Data centers as an asset class. Evaluation of specialized REITs

Master's Thesis, 2020

83 Pages, Grade: 1,7



List of Figures and Tables

List of Abbreviations

1. Background and Overview

2. History of Data Centers

3. Basic Characteristics of Data Centers
3.1. Shell and Core
3.2. Technical Fit Out
3.3. Site Requirements

4. Data Centers Through a Real Estate Lens
4.1. Data Center Types
4.2. Real Estate Specifics
4.3. Data Center Investment

5. Introduction to Data Center REITs

6. Literature Review and Data Basis
6.1. Excursus: Total Return
6.2. Literature Review on the Performance of Data Center REITs
6.3. Data Basis
6.3.1. Data Center REITs Index
6.3.2. FTSE NAREIT US Real Estate Index Series
6.3.3. 1-Month Treasury Rate
6.3.4. Stocks and Bonds
6.3.5. Fama/French Research Factors

7. Performance Analysis
7.1. Data Center REITs vs. other REIT Sectors
7.1.1. Descriptive Analysis: Total Return, Risk and Sharpe Ratio
7.1.2. CAPM and Carhart Four-Factor Model
7.2 Data Center REITs vs. Stocks, Bonds and Cash

8. Portfolio Benefits
8.1. Theoretical Foundations
8.1.1. Modern Portfolio Theory
8.1.2. Overall, Diversification and Return Benefits
8.2. Correlations
8.3. Data Center REITs in a REIT Portfolio
8.3.1. Portfolio Optimization
8.3.2. Overall, Diversification and Return Benefits
8.4. Data Center REITs in a Mixed-Asset-Portfolio
8.4.1. Portfolio Optimization
8.4.2. Overall, Diversification and Return Benefits

9. Summary and Limitations of Empirical Findings

10. Conclusion and Outlook

List of References

List of Data

List of Figures and Tables

List of Figures

Figure 1: Data Center REITs Index Weights

Figure 2: FTSE NAREIT Property Sectors

Figure 3: Compounded Total Return Data Center vs. All Equity REITs

Figure 4: Risk-Return Spectrum Data Center REITs vs. Other REIT Sectors

Figure 5: Efficient Frontiers for REIT Portfolios

Figure 6: Asset Allocation in a REIT Portfolio incl. Data Center REITs

Figure 7: Risk-Return Spectrum Data Center REITs vs. All Equity REITs, Stocks, Bonds and Cash

Figure 8: Efficient Frontiers for Mixed-Asset Portfolios

Figure 9: Asset Allocation in a Mixed-Asset Portfolio incl. Data Center REITs

List of Tables

Table 1: Equity REITs by Property Sector

Table 2: Data Center REITs

Table 3: Data Basis

Table 4: Total Return, Standard Deviation, Sharpe Ratio: Data Center REITs vs. other REIT Sectors

Table 5: CAPM and Carhart Four-Factor Model; Full Sample Period Data Center, Diversified, Office, Retail, Residential

Table 6: CAPM and Carhart Four-Factor Model; Full Sample Period Data Center, Industrial, Lodging, Health Care, Self Storage

Table 7: CAPM and Carhart Four-Factor Model; GFC Period Data Center, Diversified, Office, Retail, Residential

Table 8: CAPM and Carhart Four-Factor Model; GFC Period Data Center, Industrial, Lodging, Health Care, Self Storage

Table 9: Total Return, Standard Deviation, Sharpe Ratio: Data Center REITs vs. Stocks, Bonds and Cash

Table 10: Correlations

Table 11: REIT Sectors - Total Return, Standard Deviation and Correlation with Data Center REITs

Table 12: REIT Sectors - Overall, Diversification and Return Benefits of Data Center REITs

Table 13: All Equity REITs, Stocks, Bonds and Cash - Total Return, Standard Deviation and Correlation with Data Center REITs

Table 14: All Equity REITs, Stocks, Bonds and Cash - Overall, Diversification and Return Benefits of Data Center REITs

List of Abbreviations

CAPM: Capital Asset Pricing Model

FRED: Federal Reserve Bank of St. Louis’ Economic Database


FNCO: FTSE NAREIT Composite Index

FTSE: Financial Times Stock Exchange

GFC: Global Financial Crisis

HML: High Minus Low Factor

IPO: Initial Public Offering

MOM: Momentum Factor

NAREIT: National Association of Real Estate Investment Trusts

REIT: Real Estate Investment Trust

REOC: Real Estate Operating Company

SMB: Small Minus Big Factor

S&P: Standard & Poor's

1. Background and Overview

Data is without doubt the raw material of the 21st century and an exponential data growth has been recorded worldwide in recent years. It is expected that annual global internet traffic will reach 4,8 zettabytes (1 zettabyte = 1 trillion gigabytes) by 2022. In 2017 global traffic stood at 1,5 zettabyte per year and thus is expected to triple by 2022. Compared to 2012, internet traffic would have multiplied by a factor of approx. 11. This dynamic de­velopment is supported by numerous market drivers, which could possibly favor that this forecast will be significantly exceeded. By 2022 it is expected that the number of internet users will increase to 4.8 billion from 3.4 billion in 2017, meaning that around 60% of the world's population will have access to the internet. Not only the number of internet users but also the number of networked devices used will increase significantly and will be more than three times of the global population by 2022. Thereby broadband speed will increase, enabling faster and better internet experiences to users.1 Certainly we live in an age of constant technological improvement. Artificial intelligence, virtual and augmented reality, internet of things, smartphones and mobile data usage, adoption of cloud services and high-definition video streaming are just a few examples of rapid technological devel­opments in recent years. All these technology trends have one thing in common, namely that they require enormous data capacity and fueled the tremendous growth in data vol­ume in the past and will continue to do so in the future.2

