As global businesses demand uninterrupted uptime, lower latency, and stable connections for increasingly essential and interconnected online operations, the demand for dedicated data centers has grown. Traditionally, enterprises relied primarily on their own in-house data center resources. Given the exponential growth in businesses' data storage and computation needs, as well as the operational and cost efficiencies inherent in cloud services and third-party managed colocation centers, businesses are increasingly outsourcing their data center needs. Improving broadband speeds, the onset of 5G wireless networks, the Internet of Things (IoT), growing streaming video consumption, and the exponential growth in user-generated digital content, such as from social media networks, are among the factors contributing to the demand for more robust data operations.

In response to this rising demand, third-party data center operators have dramatically ramped up their development of new facilities. Such development must be financed, and some operators have turned to the securitization market for long-term financing. Since February 2018, S&P Global Ratings has rated four series of data center-backed securitizations from two data center operators, and has recently rated a few conduit commercial mortgage-backed securities (CMBS) transactions that included loans backed by data centers. In light of the growing interest in securitization financing from data center operators, as well as increased investor interest, we are revisiting the basics of the sector and our approach to the risk.

The four main types of data centers

Data centers are buildings that provide space, power, and cooling for tenants' network infrastructure. Enterprises use data centers to store, share, and manage their data, making security, reliability, and low latency the key aims for data center providers.

Though there is a wide spectrum of data center types, including those that are hybrids of the types identified in the table below, we think it is useful to group them according to four different classifications: retail colocation, wholesale colocation, enterprise, and managed services.

Table 1

Data center pricing

In terms of pricing, retail colocation tends to be associated with higher per kilowatt-hour (kWh) lease rates. While wholesale leases might fetch between $130 and $185 kWh/month, retail colocation leases--which offer more involved services--—may be priced between $220 and $365 kWh/month (note these ranges are illustrative only, and will vary widely based on the market, tenant credit quality, lease term, and other factors). Various reasons can cause the pricing differential between wholesale and retail colocation data centers, for example:

• Wholesale contracts tend to be longer and the churn for such lease contracts tends to be lower at around 1% per month, while retail colocation generally have much shorter contract terms;
• Wholesale contracts, by nature, have larger leased capacity and sometimes offer rent rebates and allow the tenant to phase into new space over several months (i.e., "ramp-ins");
• Level of service is typically higher for retail colocation tenants;
• Utilizations tend to be higher for wholesale agreements (around 90%)--given that the space is built pre-leased with large anchor tenants agreeing to rent once construction is completed--compared with retail providers (between 80% and 60% utilization) having to both sell to many more customers and finalize the leasable space before selling; and
• Tenants signed up for wholesale contracts generally have larger capital investments and have a better credit profile (some could be investment-grade rated companies), while tenants of retail colocations generally have a smaller economy of scale.

The role data centers play in company operations

Mounting connectivity and data needs with IoT and digital content distribution in recent years have brought about "edge computing," which is the practice of processing data away from a central server (such as in a traditional cloud arrangement) and instead closer to the outer rim of a network where that data is actually gathered. Keeping this process closer to the source of the data mitigates the distance the data needs to travel, and therefore alleviates latency in data transmission while simultaneously reducing the bandwidth clogging that would otherwise take place. Edge data centers primarily serve local markets, operate as part of a broader network, and are typically smaller than enterprise-level hyperscale facilities.

The critical need for data centers lies with the stable connection they offer their tenants. Therefore, any interruption in that access to data can have a deleterious effect on the underlying tenant's ability to conduct its day-to-day operations, even if it's brief. Such interruptions can be caused by power failure (most common), network failure, software issues, or IT systems errors. This prompts data center operators to invest in backup systems to step in and facilitate traffic in case such failures occur. The backup power available for emergencies is measured in terms of "redundancy." These redundancy levels are represented in categories: N+1 (one backup uninterruptible power supply [UPS]), 2N (double the amount of equipment needed with no single power of failure), and 2N+1 (double the equipment plus one backup UPS).

