Data Centers: Essential infrastructure for digital lives

Real Assets Research team
PDF 372.9KB
01 March 2017
  • High-definition video to drive huge growth in data consumption; driverless cars and virtual reality long-term drivers;
  • Cloud at the centre of data processing and delivery, as adoption increases and data centre operators react;
  • Investors must decide how much risk they want to take – there are a wide range of potential returns available.

We are all generators of data. From the smart phones in our pockets to our watching of TV on connected devices, the amount of data we consume and generate is rising fast. The photos and videos uploaded to Facebook and Instagram, messages sent via What’s App and WeChat or movies watched on Netflix lead to one common theme – the data is stored or accessed via a data centre.

Data created at an accelerating rate

Growth in data produced globally shows no sign of slowing. According to the Cisco Global Cloud Index, global data centre traffic was around 1.8ZB (Zettabytes – 1ZB = 1 billion Terabytes) in 2010 and is expected to grow to 15.3ZB by 2020: 2 representing an annual growth rate of more than 30%.1

In the past, file sharing accounted for a large proportion of Internet traffic but video, which accounted for 63% of all internet traffic in 2015, is expected to grow to 79% of all traffic by 2020.2 Netflix has been a driver of the demand for video content, with almost 70,000 hours of video viewed around the world every 60 seconds.3 The move from standard definition to high and ultra-high definition content is likely to increase the amount of data passing through data centres. Moreover, increasing numbers of people are subscribing to this type of service.

While video content is driving demand now, in the future, driverless cars (AVs – Autonomous Vehicles), Virtual and Augmented Reality and the Internet of Things could lead to stronger growth in data generation. Intel forecasts suggest that by 2020, each AV could be generating around 4,000GB (gigabytes) of data every day, compared to the average Internet user generating 1.5GB.4 Based upon relatively conservative estimates,5 over the next 25 years, AVs could represent around 50% of the world’s car fleet, generating around 2ZB of data every day. That is more data, from this single source in one day, than was generated in total throughout the whole of 2011.

Already, governments are beginning to legislate for AVs and connected vehicles, with recent pronouncements from the US Department of Transport and the UK government.6 Those doubting that this technology is ready need look no further than Google and Tesla. Google’s AVs have driven more than 2 million miles in the US, while Tesla cars using its semi-autonomous Auto Pilot mode have driven more than 220 million miles. As take-up increases, data generation should too, placing pressure on data centres, which are likely to process and store these huge volumes of data.

The Cloud is at the centre of data delivery

The nature of data traffic is undergoing a major transformation. This is a result of the increasing prevalence, and adoption, of cloud applications, services and infrastructure. Cisco estimates that by 2020, more than 90% of data centre traffic will be Cloud traffic. In addition, forecasts from Gartner show spending on Cloud services was expected to reach USD204bn in 2016, compared to USD175bn in 2015. In all, more than 70% of enterprises are adopting hybrid Cloud services in some way.7

The unprecedented amount of data generated by people, machines and ‘things’ means growth in Cloud services should accelerate. The Cisco Global Cloud Index estimates that 600ZB of data will be generated by 2020, up from 145ZB in 2015. Most of this data is likely to be ephemeral in nature and will be neither saved nor stored (much of the ephemeral data is not worth saving), therefore, bypassing data facilities. However, Cisco estimates that approximately 10% could be useful, meaning that there would be 10 times more useful data created (60ZB, 10% of the 600 total) than would be used (6ZB) in 2020. This shows that organisations generating useful data still have some way to go in capturing and harnessing the benefits it could bring.

Operators have responded to this shift, with the development of Cloud-focused data centres, aimed specifically at delivering data across a Cloud network. In addition, hyperscale data centres, made up of small, individual servers (nodes) that are clustered together and managed as a single entity are increasingly common. These systems increase both efficiency and processing power, to the benefit of end-users and operators.

Connectivity and power availability are vitally important and drive the location of data centres around the world. Arranged in hub and spoke locations, hubs are generally in areas where power and connectivity is added cheaply (outside major cities). Spokes are smaller and tend to be within cities, with links across the various data centre networks. Further out are Edge data centres, serving smaller cities or big towns at the edge of networks.