All these data, we generate every single day, must actually be stored and managed phys­ically somewhere: in data centers, the backbone of the internet and the 21st century digital economy. Obviously, data centers are the beneficiaries of the trends stated above and have become an elementary part of our society with many businesses and social processes re­lying on their continuous operation. Especially due to the very favorable market environ­ment, demand for space in data centers is constantly increasing. It is therefore not surpris­ing that data centers have gained enormous potential as an alternative real estate invest­ment and that investors are showing growing interest in this niche segment.3 To give an example: in summer 2019, Singapore’s sovereign wealth fund GIC, set up ajoint venture with Equinix, the leading data center real estate investment trust (REIT), to invest more than $ 1 billion in six European hyper scale data centers.4 There are several reasons why investors might consider an investment in data centers. Of course, the performance of data centers can mirror the tremendous growth of the internet and technology sector while providing the tangibility and security of a real estate investment. Especially in saturated markets, non-traditional asset classes like data centers can offer higher returns with toler­able risks compared to other property types.5 Furthermore, market drivers for the data center sector differ significantly from those of traditional real estate sectors.6 It is ex­pected that the performance of data centers acts less volatile and shows weaker links to business cycles than other property types and asset classes.7 The fact that data centers seem to be rather independent from macroeconomic factors offers great diversification potential. Adding non-traditional property types like data centers could help optimizing real estate portfolios by either contributing diversification, reducing portfolio risk and/or increasing portfolio returns.8 This applies even more in the current low interest rate envi­ronment, where yields for traditional property types continue to fall.9

Until now, little attention has been paid to data centers as an alternative asset class. The literature and research on this special property type is limited and usually follows a more technical approach (see Marzuki and Newell for a brief overview).10 Only a few studies have examined data centers as a potential alternative investment opportunity and gathered their performance and diversification benefits. These will be discussed in more detail in chapter 6.2. later on. As the data center sector is rather nascent and opaque, capturing direct performance of data centers is not possible (at least not with public and traceable sources). For this reason, existing literature focused on the performance of indirect in­vestments via public data center REITs. This thesis adopts the indirect approach and aims to take up the rare literature on this specific topic and contribute to it wherever possible.

First of all, some basic background knowledge about data centers as an alternative asset class will be provided in chapters 1-4. However, the main part of this thesis will be a comprehensive empirical analysis of the performance, risks and portfolio benefits of data center REITs relative to other commercial REIT sectors and major financial asset classes in chapter 5-9. The focus will be on the comparison of REIT sectors respectively property types, as this topic has rarely been dealt with in literature so far.

Following research questions were formulated:

Research Question 1: How do investments in data center REITs compare to investment in other REIT sectors and major financial asset classes in terms of risk and return?

Research Question 2: How do data center REITs act in a portfolio and which benefits can be gained from including them in a portfolio of different REITs respectively different financial assets?

Due to the facts that data center REITs specialize in in a property type, which requires tremendous specific knowledge and the simultaneously very supporting settings stated above, the following hypotheses will be examined:

Hypotheses 1: Data center REITs achieved higher returns but also higher risk than other REIT sectors and financial assets in the past.

Hypotheses 2: Data center REITs play a substantial role in a portfolio, that can be at­tributed in particular to the outperformance of other REIT sectors and financial assets.

2. History of Data Centers

Over the last decades, data centers have evolved from an initially corporate premise to a niche asset class in the broad real estate and infrastructure universe, that is attracting more and more attention due to the rapid evolution in IT.11 This chapter will give a short his­torical overview of how data centers and the data center industry have evolved in the past.

The evolution of data centers begins with the first emergence of mainframe computers in the 1950s. At the beginning mainframe computers were produced and used exclusively for research purposes, often with a military or governmental background. During the 1960s and into the 1980s mainframe computers were adapted by larger businesses, espe­cially by banks and other financial service providers. In the early years, these mainframe computers and the associated infrastructure filled entire rooms. It was decided to store these computer systems in separate rooms, where their environment (especially the cli­mate by air conditioning) was manageable and adapted to the use. The first data centers were born, even though they were usually located within the building of the business using the facility. With the emergence of the personal computer during the 1980's, further adoption of the mainframe technology was retained. Due to the increasing spread of per­sonal computers, data centers lost much of their importance as they were not necessary for the use of personal computers. But it was not to be long before they reappeared, and their importance increased again. In the following years the performance of personal com­puters increased steadily and thus found its way into more and more areas of life. Rapid development of computer technology, the spread of client-server-computing as well as the origination and growth of the internet for public use led to an increase in the demand for offsite data storage in the 1990s. The strong demand was mainly driven by an increas­ing number of new businesses in the telecommunication and technology sector who wanted to participate in the dot-com boom. The business models and operations of these companies were heavily dependent on secure and reliable data storage facilities, so they began to outsource their IT infrastructure to standalone purpose-built facilities. Thanks to the internet, these data centers no longer needed to be close to their operators. Data centers as we know them today were born and had a new mission to fulfil: to serve as internet nodes. At the turn of the millennium, new development of data center space was at its peak. This euphoria came to an abrupt end with the bursting of the dot-com bubble in the year 2000. Demand from technology companies dropped off and was followed by a strong imbalance in the market for data center space, as increased supply met with only low demand. Following 9/11 it was above all demand from financial service occupiers, which was able to restore the balance between supply and demand. Increasing sensitivity to­wards data protection from global terrorism and digital crime led to stronger demand for high-quality, secure data center space. Even though the financial sector continues to be a major user of data center space, others have joined over the course of time and technology companies have again been the undisputed drivers of the data center industry over the last decade.12