Data center tier classifications

The Uptime Institute classifies and certifies data centers into different tiers based on design and performance. A Tier 4 data center is considered the most robust and least prone to failures, designed to host mission-critical servers with fully redundant subsystems and compartmentalized security zones controlled by biometric access control methods. Naturally, the simpler non-redundant Tier 1 facilities tend to be used by smaller in-house businesses. The tiers are defined below:

• Tier 1 (no redundancy): Guaranteed 99.671% availability (28.8 hours of downtime per year)
• Tier 2 (partial redundancy): Guaranteed 99.741% availability (22 hours of downtime per year)
• Tier 3 (N+1): Guaranteed 99.982% availability (1.6 hours of downtime per year)
• Tier 4 (2N+1): Guaranteed 99.995% availability (26 minutes of downtime per year)

A particularly illustrative event occurred in September 2018 when a lightning strike in Texas overwhelmed a cloud provider facility's surge suppressors and disabled its cooling systems, which caused the hardware to shut down amid the raised temperatures. The cross-service dependencies present at the time were a vulnerability, as it took over 21 hours to recover the services that automatically powered down to safeguard the data. To put this in perspective, a study conducted by the Ponemon Institute in 2016 across 63 U.S. data centers estimated that the average cost of an unplanned data center outage was $9,000 per minute. The study estimated the average cost of a data center outage at $740,357. Additionally, a June 2018 Uptime Institute report noted that a third of all reported outages each cost over $250,000, and the increasing levels of complexity, sensitivity, and interdependence in modern data center operations have made resiliency against such failures harder to achieve.

How lease arrangements work

Data center tenants enter into various types of contracts, with the larger hyperscale contracts tending to be more customized to the tenants' needs. Contacts we've seen in the wholesale centers are typically structured as triple-net leases or modified gross leases. In triple-net lease arrangements, the tenant is responsible for operating expenses and electricity expenses, in addition to the base rent, as well as accompanying taxes and insurance to use the facility. In modified gross leases, the data center operator pays the operating expenses as well as the taxes and insurance, although such expenses are typically estimated at the outset of the contract and priced into the tenant's base rent. Retail colocation contracts tend to be full-service in nature--where the customer pays a fixed rent and the data center operator is responsible for the power costs--which leaves the operator vulnerable to power cost increases. Alternatively, providers of wholesale colocation typically have a higher percentage of contracts with "metered power" reimbursements due to the large size of these deployments.

At the same time, the state of the data center itself can either be tailor-made for a tenant (i.e., "built-to-suit") or instead be little more than an empty building with flowing electricity and fiber connectivity. The former option, a "turnkey data center," is a custom arrangement where the tenant pays rent based off power used ($/kW), and the tenant uses the backup power and cooling equipment that the landlord has installed. In this arrangement, the landlord bears the brunt of the installation, maintenance, and operating expenses (except where triple-net leases pass it down to the tenant). The latter option, dubbed "powered shell data center," is a building with no offerings beyond the basic shell, where tenants build and install their own equipment for backup power and cooling. The tenant owns and operates the equipment therein, and the rent payments are based off square footage ($/sq. ft.) rather than power use.

Table 2

The leases often include annual rate escalators (under 2%-3%), five-year renewal options, and explicit default remedies.

The most common ownership structures in data securitizations

The ownership structures that data center securitizations adopt can vary; however, we've seen certain patterns within this asset class. The data centers themselves are typically owned by distinct asset-owning entities (AOEs), which are structured as bankruptcy-remote special purpose entities. In addition to the data centers, the AOEs own the accompanying tenant leases and operationally critical inventory, systems, and software. The issuer, as a bankruptcy-remote special purpose entity, owns the AOEs and pledges all of its assets, including its ownership interest in such AOEs, to the bondholders. The manager is bound by the management agreement to maintain, market, operate, and administer the data centers.

