025, which is a 20 percent increase from the actual 436TWh consumed in 2024 (see Exhibit 1).

Even larger hyperscale data center projects with capacity levels exceeding 300MW will begin coming online in 2026, exponentially increasing capacity in some markets while also creating new markets. These new massive facilities will further accelerate the AI and cloud race.

Most new data center capacity coming online will be deployed for cloud services by hyperscalers like Alphabet Inc.’s (Aa2 stable) Google subsidiary, Microsoft Corporation (Aaa stable), Amazon.com Inc.’s (A1 positive) Amazon Web Services subsidiary and Oracle Corporation (Baa2 negative), which have pre-leased the capacity to meet their rising customer backlogs for new computing demand.

As a result, most of this new capacity should generate revenue and cash flow for the tenants. New AI-focused data center capacity will enable the development and deployment of new AI models and products as this nascent industry continues to scale up in tandem with the growth in available high-performance computing capacity.

The path toward revenue generation will become increasingly important in the growing AI ecosystem to deter the rising questions around an ‘AI bubble.’

Capital investments by six hyperscalers in the US (Aa1 stable) – Microsoft, Amazon, Alphabet, Oracle, Meta Platforms Inc. (Aa3 stable) and CoreWeave, Inc. (Ba3 stable) – approached $400 billion in 2025 and are on track to reach $500 billion in 2026 and $600 billion in 2027.

These large investments by strong big tech companies continue to fuel the rapid growth across the entire data center value chain and AI ecosystem, supporting our forecast for global data center related investments of at least $3 trillion over the next five years (see Exhibit 2). Moreover, the US cost assumptions we applied to this global forecast may prove to be excessively high because APAC, which has lower construction costs, accounts for half of the demand growth.

Capacity growth will be robust across all regions

North America

Data center development continues to be dominated by new data center projects across the largest US markets. Capacity coming online in 2026 will remain mostly near tier one markets, yet new markets can emerge rapidly given the scale of new developments and the tendency of hyperscalers to cluster their data center developments together like in Columbus, OH (Aaa stable), Atlanta (Aa1 stable) and the Texas (Aaa stable) cities of Dallas (A1 negative), Fort Worth (Aa3 stable), San Antonio (Aaa stable) and Austin (Aa1 stable).

Sharp growth is also occurring in remote regions of Texas and the Midwest, where the AI factories planned could eventually dwarf the size of some tier one markets in the country on a capacity basis. New developments in Canada (mostly in the greater metropolitan areas of Toronto, Vancouver and Montreal) are moving forward now given the more supportive development stance of federal and provincial governments and new investments by the same hyperscalers investing in the US.

Asia Pacific

Hyperscale, co-location and Edge data centers in the region have combined capacity of about 16GW, which we expect will more than double over the next five years. These data centers make up about 60-65 percent of APAC’s total data center capacity, with enterprise data centers representing the rest.

This reflects a compound annual growth rate (CAGR) of about 20 percent through 2030, and we estimate this growth will require investment of up to $900 billion. China (A1 negative), APAC’s largest data center market, will continue to grow but mostly within its own ecosystem through large tech companies, domestic developers, operators, and users. Overbuilding could be a concern in China as supply-demand imbalances and rapid technology shifts could lead to underutilized data center capacity.

Europe

By the end of 2025, data center developers in Europe planned to deliver a record 871MW of new capacity, up 34 percent from 2024. Nearly three-quarters of this additional capacity is concentrated in the five largest markets, known as FLAPD: Frankfurt, London, Amsterdam, Paris (Aa3 negative) and Dublin. We anticipate roughly 500MW of new capacity in the FLAPD markets in both 2026 and 2027.

Europe is ramping up its AI infrastructure by adding six new AI factories, bringing the total to 19 facilities across 16 EU (Aaa stable) member states. These new sites will join 13 existing ones, forming an interconnected high-performance computing network designed to accelerate the development, testing and scaling of AI models and applications. This expansion is a crucial part of the EU’s strategy to strengthen digital sovereignty and reduce dependence on non-European technology ecosystems.