Further innovations in IT and the progressive digitalization created ever-increasing data storage requirements and thus data centers have become a crucial element in 21st century’s digital society. As IT has become increasingly complex, companies stopped managing their own data centers and started housing their IT equipment in data centers managed by specialized third-party operators. With the ongoing trend of cloud computing, outsourc­ing goes one step further. Businesses can now decide, if they want to manage their own IT equipment or utilize the cloud.13

After this brief overview, it should be clear that the data center industry is evolving rap­idly. The forecast for this sector is exponential growth. However, investments in data centers are subject to risks due to their high specification and their dependence on the evolution of the underlying technology. Exponential growth in demand for telecommuni­cation infrastructure was also expected at the end of the 20th century. These high expec­tations came to an abrupt end at the beginning of the 21st century when the dotcom bubble burst.14 Potential investors should take a closer look at the characteristics of this niche asset and its fast-changing environment. In the following, the unique characteristics of data centers will be discussed more detailed in order to shed some light on this rather special property type.

3. Basic Characteristics of Data Centers

Data centers are buildings that are intended to provide a physical highly reliable, secure and optimized environment for housing IT equipment and the associated infrastructure. The overall objective of a data center is to ensure that its operations, which are mission critical for the occupiers, are never interrupted. This challenge makes them sophisticated special-purpose facilities, which differ significantly from other property types especially due to their technical characteristics. About 80-85% of the total investment costs of a data center are solely accounted for by the technologically advanced fit out, while the real estate shell only represents 15-20%. In order to fulfil their mission, data centers are sub­ject to specific physical and technical requirements, while their location criteria differ significantly from other property types. These highly specific characteristics will be dis­cussed in this chapter. First of all, the physical and technical characteristics of the data center itself will be discussed. This is followed by a brief overview of the specific site requirements.15

3.1. Shell and Core

From the outside, data centers are relatively unsightly, windowless buildings that do not look inviting at all. That's exactly what they are not supposed to be, because they are not designed for people, but solely for machines. On their external appearance, data centers are most similar to warehouses and therefore are also allocated to the industrial real estate segment. Because both share many external characteristics, it is not surprising that in the past, data centers were set up in former industrial or logistics buildings. Another possi­bility is the purpose-built construction. The choice between new construction and con­version can have a significant impact on construction and management costs as well as operational flexibility. Today, preference is given to purpose-built data centers that are optimally adapted to data center operations. Although the building structure of a data center only represents a small part of the overall investment costs, it has to meet a few requirements. Preference is given to buildings with large open floor surfaces, that allow for flexible and efficient division of space. Due to the heavy technical equipment a high floor loading capacity of 1 ton per square meter is required. A ceiling height of at least 3.5 meters must be provided, although higher spaces are desirable because of the massive heat generated. Furthermore, the building should have physically secure boundaries. These and the data center itself should be constructed in such a way, that it can withstand severe external influences.16 The areas inside a data center can be roughly divided as follows: As already mentioned, data centers provide the optimal infrastructure for IT equipment such as servers, which store and manage the data of the facility’s occupiers. Therefore, server rooms usually take up the main part of a data center and are commonly referred to as white space. A typical server room consists of numerous servers stored in cabinets (“racks”). These are usually placed on a raised floor, which creates space for the cabling and cooling of the IT equipment. Several racks are usually aligned in rows, cre­ating hot/cold aisles to optimize cooling. Depending on the owner and rental structure, the white space represents the rentable area of a data center. Beside the server rooms, the data center must provide space for all the support equipment and services, which are nec­essary to ensure continuous operation of the servers. These include, for example, areas for all mechanical and electrical systems like power and cooling or space for communi­cation and networking equipment.17

3.2. Technical Fit Out

The ultimate aim when operating a data center is to ensure that it always functions reliably and never fails running the users’ mission-critical IT processes. To be able to guarantee this, comprehensive technical arrangements must be made. One essential step is to create redundancy for all critical infrastructures, including power supply, cooling and connectivity. Redundancy is achieved by duplicating these critical components. If one component fails, a backup is provided in case of emergency. The redundancy of critical infrastructure can take on different dimensions and is depending on the needs of the users. A so-called N configuration thereby describes a non-redundant critical infrastructure sys­tem. Each critical component (N) is only provided once, meaning that if one of the com­ponents fails, the whole system collapses. The N+1 configuration provides basic redun­dancy, meaning that each critical component has at least one independent backup (+1). This minimum level should be met by each data center.18 Much higher redundancy is offered by the 2N configuration, in which the entire critical infrastructure system (2N) is duplicated rather than providing backups for individual critical components. Highest re­dundancy is guaranteed by the 2N+1 setup, because a backup is provided for each critical component, even though the entire system is already completely duplicated.19 The tech­nical fit out, which is supposed to guarantee continuous operation, can roughly be divided into power supply, environmental control, network connectivity, fire protection as well as several security measures.