Data center securitization ratings

S&P Global Ratings has rated four data center-based series across two master issuers to date: Vantage Data Centers Issue LLC (Vantage) series 2018-1, 2018-2, and 2019-1, as well as Stack Data Centers Issuer LLC (Stack) series 2019-1. The six to 10 data center sites included in the securitizations were spread among a handful of U.S. states, as well as Quebec City and Montreal, and ranged from less than 600,000 sq. ft. to over 1.1 million sq. ft. Of the 63-97 megawatt total leasable capacity in each deal, only 1%-14% of availability was vacant at the time of closing. We assigned our 'A- (sf)' ratings to the bonds, each of which had five-year anticipated repayment dates. There were 15-25 underlying tenants in each deal, with the top five accounting for more than 60% in annualized base rent (ABR), which is the cash flow paid to the securitization. The ABR generation among the different data centers was somewhat diversified, with less than 42% of ABR generated by one data center. In each deal, the majority of tenants were wholesale investment-grade businesses (i.e., rated 'BBB-' or higher)--including industries like cloud, big data, manufacturing, and media communications--that require 500 kW or more of power to operate their networks and equipment.

Table 3 shows the pool characteristics of the master trusts from both Vantage and Stack.

Table 3

The factors S&P Global Ratings uses in its analysis

Although the third-party data center industry itself is not new, data center securitizations are still very much in their infancy. As a result, S&P Global Ratings continues to reevaluate its analytical approach for rating such deals as we receive feedback from the market and observe additional performance data. So far, however, there are a number of considerations we expect to remain integral in determining our opinion of the timely interest and ultimate principal projected to be paid toward any data center lease-backed bond.

The explicit collateral underlying a data center securitization are:

• Primarily mortgages, deeds of trust, and deeds to secure debt creating first-mortgage liens on the interests in the data centers;
• A perfected security interest in all personal property and fixtures owned by the issuer's subsidiaries located in the data centers; and
• Any reserves and escrows related to the data centers.

The cash flows that service the rated debt stem from leases the underlying tenants pay to use the data center facilities. As with other securitizations that S&P Global Ratings rates, circumstances that might delay, compromise, or in any way hamper the cash flows from being generated and paying off the bonds play a key role in shaping our methodology.

Although historically low customer churn rates lessen the chances of a shortfall in lease cash flows, as do the high costs a tenant would incur by relocating to an entirely new facility, S&P Global Ratings assesses the scope of the manager's responsibilities. After all, if there's a critical function a tenant expects to be handled by the manager that is overlooked, it's possible that such a failure might interrupt the tenant's operations and therefore prompt that tenant's migration (as well as its ongoing lease payments) out of the data center. Such is the nature of an operating asset. The manager's operating duties include items like maintaining property, casualty, and business interruption insurance, paying real and personal property taxes, keeping data centers in compliance with laws and regulations, maintaining physical security, and providing uninterrupted levels of electricity, access, and cooling to the tenant detailed in the service-level agreements between the tenant and the data center manager.

Such priority expenses, operating costs, and maintenance capital outlays are carved out at particular dollar-per-kW rates in the securitization documents, with designated annual escalators each year. When these obligations are left unmet, this can result in financial reimbursement to the tenant for any resulting network interruption. Such reimbursements would therefore not be remitted to service the rated debt. As such, a data center operator's management team is crucial to S&P Global Ratings' analysis of future cash flows.

Typical credit factors like tenant concentration and credit quality, geographic dispersion of ABR-generating sites, the industry diversity present among the underlying tenants, and the remaining average contract terms (including early termination conditions) will continue to play a key role in evaluating the reliability of future cash flows over the rated bonds' life. The data center facility locations and the real estate and data center market conditions are also factored into our considerations. The existing market condition is reflected in the current vacancy and lease rate factors of the existing contracts, and future values are considered through the calculation of the liquidation value of the assets.

Cash flow and default projections

The primary drivers for projecting incoming lease cash flows for data centers are:

• The potential for the underlying tenants to default;
• The ability of the property manager to re-lease the data center properties vacated by defaulted lessees to new tenants;
• The renewal rate of lessees who reach the end of their lease;
• The rental rate and term of renewed leases and leases to new tenants; and
• The ultimate liquidation value of the real estate by legal final maturity, considering factors like market capitalization rates and the appraised value.

