Our research indicates that network-dense data center REITs should continue to grow cash flow per share faster than their non-network-dense peers and the broader REIT universe. In addition, they carry lower supply, re-leasing, and obsolescence risk than other tech-oriented real estate. Finally, they trade at lower FFO and EBITDA multiples relative to many high-quality REIT peers, which builds a margin of safety into current pricing. As a result, we believe that the most network-dense data center REITs should outperform the MSCI US REIT index (Bloomberg: RMZ) for the foreseeable future.
A data center is any physical location where an organization stores its important information technology (IT) equipment, such as computer servers, data tapes, and network switches. Figure 1 depicts the prototypical data center design: the data center pod or data hall. A pod is composed of a large room, often with a raised floor, called a computer room. Leasable computer room space usually makes up 40-60% of a data center’s total square footage, and is measured by computer room square feet (CRSF).
Several rows of six to eight foot tall metal boxes called server cabinets are located at the center of the computer room, each of which holds dozens of stacked computer servers. Surrounding these cabinets are several pieces of industrial-strength equipment, each as big as a large refrigerator: power strips called power displacement units (PDUs), backup batteries called uninterruptible power supplies (UPSs), and HVAC units called computer room air conditioners (CRACs). The PDUs control the flow of power to the rows of cabinets, the UPSs provide temporary backup power until stronger backup diesel generators power on, and the CRACs keep the servers from overheating.
Each data center building can have one to ten of these pods, surrounded by an outer ring of shared spaces housing connections to at least one internet service provider (ISP, e.g. AT&T), each pod’s backup diesel generators, multiple layers of security, and conference rooms. The servers run 24/7 and are connected by miles of network and power cables that must be maintained.
Figure 2 shows a decision tree that outlines the three approaches companies can take to managing their IT real estate. In the cloud computing approach (1), firms outsource the highest percentage of their IT management. This typically lowers total IT costs per employee, has minimal up-front fixed costs, and allows companies to upgrade and back up their IT systems. However, it sacrifices control over hardware and can raise security and privacy concerns. In its 2016 survey of global IT professionals, the Uptime Institute found that only 9% of enterprises (organizations with more than 1,000 employees) manage the majority of their IT assets via cloud computing.
In the external data center approach (2), companies own their own servers but lease some of a data center pod and contract for varying amounts of equipment maintenance. This approach achieves much of the same cost savings, operational flexibility, and speed as cloud computing, but it is preferred by many companies because they can retain control over more sensitive hardware and data. However, it usually costs more than a pure cloud approach, particularly at the beginning. The 2016 Uptime Institute survey found that 20% of global enterprises have the majority of their IT assets in external data centers.
With the internal data center approach (3), firms in-source almost every aspect of IT management. Internal data centers include all single-tenant user-owned IT real estate, as well as industrial and office space that houses IT equipment.
Companies taking this approach must work directly with utilities and network providers to bring additional power and fiber to their sites, hire data center development, space planning, and construction teams, and permanently retain a full-time IT staff to acquire, install, and maintain the equipment. According to a 2016 Uptime Institute survey, 71% of global enterprises have the majority of their IT assets in internal data centers.
As one can imagine, a pure internal data center approach typically has the highest IT cost per employee of the three approaches and is widely believed to be inefficient for all but the most sensitive, customized, and secretive activities. The most well-known owners of giant internal data centers include Facebook (NASDAQ: FB), Google (NASDAQ: GOOG), and Microsoft (NASDAQ: MSFT).
External data centers can be divided into wholesale assets and retail assets. The key differences between the two are tenant size and lease duration.
Wholesale data centers are 50,000-1,000,000 total square foot (sqft) properties that primarily lease power, as opposed to server cabinets (retail data centers) or sqft (other real estate). Rental rates are usually quoted on a monthly per kilowatt (kW) or megawatt (MW) basis, with one MW being equal to 1,000 kW.
These owners prefer one to three tenants per building, but they occasionally accept up to ten tenants at a single property if necessary. Rents average $100-200 per kW per month, or about $240-480 per CRSF per year assuming 5 CRSF per kW. These rents sound high, but after backing out the 40-60% of data center sqft that is not CRSF and the high upfront investment in depreciable electrical and HVAC equipment that they do not have to buy themselves, they are not much different than traditional office and industrial rents. In all three sectors, tenants sign 7-20 year leases, have separately metered utilities, and buy, install, and maintain their operating and office equipment.
