Limits to off-premises data center growth in some areas require new concepts
Many different factors are driving the global growth of off-premises data centers. Most user organizations accept that the operation of an on-premises data center is inefficient below a certain size (<1,000sqm). The use of hyperscaler clouds and SaaS solutions requires broadband access and, for many workloads, redundancy in the data center outline. Additionally, hyperscalers and hosting providers use housing space for their offerings. All these factors have been driving constant growth in the data center housing market over the past few years and there are no signs of a slowdown in the foreseeable future.
The limits of growth
Growth is limited in some areas for three main reasons:
- Power supply limitations – Data centers consume large amounts of electricity, which is why housing providers specify the size of their data centers in MWh of available energy rather than in sqm. This high energy consumption is an increasing challenge for utilities. For that reason, two large data center projects in Ireland, for example, have been stopped. Microsoft and Amazon are not allowed to build new data centers in Ireland because the grid operator EirGrid expects they would have led to power outages. No new data centers in and around Dublin will be supplied with electricity before 2028. The same issues have arisen in Singapore.
- Shortage of suitable land for building – The requirements on data center locations are high. Floods, earthquakes, and hurricanes have to be ruled out. There has to be secure supply of water and, even more critical, electricity (one reason for the popularity of Frankfurt/Main, Germany, is that the southern and northern European power distribution grids meet there, meaning that data centers can easily be redundantly connected to both subgrids, therefore making a power outage much less likely). Last but not least, connectivity in the form of Internet, carrier Ethernet, and the like has to be available with maximum bandwidths.
- Community resistance – More and more societies focus on sustainability and the environment. Data center operators have to be good neighbors, explaining why their campuses will benefit the community and not cause noise-related issues, be a monstrosity, or be a drain on precious resources such as water and energy. The initial approach is based on ideas such as the free transfer of waste heat to local district heating networks (although, in many cases, this is no more than a fig leaf) and the use of renewable energies. Lower land use through higher densities in data centers and underground data center space on suitable sites can also contribute to acceptance. Sooner or later, however, the question will be whether it generally makes sense to build data centers near residential areas. Amsterdam shows how designated areas for data centers can be used to find solutions that satisfy everyone involved.
Growth will continue in some form
The need for more data center space is a given. If existing data center hotspots such as Phoenix and North Virginia in the US, London, Paris, and Frankfurt in Europe, and Mumbai, Tokyo, and Hong Kong in APAC cannot handle the growth and customers are not willing to pay excessive location surcharges as a result of limited ramp-ups, other locations will be found that are happy to develop new capacities. Examples could be Dallas, Chicago, and Atlanta in the US, Madrid in Europe, Johor in Malaysia, and Melbourne in APAC. In all these cities, data center space grew significantly in 2022 and will continue to do so in the foreseeable future.
There are five solutions to overcome the limits to growth in some areas:
- Higher density in existing data centers – High-density data centers (>80kW/m2) massively reduce space requirements, but also require completely different cooling concepts that older data centers can hardly provide without massive refurbishment. Blade servers are the norm, but blades with multiple (more than two) CPUs are still not always used.
- Workload-specific CPUs/server hardware – The hyperscalers have started to design/ manufacture their own processors in addition to their own hardware, some of which are optimized for very specific workloads, such as deep-learning training, and promise significantly higher efficiency. Some of these solutions are available to the general market as well.
- Green IT – Measures related to facilities and cooling in particular can help to reduce PUE to below 1.2, and thus considerably reduce the energy consumption of data centers. Water cooling and free cooling are available but are not used in all data centers as yet. Older facilities in particular need refurbishment. In addition, data centers should use 100% renewable energy.
- Workload distribution – There are many reasons, such as convenience, why it is preferable to have all workloads in one data center (and have a second data center for redundancy). However, if data center space is scarce in tier-1 locations, and therefore prices are high, it might be an economically favorable option to move specific workloads (e.g., workloads with lower connectivity or response time needs) to tier-2 locations. This approach can be complex, as dependencies have to be understood and addressed. The one-data-center idea has been diluted anyway, due to an environment of growing SaaS, cloud computing, and edge computing use, with distributed clouds and multi-access edge computing.
- Designated areas for data centers – Cities and regions such as Amsterdam and Frankfurt (currently under discussion) show possible future models for data center growth – data centers that are not in the immediate vicinity of residential areas but have redundant power supply and network connections. This provides location advantages such as available employees, Internet nodes, and proximity to other data centers and cloud providers, without affecting residents. On the other hand, it makes frequently discussed concepts for waste heat recovery even more difficult. Temperatures are generally too low for industrial purposes. Sufficiently high temperatures cannot be achieved with reasonable effort even with heat pumps, and for low-temperature district heating (70-80°C), close proximity to consumers is much more efficient.
If you want to share your feedback or personal experience, please leave a comment or drop me an e-mail.