Given the 20+ year maturities of the bonds in data securitizations, we assume two recessions will transpire over the course of a given transaction. This effectively migrates the credit quality of the assumed tenants downward to the speculative-grade level (i.e., rated 'BB+' or lower) by the second recession. Doing so addresses the technology risk that existing data center infrastructure face, as potentially cheaper or more effective means of housing/accessing data develop.

Of those that we assume to default, we project all leases to be re-leased at a material discount, rather than be accepted into bankruptcy or liquidated. That said, the methodology we employ for single-tenant triple-net lease securitizations does not take into account the multi-tenant nature of the data center properties, which are unlikely to be liquidated based on the default of just a select few tenants. We also consider the operational incentive that tenants have in maintaining their mission-critical services rather than forfeiting them to bankruptcy proceedings.

In addition to running the aforementioned stresses in the cash flow runs to see if the projected cash flows are enough to pay ultimate principal and timely interest by legal final maturity, we perform sensitivity analyses to see how much a given deal can withstand before a shortfall in debt service hampers the deal's performance. Such sensitivity analyses will tack on more stress to operational factors like gross revenue reductions, heightened maintenance capital expenses, and increased annual priority expense escalators.

Why we don't employ our CMBS methodology in rating data center transactions

The quick answer is that we believe a property-level analysis doesn't cover all of the operational aspects specific to this asset class that can impact the creditworthiness of the collateral. However, there are several elements present in data center securitizations that resemble loans within CMBS transactions. The underlying collateral are a set of extensively built-out leasable properties, the S&P Global Ratings' values of which largely depend upon a derived capitalization rate (S&P Global Ratings considers industrial flex space the closest analogous frequently seen CMBS property type), and the underlying tenants are often publicly rated companies. In fact, we have seen loans backed by data centers among the largest loans in a few 2018/2019 conduit CMBS transactions, such as BANK 2018-BNK14 and CGCMT 2019-GC41.

That said, while our CMBS methodology utilizes rating-specific credit enhancement levels derived from loan-level stand-alone and diversity credit enhancement measures based largely on property-level analysis, the analysis we've employed in considering data center securitizations has largely been operational in nature. Instead of the economics of the buildings themselves being the primary focus, as in CMBS, the cash flow methodology we use in rating data center-backed securitizations incorporates our view on the importance of the data center operator and the specific ways they secure tenants' future lease payments.

Given the importance of projecting future periodic cash flows that depend on tenants' sticking with their current operators' data centers--rather than primarily on an ultimate discounted liquidation value--we review the operator's list of responsibilities. These include the aforementioned areas like maintenance over cooling towers and electric switches and generators, as well as the presence of redundant backup systems. We also consider the operator's history of countering emergencies that have threatened the servers' performance. In this way, it's clear that the creditworthiness of the data center-backed bonds can't be summed up by analyzing the properties alone; the operator's activities play a central role in ensuring the tenants have a reason to continue their lease regular payments. Put another way, while in some of these arrangements a tenant is renting real estate, they are also renting power. As such, the integration, maintenance, and capital improvements are also crucial in determining the likelihood of regular bond payments.

The relevance of the operators' performance becomes especially acute when we consider how the legal maturity profiles of data center securitizations (25 years) outlast those of typical CMBS loans, which are usually 10 years in tenor with a balloon payment due at term. This difference in refinance risk is, in our opinion, one of the most important differences between the data center ABS and CMBS structures.