Latin America

Installed IT load in the region reached 1.36GW in 2025, with nearly 1GW under construction, and vacancy rates in major hubs such as São Paulo (Ba1 stable), Querétaro (Baa2 negative), and Santiago remain below five percent, driving lease prices to record highs. Governments are actively courting hyperscalers through fiscal incentives and streamlined permitting.

Brazil’s (Ba1 stable) ReData regime, Mexico’s (Baa2 negative) Plan México, Chile’s (A2 stable) National Data Center Plan, and Argentina’s (Caa1 stable) RIGI framework underpin a wave of hyperscale projects. Structural challenges persist, despite the abundance of renewable energy. Water scarcity in Santiago and Querétaro, coupled with power grid limitations, threatens project timelines. Brazil’s pending legislation on AI training adds complexity for AI-focused infrastructure investments.

Regulatory risk and power limitations continue to constrain development

Regulatory risk and power grid limitations constrain new development in most regions, while supportive regulation and ready access to power facilitates new developments in other regions. Local opposition to new data centers has increased in some markets amid mounting public concerns about their prodigious consumption of electricity and water and their impact on utility bills.

By contrast, regions with supportive laws are likely to continue to experience new development, while some are making changes to their regulatory frameworks to encourage new data center development for AI, like the ReData law in Brazil or US President Donald Trump’s recent executive order aimed at preempting state-level AI regulation to avoid a patchwork of rules that increase compliance costs.

Access to power remains a key constraint for new data centers in most markets, with utilities and independent power producers working overtime to meet this once-in-a-generation surge in power demand. As we expected, data centers are making greater use of onsite behind-the-meter power solutions or adjacent power generation assets wholly dedicated to their needs as they increase in size.

While behind-the-meter solutions can expedite access to new power generation, they often lead to the construction of more capacity than is needed to prepare for such contingencies as unforeseen increases in power requirements or maintenance work that temporarily take individual turbines or power plants out of service. Protracted delays in securing power grid connections often exceed the time it takes to build a data center, forcing some new data centers to use temporary onsite power until they can connect to the grid.

US

More new power generation assets like natural gas-fired plants and renewable energy projects will be constructed, as well as behind-the-meter power generation sources collocated near the data centers that they are built to supply. Many of these natural gas power plants will be located near major sources of natural gas, like the Permian Basin in Texas, the Gulf Coast and shale gas deposits in Pennsylvania (Aa2 stable) and Ohio.

Utilities will continue to work with data center projects all over the country to supply their high demand for power while minimizing the impact on existing power customers. On 18 December, the Federal Energy Regulatory Commission directed grid operator PJM Interconnection L.L.C. (Aa2 stable) to establish transparent rules to facilitate service of AI driven data centers and other large loads co-located with generating facilities, which should help independent power producers increase investment in the region.

Asia Pacific

Rapid data center growth will require substantial investment by utilities in power generation, storage and grid infrastructure, amid broader challenges of growth, decarbonization and affordability. In the region’s smaller power markets, like Malaysia and Australia, data center related power demand may require a sharp increase in investment.

For larger markets like China and India, the incremental power demand from data centers will be added to existing large power development pipelines to meet broader economic growth. Fossil fuels remain the primary energy source for data centers in the region, because of the need for reliable, affordable and continuous power supply. Nuclear power is also likely to play a role in meeting this growing demand in some countries.

Europe

Power grid constraints play a significant role in the FLAPD markets, with limitations on power availability, strict planning policies and a scarcity of suitable land. Electricity prices in Europe are often higher than in other regions, although renewables account for a higher proportion of generation than other regions.

Faced with challenges in meeting demand within Europe’s FLAPD markets, developers are scaling up in Nordic countries like Sweden (Aaa stable), Norway (Aaa stable) and Iceland (A1 stable), as well as in southern European countries such as Spain (A3 stable), Portugal (A3 stable) and Italy (Baa2 stable).

