Decoding Data Centers: Opportunities, risks and investment strategies

Data centers are critical infrastructure needed to support global digitization and technological progress. This crossover sector features characteristics of both infrastructure and real estate investments. As the market rapidly evolves, investors will want to understand the drivers of data center fundamentals, including the significant demand and supply imbalance, investment profiles of various options, sustainability considerations and other risks.

In this article we seek to help potential investors in data centers:

• Understand data centers as assets: Critical to analyzing this asset class is understanding the importance of power supply, the highly technical expertise needed for operating and developing data centers, the rapid pace of technological change in the sector and how these factors are collectively shaping the market. We explain the main types of data center assets and explore how ownership structures are evolving.
• Analyze the demand and supply: Demand for data centers is booming, driven by exponential growth in the amount of data produced globally and the rise of artificial intelligence (AI). In response, a record amount of new data center space is under construction that will nearly double the inventory in the U.S. and other major markets. Over the long term, however, mounting constraints on new supply—largely due to power access—will likely keep vacancies structurally low.
• Explore investment options: In our view, data center investments broadly fall into three access categories: direct investment, development and platforms. Investors will need to look carefully at the risk/reward of investment options, including the asset, type of access and return profile, and consider the appropriate fit within a broader portfolio.
• Assess risks and the outlook for the sector: In this unique and rapidly changing sector, investors must consider the increased focus on environmental impact and potential obsolescence risk.

Understanding data centers as assets

Continued innovation and adoption of new technologies are fueling the increasing digitization of our daily lives: large portions of activities, from communications and banking to marketing and entertainment, have moved to digital platforms. As this trend continues and AI use expands, the volume of data will increase exponentially from current levels.

Data centers—physical buildings that house complex information technology (IT) infrastructure critical for data processing, storage and high-powered computational services—will have an enormous and essential role to play.

Are data centers real estate or infrastructure investments?

The first question many investors ask is whether data centers should be analyzed as infrastructure or real estate.

Consistent with the definition of infrastructure investments, data centers provide an essential service—in this case, digital interconnectivity, including data processing, data storage and computational services that are now critical to our daily lives and global economic growth. Data centers are also closely linked to wireline and wireless infrastructure and their business models share some similarities. Data centers tend to have high barriers to entry—they require access to significant amounts of power; they also have relatively predictable revenues and earnings due to medium- to long-term contracts.

At the same time, data centers meet the definition of real estate because they are physical buildings. Similar to real estate investments, the location and the underlying land—especially access to power and the ability to obtain the necessary permits—are critical drivers of asset performance and market appeal. Many data centers offer stable rental income, like real estate investments, because they are leased to a third party.

Access to power and technical expertise is critical

While data centers share characteristics of both infrastructure and real estate, several unique characteristics are key to understanding the asset class.

Data centers require enormous amounts of power, which becomes a gating factor, as we detail further in the next section on supply and demand. Recognizing the critical need for power, data center tenants are not actually renting space, but rather the guaranteed provision of electric power. Data center rents are normally reported per kilowatt or megawatt (KW or MW).

Owning and operating data centers involves specialized experience and capabilities that are different than investing in other property types. The main reason is that the largest occupiers—Big Tech, especially cloud infrastructure service providers, and telecommunications firms—account for nearly half of the total occupied data center space (Figure 1). These users have exacting technical specifications and security protocols, including power, cooling and back-up arrangements. They entrust their mission-critical computing infrastructure only to experienced and proven operators with long track records.

Figure 1: Cumulative occupancy of U.S. data centers by largest tenants

Large technology and telecom companies account for nearly half of total occupied data center space

(in SF millions)

Three general types of assets

Many different types of data center assets exist (Figure 2), but the market can be simplified into three general categories:

• Enterprise: Owned and operated by the companies they support and usually housed on the corporate campus; these assets are not available to outside investors.
• Colocation: Multiple tenants share a data center. In a retail model, tenants rent racks or cabinets in a data center as a service contract that includes power consumption and cooling. Wholesale colocation features longer leases than retail, typically five to 20 years, and each tenant pays for power delivery.
• Hyperscale: These high capacity data centers are the most modern and are designed to meet the significantly higher technical and security requirements of the largest technology companies, as well as their pricing demands.

Figure 2: Main types of data centers differ by level of operational complexity and outsourcing

The operational complexity (technical requirements around power provision, security and redundancy required by the occupier) and the level of outsourcing (who is responsible for providing the services) can vary within each of these categories.