How data centers are financed

Data centers entering the securitization foray is a relatively recent development. In the past, we've seen some of the larger players in the industry issue rated corporate debt. S&P Global Ratings maintains issuer credit ratings on the following entities:

• Digital Realty Trust Inc. (BBB/Positive)
• Equinix Inc. (BBB-/Stable)
• CyrusOne Inc. (BB+/Stable)
• Switch (BB/Stable)
• InterXion Holding N.V. (BB/Stable)
• QTS Realty Trust Inc. (BB-/Stable)
• Rackspace Hosting Inc. (B+/Negative)
• Dawn Acquisitions LLC (B/Stable)
• Ensono L.P. (B-/Positive)
• Flexential Intermediate Corp. (B-/Stable)
• DataBridge Parent Inc. (B-/Stable)
• Cequel Data Centers L.P. (B-Stable)
• Cologix Holdings Inc. (B-/Negative)
• Cyxtera DC Holdings Inc. (B-/Negative)
• Internap Corp. (CCC+/Negative)

Large investment-grade issuers, such as Digital Realty and Equinix, have primarily unsecured capital structures to provide maximum financial flexibility. However, there is a long trail of speculative-grade issuers that often rely on a mix of first- and second-lien secured credit facilities and unsecured notes for their financing needs.

S&P Global Ratings' outlook on the data center sector

We have a generally favorable outlook supported by increased IT outsourcing, exploding growth in data, and increasing application complexity. But not all data centers are created equal. We believe data center operators with significant scale, geographic diversity, capital resources, and interconnection capabilities will be positioned best to take advantage of this growth. In contrast, we expect data center operators with limited scale and geographic reach to have more modest growth over the next few years.

We believe demand for third-party data center space will continue to increase, spurred by increasing cloud adoption, with the benefit skewed toward larger-scaled players. Cloud service providers like Amazon Inc. and Microsoft Corp. will likely outsource much of their infrastructure, as wholesale providers such as Digital Realty and CyrusOne can generally build computing nodes faster and more efficiently. These providers will also need to expand access nodes, benefiting interconnected carrier-neutral players like Equinix and InterXion.

We also believe that enterprises will continue to outsource their infrastructure as they look to deploy hybrid and multi-cloud architectures, in which certain applications reside in colocation and others in various cloud environments. We believe there is a lot of runway for growth, as we estimate more than 60% of enterprises' IT equipment still resides in-house. However, the pace of change can be influenced by economic conditions, as shifting architecture requires upfront costs that may be deferred in a recession.

Finally, the cloud transformation creates a headwind for retail data centers, as certain applications that reside in colocation may be migrated to the cloud when the contract expires. This is because the cloud removes the need for significant investments in IT infrastructure, offers quick scalability, and provides easy access to data, which will cause elevated churn over time as enterprises eventually transition. Still, colocation will continue to be necessary for network providers that make a significant portion of its customer base and workloads that are particularly complex, have relatively fixed-usage levels, or have high-security/compliance requirements.

We believe data center operators that house cloud infrastructure could benefit from some downside protection, depending on their ability to offset lower enterprise revenue with incremental revenue growth from cloud service providers, while still indirectly serving the enterprise. Interconnection is another key competitive advantage that can attract enterprises as they look to connect with multiple cloud providers, networks, customers and/or suppliers. On the other hand, providers that lack interconnection tend to attempt to differentiate themselves with managed services, such as their own private cloud solutions. In our opinion, these operators are most susceptible to the shifting trends, given the low switching costs associated with managed services that carry lower profit margins due to higher operating costs.

Related Research

S&P Global Ratings research

• Presale: Vantage Data Centers Issuer LLC And Vantage Data Centers Canada L.P. Series 2019-1, July 24, 2019
• Presale: Stack Infrastructure Issuer LLC (Series 2019-1), Jan. 23, 2019
• Cloud Disruption: Cloud Adoption And Digital Transformation Are Positives For The Data Center Industry, Sept. 7, 2018
• Despite Continued Growth, U.S. Data Centers May Face Long-Term Risks From Financial Pressures And Uncertain Tech Developments, Oct. 30, 2017
• Analyzing The Real Estate Characteristics Of Data Centers, July 25, 2016

Other research

• Uptime Institute Focus, June 2018
• 2016 Cost of Data Center Outages, Ponemon Institute, Jan. 19, 2016