New AI campuses or AI factories typically exceed 500MWs or 1GW and are often located further away from population centers, resulting in potentially fewer reuse capabilities using current technology. These facilities are built where reliable and low-cost power is available long-term.

Rapid construction, no local opposition and the ability to swiftly energize this new capacity drives the value of these more remote locations to the big-tech tenants that pre-lease the new capacity. Over time the distance to end customers may become less important as new technologies help reduce latency despite physical location. This would increase the value of these remote locations, especially as inferencing ramps up in the coming years as more AI products are deployed.

Tenants take on more risk to facilitate expedited completion times amid rising construction risks

Even as projects increase in size, developers are consistently accelerating project construction schedules to meet their big tech tenants’ focus on shortening their speed to market. To keep developers and contractors focused on expedited delivery, some tenants are willing to share certain construction delivery risks they did not take on historically, like exempting power and essential utilities availability from completion requirements or sharing in the risk of unexpected events.

These positive risk-allocation changes during construction help balance the risks associated with accelerated delivery as many of these new data centers with rapid schedules may be delivered late despite experience in the sector, construction crews working double shifts and advanced procurement of key pieces of equipment.

High global demand for skilled labor, key commodities and essential equipment continues to pressure the total construction costs of new data centers, while also raising operating costs for existing data centers.

Miners of copper and rare earth metals and manufacturers of essential cooling and power related equipment are cautiously ramping up production to meet demand from data centers. But the additional production remains insufficient to moderate price increases in 2026. As a result, newer data centers cost more to build than older ones located in similar markets, which increases tenant lease rates.

The sharp rise in hyperscale data center lease rates in key US markets over the last couple of years reflects both high demand and the higher costs of the newer facilities coming online. In northern Virginia, the largest data center market in the world, leases for hyperscale data centers with more than 4MW of capacity increased to a range of $130 to $190 per kilowatt per month in 2025 from a range of $110 to $150 per kilowatt in 2024, according to datacenterHawk via FactSet.

That represents a year-over-year increase of 18 percent to 27 percent and follows a 13.5 percent average increase for hyperscale leases in 2024 from 2023 levels. Corresponding data for the rapidly growing Atlanta market indicates a nearly 39 percent cumulative increase in average pricing from 2023 to 2025. Data center lease rates typically increase by two percent to 3.5 percent annually, so the cost of renewing hyperscale leases for data centers that come online between 2025 and 2027 will be less than those that come online thereafter.

Despite construction delays and higher costs, current market dynamics are leading tenants to accept the facilities when available rather than exercising their often heavy-handed rights in their lease. This could change once market supply and demand are more in balance, which we do not expect for several years in most markets.

We expect more operating problems to be reported over time as well, because of the sheer increase in the number of data centers and the expanding ranks of newer, less experienced operators in the market. Over time, additional supply chain capacity could help moderate price increases and help improve the economics of data center development, operations and rehabilitation.

Capital markets adapt to finance rapid data center growth; creditworthy counterparties still key

As the number and size of data centers continue to expand, the amount and diversity of development capital required has increased as well. Most of the development capital to date has come from project finance, corporate bank loans and private capital, as well as developer equity, tenant equity or both. Banks will continue to play a prominent role in financing new data centers under construction, with institutional investors increasingly lending alongside the banks given the large amount of capital needed.

As projects increase in scale and the pace of construction accelerates, lease terms and tenors are adjusting to balance the risk allocation to help ensure unknown and higher risks are borne by a well-capitalized or highly experienced party.

Start-up AI companies like OpenAI and Anthropic are large users of data center capacity and typically procure the capacity and costly internal computing equipment with the support from an investor or a creditworthy partner. We expect this relationship to continue, with long-term leases for new hyperscale data centers likely to be supported directly or indirectly by the credit of one of the large, big tech companies rather than solely by the credit of the start-up AI company that is the primary user of the new capacity.