The general trend is clear—data centers are becoming larger and highly customized to the needs of tenants. Close to 900 hyperscale data centers have been built worldwide, accounting for 37% of total capacity. Current demand trends suggest that number could double and hyperscale data centers could account for half of all capacity, while enterprise on-premise data centers are expected to fall to less than 30% (Figure 3).¹

The shift toward hyperscale data centers is also changing ownership models. The lease vs. build mix for the total hyperscale market has shifted from roughly 50:50 to 70:30, as the big players, such as Amazon, Meta and Google, move from owning to leasing.

By 2027, we expect the shift toward larger facilities will result in a market with roughly 50% hyperscale capacity, half of which we believe can be accessed by third-party investors. Combined with non-hyperscale colocation assets, investors may be able to access about 50% of the total data center market.

Figure 3: The share of hyperscale assets and leased assets is increasing

Assets available for third-party investment outlined in orange

Supply-demand imbalance supports long-term fundamentals

Demand for data centers is growing exponentially while supply faces multiple challenges. The result is a strong fundamental outlook for data center assets in the near and medium term.

AI has supercharged demand

Even before 2023 and the release of AI platforms for public use, expectations for data center demand were high due to the trajectory of data growth (Figure 4). Several trends will push demand significantly higher in the coming years: a shift toward big data, more workloads in the public cloud and robust IT spending. In addition, generative AI will significantly increase as a percentage of technology spending, initially led by the Big Tech companies before a more widespread adoption by enterprises, governments and consumers (Figure 5).

Figure 4: Worldwide data creation, 2010-2035 (zettabytes or billions of terabytes)

Exponential data growth is fueling data center demand

Figure 5: Generative AI spending forecast ($ billions)

Significant growth in AI is further contributing to data center demand

Access to power is the biggest supply bottleneck

Meeting the tremendous demand with enough supply will be challenging. Data centers already require an enormous power supply and are likely to require even more in the future as workloads increase.

For example, data centers currently account for 2.5% of U.S. electricity consumption²—and nearly 20% in Northern Virginia, where almost half of the entire U.S. data center inventory is located. U.S. data center power usage is expected to increase rapidly, from 200 terawatt-hours (TWh) in 2022 to 260 TWh by 2026. By 2030, data centers could account for 7.5% of U.S. electricity consumption. European data center power usage is expected to increase 50%, from 100 TWh to 150 TWh over 2022-2026.

Access to the power grid is the key bottleneck for increasing data center supply. Permitting and building new grids can take five to 15 years. Large operators are experimenting with on-site power generation, potentially in the form of small modular reactors (SMRs). In the meantime, data centers are competing with new power generation to get grid access—nearly 1,600 gigawatts of generation capacity (mostly solar, wind and storage projects) is awaiting regulatory and local approval to get connected to the U.S. power grid—twice as much as the entire current power plant capacity across the country.³

In addition to the power challenges, building data centers is a highly complex process with many stakeholders and considerations. Not all buildings can be converted to data centers, sustainability is a growing issue, fewer developable sites in primary markets are available and construction costs have risen. Taken together, construction of new data centers will likely decline from the current record levels, especially in major markets like Northern Virginia, Silicon Valley and Chicago.

Record low vacancies and strong rental growth

Currently, roughly half of the operating capacity globally for data centers is in the U.S., a proportion that is likely to increase, with implications for global investors that we discuss in the next section. Unprecedented demand for data center space in the U.S., fueled by the AI arms race among Big Tech firms, triggered a record supply wave that will add more than 33 million sf and 3,300 MW of capacity, with another 70 million SF—and as much as 7,000 MW—in various stages of pre-construction (Figure 6).⁴ Taken together, the supply pipeline will more than double existing data center inventory over the next decade, with much of the new supply coming online in emerging markets like Hillsboro County, Oregon, Central Washington State and Omaha, Nebraska, where amenable local governments and utilities permit access to cheap and ideally renewable power at scale.

Figure 6: U.S. data center supply, demand and vacancy (MW, semi-annual)

Tight supply and high demand have resulted in exceptionally low data center vacancy rates

Despite the record amount of new supply delivered and underway, data center vacancies in primary global markets have fallen to record low levels (Figure 7). In Northern Virginia, for example, where data centers have the most direct and lowest-latency connections to the internet, the vacancy rate is below 1%. The low vacancies have driven extraordinary rent growth after roughly a decade of flat or even falling rents for data centers in the U.S. Our investment outlook anticipates that vacancies will remain structurally lower, driving ongoing rent gains (Figure 8).

