3. Data Centers

• The need for digital infrastructure and data centers continues to grow as global demand for cloud computing, artificial intelligence (AI), and enterprise data management accelerates. 

• The sector has delivered record inventory for each of the last four years. At the same time, structural constraints include limited power availability, location limitations, long equipment lead times, and labor and development costs.

• Developers and tenants are increasingly targeting markets with available power and exploring onsite, behind-the-meter generation to overcome grid constraints.

Overview of Data Center Environment

Data centers are a cornerstone of the digital economy, serving as data storage hubs, communication gateways, and powerful processing engines. The term data center encompasses a diverse sector at the intersection of real estate and infrastructure, resulting in a wide variety of investment types, operating models, and lease structures. Data center investments such as powered shells and strategic colocation resemble traditional real estate investment types, while carrier hotels and digital connectivity infrastructure are typically infrastructure assets.

As global demand for cloud computing, artificial intelligence, and enterprise data management accelerates, the need for digital infrastructure and data centers continues to grow. While this sector has experienced robust supply growth, it has been bound by structural constraints due to limited power availability, location limitations, regulatory constraints, long equipment lead times, and development costs that can exceed $10 million per megawatt (MW). Owners and developers of data centers have reaped the benefits of strong preleasing and long-term leases to predominantly tenants with strong credit, as well as attractive development economics.

Cloud Computing Proliferation

The public cloud storage and computing ecosystem has grown substantially over the past decade as companies relocated internal data to secure data centers operated by third parties. Prior to the broad adoption of the public cloud, many companies managed their own enterprise data centers to house and use internal data. By transitioning to the public cloud, companies save money by not developing and maintaining their own facilities, and gain access to the latest hardware and software programs cloud computing companies continuously invest in. Furthermore, for corporations with growing data storage needs, utilizing a third-party operated cloud data center provides maximum flexibility to scale data center needs. In addition, the growth of the digital universe (the Internet of Things, evolution of digital content, etc.) is creating vast oceans of data daily. IP devices, which span from watches to smart home appliances to boat GPS systems, continuously generate real-time data that must connect to a data center for processing, storage, and transmission.

The global cloud infrastructure market is dominated by hyperscalers, which account for more than 50 percent of global customers. Furthermore, according to earnings filings, revenues for cloud computing companies have increased more than 300 percent since the start of 2020.

To support this growth, Bloomberg estimates the hyperscalers are forecast to surpass $350 billion in capital expenditures in 2025, with most of the investments going toward data centers and data center infrastructure. However, certain hyperscalers have noted that even with the substantial investments they are making, demand continues to outpace their existing data center capacity.

Artificial Intelligence

AI has swiftly taken the data center ecosystem by storm, reshaping the technology landscape and creating significant implications for digital infrastructure. The demands from AI regarding hardware and power have created a lasting effect on the data center sector. However, AI isn’t just one process in one place, it is a series of interrelated workloads distributed across different types of data centers. The two primary categories of AI workloads can be broadly classified as training and inference.

AI training models process vast datasets, establish parameters, and produce complex pattern-recognition models. These models can be “trained” using trillions of data points (tokens) and can require a gigawatt or more of energy. Unlike more traditional data center facilities, AI training facilities often require higher power densities due to utilization of more sophisticated hardware, such as graphics-processing units and custom AI accelerators. Furthermore, these power-intensive servers require improved cooling technology. Since the training phase occurs prior to the deployment of a model (i.e., mass customer utilization), there is less latency sensitivity, which allows the hyperscale campuses to be built further from the end user base. As the hyperscalers and AI-focused companies continue to develop training models, the desirable locations tend to be those with power and land abundance and favorable regulations. Due to the power constraints found in many primary data center markets, some recently announced large AI training data center campuses have been breaking ground in locations including Indiana, Ohio, and Louisiana. Once a model is trained, it is deployed, which is commonly referred to as AI inference.

AI inference occurs when trained models are used to deliver real-time outputs, such as generating responses in chatbots, powering autonomous vehicles, or creating predictive analytics. These inference tasks require distributed computing infrastructure located closer to end users to minimize latency and optimize performance. As a result, the desired locations for AI inference deployments are within colocation and edge data centers near population centers. Since the inference model has already been trained, power requirements to operate the model are much less than what is required in the training phase. The continued integration of AI into everyday applications is expected to drive sustained demand for low-latency data center facilities such as edge and colocation data centers.