Power Supply: Large amounts of electricity are needed to supply all the technical equip­ment in a data center with power. The power supply is therefore the most important and at the same time most critical point for continuous operations. It is vital that it does not fail under any circumstances, otherwise data will be lost. Redundancy of power supply with independent lines to different power plants is essential, should one supplier fail for a short time. The internal power supply must be designed section wise and duplicated entirely. By doing this the entire system does not have to be shut down for maintenance work. The complete power supply has to be supported by a standby system, which usually consists of an uninterruptible power supply and additional power generators and must also be connected redundantly to the IT equipment. In the case that the main power supply is interrupted, the uninterruptible power supply is immediately activated until a secondary power source is put into operation. The uninterruptible power supply is basically an over­sized battery in the data center, which is always kept on standby and is immediately avail­able if the power supply should fail. With this system, short-term power failures of up to a few minutes can be covered. If the main power supply is interrupted for a longer period and cannot be covered by another supplier, the independent emergency power supply in the form of power generators must be utilized. Depending on the data center and its location, these generators have to provide power capacities for outages ranging from a few hours to several days. Of course, sufficient fuel must be stored in the data center to guarantee the emergency power supply for this period.20

Environmental Control: Without any intervention, the temperature in a server room would rise steadily. As high temperatures can lower the performance of the servers, it is essential to install a powerful cooling system that keeps both temperature and humidity in the server rooms constant. The temperature in a server room should be around 22 °C, the air humidity at about 60%. To maintain this constant environment, the entire cooling system usually consumes more power than the servers themselves. In order to reduce power consumption, essential thermodynamic principles are applied. As already men­tioned above, a server room consists of numerous server racks, aligned in rows on a raised floor. The front sides of the racks thereby face each other, creating the so-called cold aisles. Cold air from the air conditioning system is distributed via the raised floor to the cold aisles. The cold air flows through the servers and cools them down. The servers vent out hot air to the back side of the servers, the hot aisle. From there the hot air rises to the top of the server room, where it is drawn in and cooled down by the air conditioning system and so on. This simple but useful configuration prevents hot and cold air from mixing, reducing overall cooling requirements and energy consumption.21

Connectivity: In an increasingly digitally connected world, data centers must enable their users to communicate with the world. Communication is thereby guaranteed by networks based on the internet or cross-connections within a data center or between different data centers. Constant internet connectivity is self-evident. In order to ensure this, redundancy via direct links to at least two internet providers is essential. The more links are available in the data center, the more flexible it can be used or rented. Since the use of public inter­net networks is prone to problems, data centers usually offer their users the possibility to interconnect their IT equipment directly via cross-connections with users in the same data center or other data centers. Cross-connects are much cheaper and more reliable than connections via conventional internet networks, and therefore offer considerable ad­vantages to data center users.22

Fire Protection: Prevention of major fire damage is essential and usually comprises sev­eral active and passive components in a data center. The installation of highly sensitive smoke detectors and fire alarms is obvious and can prevent worse before a fire starts spreading. In the event that a fire develops that cannot be extinguished manually, most data centers are equipped with a clean agent fire suppression gaseous system. This type of fire suppression involves filling the entire affected area with inert gases. Conventional fire sprinkler systems are only used if the fire could not be contained by the gas. Because the negative effects of the gaseous fire suppression are rather limited relative to the sup­pression by water, the former is preferable. If all active measures do not succeed, data centers are usually equipped with passive protection elements such as fire walls and doors. However, these can only prevent the fire from spreading further throughout the entire plant.23

Security: The security of the IT equipment also plays a critical role for its continuous operation. Therefore, physical access to the facility is most often restricted to authorized personnel. Usually, a wide range of security measures is used to deter physical intruders. The scope of these measures varies with the sensitivity of the data stored and processed in the facility. Possible physical security measures are among others video surveillance, security guards and doors, or even card access with biometrics and visual identification. A data center must of course also be protected against digital intrusion by cyber-attacks, so sufficient protective measures must be implemented by the operator.24

3.3. Site Requirements

Due to their high standards for continuous operation, risks arising from potential disrup­tive factors must be minimized as far as possible. Several issues must be considered when choosing a location for a data center.

Power supply: The availability, reliability and of course costs of power infrastructure is probably one of the most important location factors. Power supply with high capacity must be provided on the site. A modern large data center can have a power requirement of more than 100 megawatts. By comparison, this is roughly equivalent to the power consumption of a city with 100,000 inhabitants. Such a high availability of power at a single site is not universally available and is therefore a key factor in determining whether a location is suitable for a data center or not. As already mentioned, the site should be powered by two diverse power suppliers, to ensure reliability. Underground power is preferable, because it reduces the effects of external influences on the power supply. The smaller the distance to the power generation source, the lower the cost of electricity. For this reason, it is not unusual to build additional sub stations near to data center sites. Costs for electricity make up a substantial part of the operating costs of a data center and must be minimized. This can be achieved by either selecting locations in regions with low costs or realizing cost savings, for example, by integrating renewable energies (solar, wind, geothermal, hydro energy) or using air and water for cooling.25,