A wholesale data center costs $7-13 million per MW to build, or $800-1,500 per total sqft ($1,600- 4,000 per CRSF), which means a 350,000 total sqft project can cost over $500 million. Most players in the space spend about 5% of cost on land, 60% on movable landlord-owned equipment (e.g. UPSs, diesel generators, PDUs, CRACs), and about 35% on fixed real estate improvements (e.g. building shell, foundation, parking).
Wholesale campuses are typically built in phases, with the first speculative phase initially built out as a powered shell. A powered shell is basically an industrial warehouse with thicker walls, space for a lobby, a maximum security front office, and access to significant amounts of power and fiber. They cost $200-400 per sqft to develop, with tenants building out the interior of such properties with their own data hall designs and at their own expense. If a tenant prefers the landlord do the buildout, most experienced data center developers can take the building from shell space to a move-in ready turn-key data center in three months.
At lease expiration, wholesale landlords typically have little bargaining power due to the high percentage of their property value that depreciates each year, high concentration of tenants, and the constant innovation in data center design and IT equipment with which they must keep pace.
Wholesale data center REITs create value by having a more efficient leasing and development platform than their peers. One example of this is CyrusOne (NASDAQ: CONE), which spun off from Cincinnati Bell (NYSE: CBB) in 2013 and since then has created the most efficient wholesale data center development platform in the US. Since it has been independent, CONE’s cumulative yield on cost has never fallen below 16%, despite the company growing its total cost basis by more than 160% via development and accretive acquisitions.
According to North American Data Centers, a data center broker, CONE was responsible for five of the ten largest US data center leases in 2016, despite having only the third largest wholesale portfolio. None of its competitors showed up on the list more than once.
Retail data centers are typically 50,000-800,000 total sqft properties that primarily lease server cabinets, as opposed to power (wholesale data centers) or sqft (other real estate).
Retail rental rates are typically quoted on a monthly per-cabinet basis or in a monthly per-kW basis with a fixed allocation of power.
Retail properties are classified by their network density, which is measured by the number and size of the networks to which they provide access. The most network-dense retail centers are called internet exchange points (IXPs), while less network-dense retail centers are called retail colocation centers.
Chilton believes that IXPs have the best long-term investment attributes of all data centers. The prototypical IXP often looks like a large wholesale data center on the outside, however, in addition to the basic amenities of wholesale centers, IXPs also house the region’s main physical connection points to dozens of networks and carriers. These include ISPs, cloud providers, content delivery networks (e.g. Akamai (NASDAQ: AKAM), and content providers themselves (e.g. Netflix: NFLX).
One of the founders of Equinix (NASDAQ: EQIX), the largest data center REIT in the IXP business, compares IXPs to “international airports where passengers from many different airlines make connections to get to their final destinations.” Just as each large MSA only needs one large carrier-neutral international airport, each large MSA typically needs just one big carrier-neutral IXP (see Figure 3).
Both businesses are natural monopolies, primarily due to ‘network effects’, an economics concept that refers to the durability of an asset’s profitability increasing as its market share increases. Network effects tend to be self-perpetuating, particularly with natural monopolies.
IXPs generate the majority of their revenues by executing one to three year leases in small increments for high rents. They typically cost about $1,500-2,500 per total sqft to build.
IXPs also generate revenue by selling cross-connects in a process they call interconnection. Cross-connects are fiber or copper cables that connect the equipment of two different retail data center tenants. IXPs charge $200-400 per month per cable for the one-time act of stringing a cross-connect cable between the servers of different tenants or between tenants and carriers. It is a recurring-revenue business with a 95%+ profit margin that greatly increases the network-density of an asset. It also reduces re-leasing risk as the costs of moving interconnected equipment to another facility become increasingly prohibitive as more companies cross-connect at a facility.
EQIX dominates the global market for interconnection with 230,000 cross-connects, 50% more than the next five largest interconnection companies combined. EQIX and the next two most network-dense data center REITs, CoreSite (NYSE: COR) and Digital Realty Trust (NYSE: DLR), leverage the networks to generate the highest re-leasing spreads and rent growth in the data center universe at their IXPs. On the private market, IXPs trade for 5-7% cap rates (14-20x operating cash flow) versus the 7-10% cap rates (10-14x operating cash flow) at which other data centers trade, reflecting direct real estate investors’ recognition of their low re-leasing risk and high residual values.