Government Support for Data Centers and AI

In early 2025, President Trump issued Executive Order 14179, “Removing Barriers to American Leadership in Artificial Intelligence,” which outlines the government’s goal of bolstering AI development to benefit “human flourishing, economic competitiveness, and national security.” While this was not the first administration to release policies encouraging the growth of AI and data centers, this executive order has encouraged activity across major public and private organizations, with the most notable project being Stargate.

While the sector has delivered record inventory for each of the past four years, meaningful supply constraints pertaining to power availability, labor shortages, and semiconductors have lengthened the delivery timeline.

Power Availability 

The challenge of energy availability and how it can support the growth of the digital sector is a top-of-mind issue for investors and constituents participating in the digital ecosystem. The data center industry faces energy shortages, with lead times for grid interconnection requests in primary markets ranging from two to seven years, according to S&P Global Market Intelligence.

Electricity consumption in the United States has remained subdued for much of the past two decades due to increasing demand from population growth, economic growth, and electrification being offset by gains in energy efficiency. However, as data center development has quickly evolved, electricity consumption is rising and is anticipated to grow meaningfully faster.

While a mix of energy sources is going to contribute to the expansion of the power grid in the coming years, natural gas-fired electricity generation is likely to play a vital role in the expansion in the short to intermediate term. The development of data centers is crucial to the growth of the U.S. economy and is even being labeled as a matter of national security by the U.S. government, creating a sense of urgency to develop power generation. Natural gas generation provides cost-effective, reliable, and high-density electricity. Global Energy Monitor reports that as of the end of 2024, 85 gigawatts (GW) of natural gas capacity—representing 15 percent of total existing capacity—were either under construction, in preconstruction, or announced construction. In turn, orders for new gas turbines have surged; however, since turbine manufacturing had adjusted to a low energy consumption era, there are substantial backlogs. Turbine deliveries for new power plants now face potential delays of several years. Additionally, generators, switchgears, and transformers are taking a longer time to deliver.

Further compounding electricity constraints is the limited development of high-voltage transmission lines. In 2023, Grid Strategies estimated that 55 miles of high-voltage (345-kV) transmission lines were added across the United States, well below the 3,500 miles built in 2013 and nearly 700 miles built annually from 2015 to 2023. A robust high-voltage network is necessary to support the growth in electricity generation and data center developments.

To circumvent some power constraints, developers and tenants are targeting markets where power is readily available and proposing generating power on-site, also known as behind-the-meter generation. Data center developments in historically secondary markets such as Columbus, Austin, Reno, and others have soared. In addition, more conversations are being had about the development of behind-the-meter generation; however, this remains a costly alternative.

Labor Shortages

Construction and operation of thousands of data center projects, as well as the development of new power generation, require a robust workforce of skilled laborers. Due to the technical requirements of data center projects and the finite skilled labor force in the United States, some data center developments struggle to meet project schedules. 

There is an inherent conflict between data center developers, hyperscalers, and utility providers as they compete for many of the same employees. Skilled laborers, on average, receive higher compensation to work for the hyperscalers, which has created an exodus from the utility industry that is already struggling to develop power and transmission infrastructure. With data centers requiring power, but utilities remaining understaffed, the likelihood of delays is higher.

Certain data center facilities, such as AI training models, need abundant power and land with limited latency sensitivity. These projects have broken ground in rural areas requiring developers to import skilled labor, potentially drawing this workforce from locations that may be more proximate to primary data center markets. By relocating these employees, the pool of workers in primary markets is further depleted. 

The seemingly insatiable demand and growing supply constraints are providing a strong backdrop for operating fundamentals. As of 2Q 2025, datacenterHawk reported that the national vacancy rate remains below 2.0 percent, with virtually all developments pre-leased prior to construction commencing. While commissioned power growth in the first half of 2025 (1,440 MW) has lagged behind the growth realized in the first half of 2024 (2,299 MW), it is important to consider that limited vacancies and supply constraints are hindering higher levels of growth.

As vacancies have tightened, asking rents have grown by more than 15 percent per year from 2021 to 2024, according to CBRE data center broker reports. Without quick solutions to the near-term supply constraints as demand continues to build, operating fundamentals should be strengthened from this backdrop.

– Harrison Street Asset Management