Connectivity: Besides the power supply, the supply and the associated costs of commu­nication infrastructure is essential. Because a data center usually produces enormous amounts of network traffic, the site must offer sufficient fiber capacity. Access to the internet should be provided by multiple independent service providers, that use diverse paths to connect with the data center. Primary goal of the servers in the data center is to handle requests from the end user as fast as possible. The latency, time between request of the end user and response by the server, must be minimized. Low latency is essential for most data center users and is reduced by minimizing the distance to internet exchange points. The proximity to internet nodes therefore represents a location advantage, why many data centers are located near to them.25 26

Climate and Natural Hazards: Due to the enormous power consumption resulting from the required cooling of the IT equipment, cold and dry regions are preferred over warm and humid regions when choosing a location for a data center. Natural hazards can have a significant impact on the operation of a data center and correspondingly have enormous risk potential. These include in particular floods, extreme weather events and seismic ac­tivity. Their historical frequency should be carefully analyzed to determine the probability of recurrence.27

Man-Made Hazards: There are of course not only natural hazards, but also man-made ones. They must be taken into account when choosing a location. It is important to reduce the proximity to as many sources of man-made hazards as possible. Potential hazards are numerous and include for example: transport hazards (e.g. proximity to airports and rail­ways); industries with hazardous processes (e.g. proximity to nuclear power plants, chem­ical production facilities and refineries), interferences due to nearby electro-magnetic fields (e.g. proximity to power lines and substations), criminal activity.28

This rough listing of site requirements does not make any statement about the priority of the individual factors, nor does it claim to be complete. Many other factors have to be taken into account when choosing a location for a data center, e.g. various socio-economic factors such as the availability of trained workers, tax incentives and regulations.29 Nev­ertheless, this overview offers a first insight into the complex site selection process for data centers. With the preceding overview of the physical and technical features of data centers themselves, a superficial basic knowledge should have been obtained.

4. Data Centers Through a Real Estate Lens

After the rather general introduction to the specifics of data centers in the last chapter, this chapter will take a real estate perspective before the actual empirical analysis takes place in the next chapter.

4.1. Data Center Types

Depending on the owner, operator and tenant structure, a distinction can be made between four different types of data centers: corporate, wholesale, colocation and carrier owned data center. Boundaries between these types are partially blurred and a clear distinction is difficult.

Corporate Data Center: The first type is a conventional owner-occupied facility, most often used by large companies or government institutions to store their own data. They are often purpose-built and located on site. The IT equipment is usually managed by the company’s own staff. Depending on how much IT equipment is housed, the size of a corporate data center varies greatly. For some users it makes sense or is even necessary to maintain their own data centers. However, in view of increasing technological com­plexity and thus also costs, it is often cheaper to outsource IT equipment to wholesale or colocation data centers. The trend towards outsourcing has developed in recent years and it is expected that many conventional corporate data centers will disappear from the mar­ket in the next few years. Another possibility of outsourcing is of course to not only out­source own IT equipment to another data center, but to simply use the IT equipment of cloud providers via the cloud. Strictly speaking, this is also located in corporate data cen­ters. These are called hyperscale data centers and are used by hyperscalers, providers of cloud and big data solutions, such as Amazon Web Services, Microsoft Azure or Google Cloud.30

Wholesale Data Center: Wholesale and colocation data centers are occupied by whole­sale respectively colocation operators (e.g. data center REITs). The main difference be­tween the two is that they provide different sizes of space to a different number of tenants, who are offered a differing range of services. Thereby, wholesale data centers offer their space to larger companies and organizations, which require larger spaces for their IT equipment and more autonomy as well as physical control. The wholesale operator often owns the facility and leases it to a small number of tenants, typically less than 100. The wholesale operator usually provides all services up to the power supply and cooling of the white space. Tenants can lease entire server rooms (“suites”) or larger separate areas in a server room (“cages”), installing and maintaining their own IT equipment based on their expectations. Because tenants need to make considerable investments, longer lease terms between 5 and 15 years are common in the wholesale segment. Due to the small number of tenants, interconnection between them plays a subordinate role in the whole­sale segment. One of the major players in this segment is the data center REIT Digital Realty, which offers its services in over 200 data centers worldwide.31

Colocation Data Center: In contrast to wholesale operators, colocation operators offer their services to a wide range of diverse tenants. Colocation operators do not necessarily have to own the facility. It is common, that they lease larger spaces in other data centers, e.g. from wholesale operators. Tenants do not have to rent large areas from the colocation operator. Depending on their requirements, they can rent single server racks, or some­times even parts of them. Usually, the space is let fully fitted out up to the server racks by the operator. The colocation operator usually covers the entire range of services (e.g. power, cooling, telecommunication infrastructure, IT and maintenance staff). Maximum service and flexibility are the maxim for this data center type. Leases are therefore usually short-term, between 1 and 5 years. Colocation data centers are in general carrier-neutral, i.e. not bound to the service of one particular network provider. The aim is to combine as many carriers as possible in the colocation data center to attract many different customers. Because colocation data centers are home to a large number of companies and organiza­tions, cloud and other IT service providers usually house their IT equipment in these fa­cilities and deliver their services by cross-connecting with their customers. In colocation data centers, the interconnection between individual tenants thus plays a much greater role than in wholesale data centers. By far the largest player in the colocation and inter­connection market is the data center REIT Equinix. As the business models of wholesale and colocation operators is increasingly merging, clear separation of both is becoming blurred. Traditional wholesale operators, for example, start adding colocation spaces to their data centers to meet flexible tenant requirements.32

Carrier Owned Data Center: As the name suggest carrier owned or telecom data centers are owned by telecommunication carriers like AT&T, Verizon or Telekom. These types of data centers are connected to the network of that carrier. Users are mainly the carriers themselves. It is not unusual for carriers to run their telecom data center in another colo­cation or wholesale facility. It is equally possible that this carrier uses its existing space to offer additional services such as colocation. By now at the latest, it should be clear why it is almost impossible to draw a clear distinction between the individual data center types.33

4.2. Real Estate Specifics

Data centers differ considerably from other property types and thus different criteria must also be considered during the real estate valuation process. This section is intended to bring together the specifics from a real estate point of view.

