Navigating power demand in the age of AI

The continuous growth of AI services since the launch of ChatGPT in 2022 means that data centre power demands will continue to rise. In fact, the International Energy Agency’s (IEA) Electricity 2024 report predicted that global data centre electricity use could double by 2026, reaching 1000 terawatt-hours (TWh) – that’s equivalent to the annual electricity consumption of Japan.

So, data centre professionals face several challenges when looking to meet that demand in a cost-effective, sustainable way. Here are a few considerations.

Think modular

If you’re building or enlarging a data centre, take advantage of a modular approach. Modularity offers a way of scaling sustainably in a way that meets the demand for power and availability while simplifying the specification and installation process. 

Rather than constructing or expanding a data centre all at once, you can build it up in blocks. As an example, you can build a projected 200-megawatt (MW) data centre in 20 MW sections – onboarding customers as demand increases. That makes it more cost-effective because it avoids wasted capacity along the way.

It’s also an approach that minimises downtime and risk, especially with prefabricated solutions – like skids and eHouses – that are ready for quick and easy installation as they are constructed off-site and then factory-tested before delivery. Besides, buying one prefabricated product instead of a range of individual components that require assembly saves money and time.

This approach is favoured by local utilities and governments too, with the smaller incremental growth of a data centre more likely to be approved than the full-scale build of a new one.

Consider shifting to medium voltage

Medium-voltage (MV) equipment has become increasingly viable and cost-effective especially in the case of rising power demand . 

MV UPSs, for example, can be installed modularly. As an example, ten 2.5 MW UPS blocks can be wired in parallel to create a 25 MW system. This enables faster deployment and increases overall system capacity without any additional complexity. It also avoids the challenge of having multiple LV UPSs that need to be regularly maintained and serviced. In addition, the lower currents at MV level mean cables can have a smaller cross-section, leading to additional savings.

Furthermore, MV UPSs are often more energy efficient than their low-voltage (LV) counterparts, and can provide power protection for the entire data centre, not just the server racks. So, switching to MV can offer long-term cost savings.  

Be more sustainable with BESS

Meeting sustainability goals is a challenge for data centre operators when it comes to the increase in power demand. 

One solution is using battery energy storage systems (BESS) to reduce reliance on diesel gensets as they can integrate renewables, like wind and solar energy, into the local energy mix. By using a BESS, excess energy from solar panels on the roof can be stored and used at another time, optimising renewables usage.

Using BESS can also provide load shifting and frequency response services to the grid, further aiding negotiations with the local utility, and creating potential new revenue streams via give-back schemes during peak demand.

First, though, it’s important to plan carefully and consider what your energy demands are to avoid over-specifying – after all, specifying for five hours of uptime would require a lot of batteries.

Make use of AI

The onward development of AI might be driving the increase in power demand, but it can also help to make data centres more reliable and efficient. This could happen through optimised cooling, predictive condition-based maintenance, data access and transfer, and demand balancing.

The key to this is connectivity. An automation system running an AI suite can keep track of the hundreds of thousands of monitoring points deployed in a typical mid-to-large data centre. This provides operators a 360 overview of the data centre’s performance, energy use and asset health at any given point. 

This data can then be used to make efficiency improvements. Take the example of a cooling system. While the upstream chiller and the distribution system are often viewed separately – leading to operational inefficiencies when attempts are made to make it more efficient – an automation system lets operators see holistically how one part affects the other. As a result, operators can make more informed decisions to improve overall efficiency.

Invest in SF6-free equipment      

Moving to SF6-free equipment now will ensure that you’re prepared for incoming regulations on the use of electrical equipment which contains the insulating gas. This will make your sustainability reporting easier. Regulations have been proposed because SF6 can leak to the atmosphere, where it has a global warming potential around 25,000 times greater than CO2. 

Adopt a TCO mindset

When considering how to grow efficiently and sustainably, data centre managers need to adopt a total cost of ownership (TCO) mindset.

This involves calculating and assessing all the direct and indirect costs of an asset over its entire lifecycle to determine its cost, as opposed to a more traditional view of separating capital and operational expenditure. By viewing expenditure with a TCO mindset, future operating cost savings are seen as net present value.

A simple way of looking at this is where you have a machine that will be running continuously for around 20 years; by investing in a more energy efficient model now, you will see significant cost and emission savings over the long term. 

Power demands might be growing as AI and digital services expand exponentially, but data centre managers and operators can meet these challenges with the right strategies in place. While these challenges will become more urgent as demand increases, acting now, with a TCO mindset, will help to ensure data centres continue to make a positive impact.