In contrast to IXPs, lower-network density retail colocation centers most closely resemble smaller regional airports. They may have exclusive control over a few short routes, particularly to other small airports, but they don’t have enough runways, surrounding population, or carrier density to overcome the barriers to entry and compete directly with large international airports.
Retail colocation properties usually have some competitive advantage, whether it be geographic, infrastructure, or key tenant-based, which prevents them from having to compete with more commoditized wholesale space.
Rents are similar to turn-key wholesale, but, similar to IXPs, they sign much smaller, shorter-term leases to a larger number of tenants. While they have less pricing power than IXPs, retail colocation data centers are less risky than wholesale data centers due to their tenant diversification.
The data center sector has two huge secular demand drivers: data center outsourcing and the Internet of Things. As we discussed earlier, a 2016 Uptime Institute survey found that 71% of current IT assets still reside in company-owned internal data centers. We believe the compelling economics of data center outsourcing (cloud computing and external data centers) will cause this to change dramatically over the next five to ten years. That same Uptime Institute survey found that CIO/corporate IT executives expect the 71% number to fall below 50% by 2020. In other words, they expect the combined market share of external data centers and the cloud to go from 29% today to at least 51% in four years. The IT research firm Gartner (NYSE: IT) took their prediction out farther, concluding that by 2024 approximately 80% of enterprise IT assets would be in external data centers or the cloud by 2024.
If we haircut the more conservative Uptime’s prediction by 1/3, we are still looking at outsourcing’s piece of the pie going from 29% to 44% in four years. This implies 10-15% annual growth in demand for external data centers over the period, depending upon various assumptions (details available upon request). And that doesn’t account for any growth in the pie itself (i.e. total demand for IT real estate).
The second secular demand driver, the Internet of Things, is the network of web-enabled devices (e.g. smart TVs), content providers (e.g. Netflix), and the ISPs needed to make them work effectively. Per market research firm IHS (NASDAQ: INFO), there are currently 16 billion connected ‘things’ in the world, a number that is expected to more than triple over the next 10 years.
Every one of these devices requires a high speed internet connection to function, with most having several content providers that will each need outposts at IXPs around the world. YouTube and Netflix didn’t take off until internet connections were fast enough to stream high-quality video without extended buffering time, and the standard is even higher for self-driving cars, smart pacemakers, and self-flying delivery drones.
Finally, Figure 4 shows that network-dense data center REITs trade at discounts to their blue chip peers, especially when looking at 2019 consensus estimates. EBITDA and FFO are the best multiples for the space, as data centers calculate AFFO in a way that is not comparable to AFFO in other REIT sectors.
Their valuation discounts provide a margin of safety for investors that understand network-dense data centers REITs’ strong economics and the powerful tailwinds that are growing demand for their properties. Chilton believes that this should cause these REITs to produce higher total returns than their peers and the RMZ for the foreseeable future.
Matthew R. Werner, CFA firstname.lastname@example.org (713) 243-3234
Parker T. Rhea, email@example.com (713) 243-3211
Bruce G. Garrison, CFA firstname.lastname@example.org (713) 243-3233
Blane T. Cheatham, email@example.com (713) 243-3266
RMS: 1926 (4.30.2017) vs. 346 (3.6.2009) and 1330 (2.7.2007)
An investment cannot be made directly in an index. The funds consist of securities which vary significantly from those in the benchmark indexes listed above and performance calculation methods may not be entirely comparable. Accordingly, comparing results shown to those of such indexes may be of limited use.
The information contained herein should be considered to be current only as of the date indicated, and we do not undertake any obligation to update the information contained herein in light of later circumstances or events. This publication may contain forward looking statements and projections that are based on the current beliefs and assumptions of Chilton Capital Management and on information currently available that we believe to be reasonable, however, such statements necessarily involve risks, uncertainties and assumptions, and prospective investors may not put undue reliance on any of these statements.
This communication is provided for informational purposes only and does not constitute an offer or a solicita
Institutional Investment Strategies
Chilton Investment Outlook
Media & Press
Client / E-Path Login