First of all, key figures are completely different from conventional property types. Ca­pacity of the rentable area in the white space is measured in mega- or kilowatts of avail­able power rather than square meters. Rents are described in dollars per unit of available power and not per unit of rentable area and can therefore not be compared to rent levels of other property types. The ability of a data center to generate rental income depends primarily on how much IT equipment can be housed in the white space. That in turn depends mainly on the availability of power in the data center and how much of it is actually available to the tenant’s IT equipment and not consumed by the supporting in­frastructure (e.g. cooling, back-ups, telecommunications). This ratio is called power us­age effectiveness. The lower this figure, the higher the efficiency of the data center. The rent level is highly variable and strongly depends on the technical specifics and layout of the respective data center. While in other types of use, rent is generally determined by supply and demand, in the case of data centers, the focus is primarily on the costs and return expectations of the supply side and the rent level is often determined using the front door approach. A comparison of the rent level between different data centers is therefore almost impossible, the determination of a common market rent anyway. Furthermore, lease contracts differ to those of conventional commercial real estate sectors. Thereby, the structure of a lease contract varies according to the requested area. The contract for larger occupiers (i.e. wholesale) is usually based on a conventional property lease agree­ment, whereas the contract for smaller occupiers (i.e. colocation) is based on a service level agreement. Conventional lease agreements with larger occupiers ensure that the landlord's costs of maintaining and managing the data center and its mechanical and elec­trical equipment are covered by an additional service charge. Lease contracts for white space let on a colocation basis are usually based on service agreements, under which ten­ants pay a single fee covering the rent and all services provided by the operator. For both, wholesale and colocation, tenants will be charged additionally for the power supplied. Large parts of the investment costs are accounted for by the fit-out with mechanical and electrical equipment. Maintenance of this equipment is usually the responsibility of the landlord and the associated costs usually cannot be passed on to the tenants. However, the owner should not shy away from the considerable costs of maintenance and regularly upgrade and replace the provided equipment. By doing this, a data center remains com­petitive. This is usually rewarded with high tenant loyalty. For tenants, it involves a great deal of effort and costs to move their IT equipment to another data center. As long as the owner keeps the equipment up to date, there is usually no reason for them to leave it. The probability that tenants will repeatedly renew their leases is then very high, which pro­vides the owner with a long-term and secure cash flow. The value of a data center depends to a large extent on the quality of its technological equipment. Its identification is there­fore the most important part during a valuation. Due to its high share in the overall in­vestment costs, investment valuation for data centers differs from other conventional property types. For investments in most property types, income and residual value are taken into account when determining the value. The majority of the value is usually based on the residual value, which is calculated by capitalizing the cash flow at the end of the investment period. When determining the value of a data center, often a residual value of zero is assumed. This is because the majority of the value is due to the technological components, which have to be depreciated much faster than the building itself, which represents only a very small part of the value. Nevertheless, the value of a data center can be determined using the discounted cash flow method as with other property types. How­ever, there are a number of factors that make this method very difficult to utilize. The market for data centers is very small. Transactions are very rare, because the pure number of data centers is rather limited. The income as well as the outgoing costs of a data center can reach considerable levels, at least for investment-grade data centers. Thus, lot sizes can be substantial, further limiting the investment market for data centers. As a result, the investment market is immature and opaque. For example, the performance of data centers is not tracked by MSCI as a separate property type. Even though some large real estate brokers (e.g. JLL, Cushman & Wakefield, Savills) now set up their own teams solely focusing on data centers, the availability of yields is limited and little standardized. Whether the few data provided can be used is questionable due to the strong heterogeneity of this property type. Therefore, market data can only be used for valuation purposes to a very limited extent. In order to determine the value of a data center, sound assumptions, deep market knowledge and high technical expertise are essential.34

4.3. Data Center Investment

Interested investors have several possibilities to access investments in data centers. A rough distinction can be made between direct and indirect investment. Direct investment offers full control over the investment and full participation in potential returns but re­quires considerable knowledge, experience and active management. With indirect invest­ment, the management is outsourced to a third party who contributes the necessary knowledge and experience but requires no knowledge. However, this involves additional costs, which reduce potential returns.

The possibilities of direct investment can again be broken down, depending on the risk and return profile, as follows. Direct investment should only be pursued if sufficient spe­cialist technical management is available in-house or can be covered by partners.