Figure 7: Data center vacancy: H2 2019 v. H2 2023

Data center vacancies in primary global markets have fallen to record low levels

Figure 8: U.S. data center asking rent per MW and annual change

Data center rents surged in 2023 and are forecast to grow in the coming years

Investment strategies

After several years of frenetic M&A activity (Figure 9), the data center market has consolidated and is dominated by large developer-operator platforms. Private infrastructure funds completed a series of take-private transactions or acquired fast-growing, privately owned data center developer and operator platforms. The two largest listed REITs, Equinix and Digital Realty, offer access to the data center theme with a combined market capitalization of close to $120 billion (CapitalIQ, as of July 12, 2024).

Figure 9: Global data center deals by type, value ($ billion) and number of deals

A variety of deal types contributed to heavy M&A activity in recent years

Investors seeking data center exposure can choose from a variety of investment approaches. We comment on the advantages and disadvantages of three currently prevailing strategies:

• Directly invest in stabilized or near-stabilized data centers
• Develop new data centers
• Acquire a minority or controlling stake of in a data center platform

Direct investment in stabilized or near-stabilized data centers

Data center tenants’ specific and highly technical requirements have resulted in a concentrated market in the U.S.—the top 20 owners, which include Digital Realty Trust, Equinix and Big Tech firms, control more than half of the inventory (Figure 10).

Figure 10: Largest data center owners’ share of U.S. market by SF

The U.S. data center market is highly concentrated among large operators and big technology firms

Many of these firms also rank among the largest data center tenants. Collectively, the relative handful of players constitutes a “club” that puts smaller owners and operators at a disadvantage in acquiring quality assets, setting rents, and winning tenants. As a result, direct ownership of data centers entails specific risks that include:

• Adverse selection bias: Individual data centers for sale by large owners may be older, lower quality properties.
• Limited demand pool: The largest data center users prefer to—and are often required to—lease space from the large, proven owner-operators; new entrants with limited scale will face challenges attracting credit tenants.
• New supply: The supply of data centers currently under construction will increase the existing stock by more than 50%. The new supply is, on average, much larger and provides an order of magnitude more power than existing data centers.
• Obsolescence: Computer processing capabilities continue to increase exponentially, per Moore’s Law, while consuming more and more power, leading to inefficient space use in older data centers. The abundance of new hyperscale supply and the rapid pace of technological innovation create uncertainty around the suitability of existing data centers for the next generation of users.
• Headline risk: The voracious power consumption of data centers and negative sentiment toward AI may create negative publicity for data center owners, especially those with a commitment to sustainability and net zero emissions.

Given these risks, acquisitions of older and less-efficient data centers likely do not meet the standards of a core investment. Assets available for sale could face obsolescence and a shrinking pool of would-be buyers at exit. Value-add opportunities for existing data centers are also limited, given the bureaucratic and regulatory obstacles to increasing the power supply. Most likely, a new entrant acquiring an individual existing data center would end up operating a colocation facility for a set of smaller, non-credit users on short leases.

Looking ahead, however, we expect opportunities to acquire interest in newly built modern hyperscale data centers with long leases to creditworthy tenants to materialize as merchant builders seek to raise capital for future developments. Investments in newer data centers would avoid most of the risks related to direct investment, most notably obsolescence and limitations on future demand.

Development of data centers

Participating in ground-up data center development obviates many of the risks involved in acquiring existing data centers, especially the “lemon” problem—picking an asset that proves challenging to maintain—and obsolescence risk. More efficient operation and reliance on renewable power mitigate the headline risk problem. Modern data centers are more likely to attract large credit tenants and the development sponsor will realize the full benefits of the long first-generation lease. New developments will likely attract a larger pool of institutional buyers at exit. Both Digital Realty Trust and Equinix have driven growth through new development rather than pushing rents on existing centers. Sizable development costs ($10-$14 million/MW of capacity, according to our estimates) also create a barrier to entry and require scale to keep pace with the growth of cloud service providers, especially as AI-driven workloads increase.

Various models for data center development exist, each with advantages and disadvantages:

• Powered land: The investor acquires land, completes entitlements, secures the provision of power and then sells the shovel-ready site. The investor avoids operational responsibilities and risks around obsolescence and tenancy—but also misses out on recurring cash flows and long-term upside in valuations.
• Hyperscale build-to-suit: The most common development strategy at present, the developer builds a bespoke data center to the exact specifications of a major user (likely Big Tech) under contract for a long-term lease (often 15 years or longer). Advantages include a tenant in place and long lease term to a credit tenant. Risks center around operations and meeting the exacting requirements of Big Tech tenants.
• Speculative development (hyperscale or colocation): Speculative development of a single-tenant hyperscale data center capitalizes on the current robust demand and potential rent upside from a bidding war but risks not meeting the idiosyncratic technical requirements of the user and/or delivering into a period of slowing demand. Developing smaller data centers that serve multiple clients avoids the single-tenant risk and highly specific technical requirements in hyperscale development. Little speculative development is currently underway.