The real estate solution

Real estate investors are able to capture this growth in a number of ways, providing the vital infrastructure needed to deal with data creation on such a large scale. Real estate investors have several options when it comes to accessing data centre investments. In their simplest form, the options are to either replicate the traditional landlord model or to move up the risk curve and invest directly in a data centre operating platform.

Landlord model

The traditional landlord model is perhaps the simplest. The real estate investor can choose to buy or develop a shell, with or without power, with internal facilities paid for and owned by the operators (tenants). In such an example, the real estate investor carries no technological obsolescence risk, as this sits with the tenant. Given their investment in the fit-out of the building, the tenant often signs an initial lease for an unbroken term of 15-years plus, with many tenants remaining in facilities for an extended period, should there be no operational issues.

Tenants utilising an asset-light data centre strategy often tend to renew leases, as moving in and out of facilities can be expensive, particularly given the cost of the internal IT equipment.

There are, however, still risks associated with this type of investment model. Tenants can default on rental payments or, in extreme circumstances, seek to renegotiate a lease or leave at lease expiry. In the event of a tenant vacating, due to the nature and complexity of moving a data centre operation, a landlord would typically receive two to three years’ notice.

Colocation/Retail model

Retail colocation could be defined as under 250kw of capacity and typically shorter-term contracts (1-5 years), whereas wholesale colocation is larger scale requirements of between 250kw and 40MW in some cases, and service contracts range from five to 15 years.

In the retail model, the operator provides services down to the computer rack level and installs physical separations between tenants, usually in the form of a cage. The operator provides additional services such as power, telecom and physical IT engineering services known as “hands and eyes”, which ensures the customer does not need its own staff to travel to the facility to undertake work on their servers.

In the wholesale model, the operator can lease larger space and capacity directly to an end user and the service obligation will stop at the power distribution unit (PDU) for the space. It is not uncommon to find an asset-light retail colocation operator leasing space from a wholesale colocation provider.

Colocation was generally provided on either a retail or wholesale basis, with the two models now merging. The colocation model carries a greater degree of exposure to the operating business of a data centre, so the risks and potential rewards for investors are greater. The colocation model, therefore, carries significant obsolescence risks. However, potential returns associated with this activity can be commensurate with the risks. For inexperienced data centre investors, pricing these risks can be difficult. In order to fully underwrite a potential investment, detailed operational knowledge is often required. A dual landlord and colocation investment model, spreading risks between the two strategies, could mitigate this risk.

A sector easily over-complicated

Data centres remain a relatively complex asset class when compared to traditional commercial real estate. However, much of the complexity is due to the technology within the data centres themselves and the use of jargon when detailing technical operations, making the sector easy to over-complicate. That said, underlying growth drivers are easier to understand and remain strong, being counter-cyclical and relatively protected from the broader economic environment. Furthermore, with technological advances such as driverless cars, virtual reality and the ‘Internet of things’ almost upon us, growth could accelerate even further.

The opportunity exists for real assets investors to benefit from the strong growth story currently unfolding. Investors able to participate in this growing market could have the opportunity to take advantage of a non-traditional, yet increasingly attractive growth opportunity.

1 Cisco Global Cloud Index
Disclaimer: © 2017 AXA Investment Managers and its Affiliated Companies. All rights reserved. These pages are not intended as an offer or solicitation, or as the basis for any contract, for the purchase or sale for any fund or instrument. The information contained in these pages shall not be deemed to constitute advice and should not be relied upon as the basis for a decision to enter into a transaction or for any investment decision. The analysis expresses the views of AXA Investment Managers Real Assets Research and Strategy team and may be subject to change without notice. AXA Investment Managers expressly disclaims any responsibility for (i) the accuracy or completeness of third party data (ii) the accuracy or completeness of the models or estimates used in deriving the analyses, (iii) any errors or omissions in computing or disseminating the analyses or (iv) any uses to which the analyses are put. The information contained in this document may not be reproduced or circulated without our written authority.