Speculative Development: Probably the most direct method is to acquire a suitable site of land with high availability of power and sufficient capacity of fiber. The investor pro­vides the building structure and takes care of the technical fit out down to the racks as well as the subsequent leasing to an operator or individual tenants. The investment costs are enormous due to the fit out, thus the barriers of entry are rather high. Due to the high obsolescence of the technical fit out the risk taken is very high. Additional risks arise from independent letting, upon which a successful exit is largely dependent. However, by taking on this risk, investors are usually rewarded with the highest expected returns.35

Shell and Core: In this model, the investor contributes a suitable site and the building structure. In addition, the investor is responsible for the availability of sufficient power and the fit out with mechanical and electrical equipment. The investor then rents out this so-called "powered shell" to an operator, whether wholesale or colocation, who comple­ments the technical fit out down to the racks according to his requirements. In this case, the investor bears much lower costs, as the costs for the fit out are borne by the operator. The whole thing also works the other way around in a classic sale and lease back trans­action, where an operator as the owner sells his data center to an investor but continues to run it. Most of the main advantages for an investor pursuing a shell and core strategy are obvious. In contrast to the speculative development less technical expertise is needed. Operators usually offer to sign long-term leases, as they are highly committed due to their high investment in the technical fit out. Therefore, the risks are much lower than with the first model, as is the expected return. The barriers of entry are still not to be neglected due to the high costs of providing the mechanical and electrical infrastructure. In addition, the investor becomes dependent on the cooperation with a third party, the operator, and loses control over his investment. A possibility to reduce the latter problem is to set up a joint venture with an operator to develop or acquire data centers. In practice, this approach has been pursued more intensively in recent years, such as the aforementioned joint venture between GIC and Equinix or the joint venture between the Mitsubishi Corporation and Digital Realty in order to expand their regional footprint in Europe and Asia Pacific re- spectively.36

In addition to the direct investment opportunities described above, there are numerous indirect investment opportunities. Some of these, as well as all direct investment possi­bilities, are only available to strong institutional investors. These can, for example, buy up entire operating platforms to instantly establish a strong exposure in the data center sector. Another possibility is the indirect investment via commitments in private real es­tate or infrastructure funds, which allocate the raised capital to data centers or operating companies. The last option, which is also open to smaller private investors in principle, is to buy shares of listed operators. Here, the risk and the possible return depends strongly on the business model the operator pursues. In the case of a wholesale operator, the risk is much lower because less technological and operational risk is involved. An investment here is much closer to a classic real estate investment. By contrast, an investment in a colocation operator involves a much higher technological and operational risk and is more similar to an infrastructure investment. Due to the higher risks, higher returns can also be expected. In both strategies, investments can be made indirectly via REOCS or REITs. The indirect investment in the latter, its performance, risks and portfolio benefits will be evaluated empirically in detail in the next chapters.37

5. Introduction to Data Center REITs

Now that the most important fundamentals of data centers have been presented, the rest of this thesis, as the title implies, will focus on the performance, risks and portfolio ben­efits of data center REITs. Roughly speaking, REITs are companies that own, finance and manage portfolios of income-producing real estate. Individual investors can hold shares from these companies and thus participate in the income stream generated by the com­pany’s portfolio of real estate assets. Companies with a REIT status are allowed to deduct dividends paid out to their shareholders from their corporate taxable income. In order to take advantage of this tax benefit, companies must meet certain criteria to qualify as a REIT. A distinction between different types of REITs can be made according to their investment focus (equity or mortgage REITs) or their listing at a stock exchange (private, public-non-listed, public-listed REITs).38 Furthermore, REITs can of course be divided into their different property type specializations. Although there are a few REITs pursuing a diversified strategy, most REITs focus on particular property sectors. While some REITs focus on more conventional commercial real estate sectors such as office, residen­tial or retail, others specialize in more alternative usages like health care self-storage and last but not least data centers.39

Data center REITs focus solely on owning and operating data center facilities.40 Looking at table1 below reveals that data center REITs are a rather small sector in the overall US REIT universe. At the end of 2019 the market capitalization of the five public-listed data center REITs was approximately $ 89.5 billion, making up a share of 7.2% in the overall equity REIT market. Although the market capitalization is only a half of the major sectors residential and retail, the size of the data center sector can compete with that of traditional commercial real estate sectors, e.g. offices with a share of 8.3%.41 What the table does not demonstrate is the growth of the data center sector over time. At the end of 2016, the market capitalization was only $ 53.2 billion, which represents a growth of almost 70% over three years. The overall equity REIT market capitalization grew by hardly 30% in the same period.42 This proves the increasing importance of this sector in the overall REIT market.43

Table 1: Equity REITs by Property Sector (Data as of 31/12/2019); following: NAREIT REITWatch January 2020, p. 4

Abbildung in dieser Leseprobe nicht enthalten

The exponential growth in data during the recent years goes hand in hand with the growth of the data center industry. Parallel to this growth data center REITs have become key players in this industry.44 Data center REITs’ ownership share in investment-grade data centers amounts to approximately 30% in the US and 20% globally.45 Taking into account mergers and acquisition activities in the previous five years, e.g. Equinix’s acquisition of 29 Verizon data centers valued $ 3.6 billion in 2017 or Digital Realty’s acquisitions (see below), this share should increase further in the future.46 The answer to the question, why the REIT structure is prominent among developers and operators of data centers, is obvi- ous.47 First of all, of course, REITs are favored by the exemption from corporate taxes as already stated above. Beyond that, REITs benefit from low costs of capital due to their public listing and thus stricter regulations. The data center business is very capital inten­sive. Consequently, access to cheap capital is an enormous advantage for data center REITs. Further advantages are economies of scale, fast expandability and geographic reach in operations which also provide benefit and thus retain customers in the increas­ingly interconnected world.48