Across all development strategies, the unique requirements of data center development—securing and delivering uninterrupted electric power and meeting the specific technical requirements of large occupiers—magnifies the complexity and risk inherent in any real estate development project. The most important decision for investors will be choosing a development partner who has good working relationships with local governments and utilities and a track record of successful projects.

Acquiring a minority or controlling stake in a data center platform

Platform strategies feature both development and operating opportunities and offer investors access to management teams with long-standing expertise and track records of dealing with tenant requirements and changing data center formats. Deploying new capacity mitigates obsolescence risk and platform scale reduces client concentration.

The projected growth in capital expenditures will trigger balance sheet constraints and we expect the large data center platforms to recycle capital by opportunistically selling stabilized or near-stabilized data center assets. The companies continue to operate a large portion of their existing assets, adding a stable stream of operating income to the currently strong development margins.

Platforms currently offer a mix of benefits and considerations:

• Strong development pipelines: Many platforms have accumulated strategic landbanks in desirable markets with power supply that can satisfy the accelerating cloud demand as well as the new wave of AI-driven workloads. In the next three years, the two largest listed REITs are expected to spend annually upwards of $5.0 billion in new development even after accounting for their existing joint ventures with financial investors (CapitalIQ, as of January 15, 2024).
• Continued access to debt financing: Data center operator platforms frequently access the debt markets via traditional loans, asset-backed securities (ABS) transactions, project finance and publicly traded bonds. Despite higher interest rates, the debt markets have remained open for data center financing at rates around 5%-6% on average.
• Solid operating profit margins: Profitability is generally high and stable, although it is affected by rising costs for specialized labor. Most hyperscale and wholesale colocation providers are able to pass through power costs to their tenants. However, we expect increasing externality costs, such as carbon offset purchases or premium power prices for renewable generation as the Big Tech companies and data center operators seek to voluntarily mitigate their carbon footprints.
• Elevated valuation multiples: Over the last four years, acquisition multiples for data center platforms ranged between 25x and 30x in terms of enterprise value to earnings before interest, taxes, depreciation and amortization (EBITDA), compared to an average of 16x for the wider universe of private infrastructure deals.⁵ This reflects the growth in development and the intangible value of strategic landbanks as well as strong investor appetite. Multiples for listed REITs have started to recover in 2023 from the interest rate-driven compression in 2022 (Figure 11).

Figure 11: Listed REITs trading enterprise value (EV) to EBITDA multiple

Choosing the most appropriate type of investment

For investors who have determined that they are interested in gaining exposure to data centers, (Figure 12) summarizes some of the key considerations for choosing an appropriate investment. While these investments are available to both real estate and infrastructure investors, traditionally real estate investors have focused on data center investments with lower operational complexity. In contrast, infrastructure investors have provided a wider spectrum of data center services related to the essential operation of the facilities, including the connectivity requirements of their tenants.

Figure 12: Key considerations for choosing an appropriate type of data center investment

Assessing the risks and outlook

Against a strong fundamental backdrop driven by the significant demand and supply imbalance, we see two key considerations for investors that are likely to play a role in shaping the data center market going forward. First, the speed of technological innovation could impact the asset class in ways that are difficult to predict. For example, the conversion to hyperscale data centers is already making some older data centers less desirable. How easily and at what cost can today’s data centers be upgraded to the needs of the next decade is unclear.

Sustainability is another major consideration for this asset class. Data centers and data transmission networks are already responsible for roughly 1% of energy-related greenhouse gas (GHG) emissions (estimated by the International Energy Agency) and with current data center projections, this percentage is likely to increase.

As sustainable standards are implemented across many industries and countries, data centers are facing mounting pressure to shift to renewable power sources and secure carbon-free energy supplies.

The focus on sustainability and a transition to renewable sources of power could impact the market in multiple ways—from the cost of development to changing which assets command a premium. Some investors may also need to consider the reputational risk from investing in power-hungry and AI-focused data centers.

Looking forward, we expect the massive demand and supply mismatch to persist; projections of exponential data growth will drive robust demand for data centers while supply bottlenecks persist. As a result, vacancy rates should remain low leading to healthy stabilized rents.

The strong fundamentals in the data center market should continue to offer opportunities to both real estate and infrastructure investors looking for direct, development and platform investments.

¹ Synergy Research Group, July 2023.
² Boston Consulting Group, International Energy Agency Electricity 2024. 
³ Berkeley Lab, Queued Up: 2024 Edition, study funded by the U.S. Department of Energy.
⁴ CBRE North American Data Center Trends H2 2023.
⁵ Realfin, average market-wide multiple last five years, as of January 2024.