For example, OpenAI has signed sizeable contracts for gigawatts of new data center capacity and other hard infrastructure assets while it continues to grow its revenue base. Most of the financing for these assets is supported by long-term leases with a hyperscaler like Microsoft and, increasingly, Oracle. While the financings benefit from a creditworthy counterparty, OpenAI’s growing presence in the AI ecosystem poses a growing credit risk depending on its success.

We expect to see more crypto miner financings with credit positive structural features and new large scale project financings with amortizing debt and no lease renewal risk.

Some companies like Flash Compute LLC (Ba3 stable), which is owned by a joint venture between bitcoin mining company Terawulf Inc. and AI cloud company Fluidstack Ltd.; WULF Compute LLC (Ba2 stable), a wholly owned indirect subsidiary of TeraWulf; and Cipher Compute LLC (Ba3 stable), a wholly owned indirect subsidiary of bitcoin mining company Cipher Mining Inc., are developing data center capacity for ‘neocloud’ providers that specialize in serving AI computing workloads.

Bitcoin miners can accelerate a hyperscaler’s speed to market by using their existing access to the power grid. But their data center projects are often located in non-core markets where the low cost of power may ultimately prove to be an insufficient enticement for data center tenants long-term. While valuable for AI model training today, their ability to be used for inferencing may be more limited until new technologies help reduce latency.

While a growing number of long-term investors are more comfortable with taking construction and ramp-up risks, they typically do not take on long-term lease renewal risk given their preference for steady, long-term predictable cash flow. To balance the uncertainty of a rapidly growing market, an increasing number of new financings are being structured with the ability to fully repay their construction debt within their initial lease term without any extensions or renewals.

Initial lease terms will continue to extend beyond 15 years to allow more time to repay the initial financing as projects increase in scale. This lowers the credit risk of the data center project financing compared to data centers that are exposed to lease renewal risk to repay their development financing. Lower credit risk enables a data center project to access a wider base of investors, like certain long-term private credit lenders or investment-grade focused infrastructure funds.

As more US data center projects are completed, their initial financings will begin to be refinanced in the asset-backed securities (ABS), commercial mortgage-backed securities (CMBS) and private credit markets. New financings will increase in size and concentration as the ABS and CMBS markets reach new highs in 2026 after record levels of issuance in 2025. Single asset/single borrower (SASB) data center CMBS debt issuance in the US totaled about $11 billion in 2025, surging from just $3 billion in 2024.

We expect 2026 data center SASB CMBS issuance to grow further, but institutional investors’ limitations on property type and tenant concentrations may dictate the extent of the expansion. In the US ABS market, about $15 billion was issued in 2025 and US issuance volume will likely grow considerably in 2026, driven by the refinancing of data center construction loans and of ABS term notes reaching their refinance dates.

Private equity-backed firms will remain the most common sponsors of data center ABS transactions. We also expect to see new data center ABS financings in Europe and potentially the first one in the Asia-Pacific region in 2026 as global data center developer owners replicate their successful US data center financings elsewhere.

High GPU costs prompt developers to look for alternative financing options

Users and owners of GPUs, the single largest capital expense in the construction of AI hyperscale data centers, are likely to seek alternative financing options and sources beyond traditional banks or their own balance sheets. Access to sources of new capital is crucial because of the high cost of computing equipment used in AI data centers.

We understand that the tenants typically replace their internal servers every four years for the most cutting-edge uses, like large language model training. As a result, an AI data center’s computing equipment, including GPUs, can cost multiples of the initial cost of building the facility.

GPU-as-a-service providers that contract with large tech companies and have the most experience operating the newest GPUs are increasingly able to use them beyond their initial life cycle. If older GPUs can demonstrate a longer useful life, they may be able to extend and widen their financing options.

Read more about Moody’s thought leadership in the data center space here