1 Cisco (2019), p. 1

2 AXA Investment Managers (2017), p. 2; Guest, Rymell (2018), p. 3

3 Reid, Bryan (2019); Schneiders (2019)

4 Lowe (2019)

5 Bellintani, Ciaramella, Celani (2018), p. 1

6 Schneiders (2019)

7 AXA Investment Managers (2017), p. 3; Bellintani, Ciaramella, Celani (2018), p. 8

8 McIntosh, Fitzgerald, Kirk (2017), p. 66-70

9 Newell, Wen (2006), p. 157-165; Schneiders (2019)

10 Marzuki, Newell (2019), p. 141-142

11 Jacobius (2019)

12 Balodis, Opmane (2013), p. 194-195; Guest, Rymell (2018), p. 2; RICS (2011), p. 2, 4

13 Duncan (2018)

14 O’Brien, Leung (2019), p. 2

15 Bellintani, Ciaramella, Celani (2018), p. 1-3; Duncan (2018); Jones, Hillier, Comfort (2013) p. 104; Schneiders (2019)

16 Bellintani, Ciaramella, Celani (2018), p. 4; RICS (2011), p. 6-9

17 Balodis, Opmane (2013), p. 184; Bellintani, Ciaramella, Celani (2018), p. 4

18 RICS (2011), p. 5-6

19 Balodis, Opmane (2013), p. 190-192

20 Balodis, Opmane (2013), p. 180-181; Bellintani, Ciaramella, Celani (2018), p. 5-6; Herzog (2013), p. 18-30; RICS (2011), p. 5-6

21 Balodis, Opmane (2013), p. 182-183; Bellintani, Ciaramella, Celani (2018), p. 6; Herzog (2013), p. 36­43; Jones, Hillier, Comfort (2013) p. 105; RICS (2011), p. 6;

22 Balodis, Opmane (2013), p. 181; Bellintani, Ciaramella, Celani (2018), p. 5; Poole, Jim (2013); RICS (2011), p. 5

23 Balodis, Opmane (2013), p. 183; Bellintani, Ciaramella, Celani (2018), p. 6; Herzog (2013), p. 45-50

24 Balodis, Opmane (2013), p. 183-184; Herzog (2013), p. 51; Jones, Hillier, Comfort (2013) p. 104; RICS (2011), p. 6

25 Covas, Silva, Divas (2013), p. 278; Mena et al. (2014), p. 5; RICS (2011), p. 5

26 Mena et al. (2014), p. 4-5; O’Brien, Leung (2019), p. 5; Schneiders (2019);

27 Bellintani, Ciaramella, Celani (2018), p. 3; Covas, Silva, Divas (2013), p. 276; Mena et al. (2014), p. 3­4

28 Mena et al. (2014), p. 6; RICS (2011), p. 5

29 Mena et al. (2014), p. 8-9

30 AFL Hyperscale (2019), p. 1,3; Dines (2011), p. 1; Jones, Hillier, Comfort (2013) p. 105; RICS (2011), p. 4; Rosenbush (2019)

31 AFL Hyperscale (2019), p. 2; AXA Investment Managers (2017), p. 3; Jones, Hillier, Comfort (2013) p. 106; O’Brien, Leung (2019), p. 4; RICS (2011), p. 4,13;

32 AFL Hyperscale (2019), p. 2; AXA Investment Managers (2017), p. 3; Jones, Hillier, Comfort (2013) p. 106; O’Brien, Leung (2019), p. 4; RICS (2011), p. 4

33 AFL Hyperscale (2019), p. 3; Jones, Hillier, Comfort (2013) p. 106; RICS (2011), p. 4

34 Bellintani, Ciaramella, Celani (2018), p. 7-8; McAllister, Loizou (2009), p. 67-73; RICS (2011), p. 6­12; Schneiders (2019)

35 Guest, Rymell (2018), p. 5

36 AXA Investment Managers (2017), p. 3; Guest, Rymell (2018), p. 5; Lowe (2019); Wang (2017)

37 AXA Investment Managers (2017), p. 3; Guest, Rymell (2018), p. 5; O’Brien, Leung (2019), p. 4-5

38 Krewson-Kelly, Thomas (2016), p. 4-6; 34; 92-95

39 NAREIT (2020a)

40 NAREIT (2020b)

41 NAREIT (2020c), p. 4

42 NAREIT (2017), p. 4

43 Wimmer, Magerman (2019), p. 2

44 JLL (2017), p. 1

45 Hoya Capital Real Estate (2020)

46 Equinix (2017)

47 Schnure (2017)

48 JLL (2017), p. 2

Excerpt out of 83 pages


Data centers as an asset class. Evaluation of specialized REITs
University of Regensburg  (International Real Estate Business School, Institut für Immobilienwirtschaft)
Catalog Number
ISBN (eBook)
ISBN (Book)
Data center, REITs, Real Estate, Alternatives, Investment, Real Estate Investment Trust
Quote paper
Tobias Hagenah (Author), 2020, Data centers as an asset class. Evaluation of specialized REITs, Munich, GRIN Verlag,


  • No comments yet.
Read the ebook
Title: Data centers as an asset class. Evaluation of specialized REITs

Upload papers

Your term paper / thesis:

- Publication as eBook and book
- High royalties for the sales
- Completely free - with ISBN
- It only takes five minutes
- Every paper finds readers

Publish now - it's free