07 May 2024
Steven Carlini, VP Of innovation and data centre, Schneider Electric
The UK’s data centre world is an interesting place to be right now, and one facing critical pressures stemming from the need to expand to support more AI – particularly when it comes to training those in the ecosystem - and the overwhelming unmet demand for capacity. Today’s need for data centre capacity is literally accelerating faster than we can build it.
There’s also pressure being put on data centres for sustainability reporting. In Europe it’s becoming mandatory to report environmental data, and impossible to gain new permits unless the data centres meet certain efficiency designs and utilise renewable energy. In the UK, meanwhile, data centre operators are coordinating on how to manage peak power situations.
Prioritising sustainability
Building the most sustainable facility means exploring efficiency, water use, and circularity.
There are three ways to do that: utilising the most efficient architectures; using the most efficient components; and managing the deployed facilities efficiently. But operators are also having to start recording and minimising water use, while exploring recyclability and building things in the most circular way. One of the things a lot of operators are doing is extending their IT refresh cycles from three years, some to as high as ten years.
“Today’s need for data centre capacity is literally accelerating faster than we can build it.”
Almost all data centre operators have either carbon neutral or net zero commitments that they’re trying to meet; not just because of the regulations, but because of power supply issues. When designing a traditional data centre, there’s UPS, and there’s a generator backup – and management of these resources becomes a priority. The operator might wish to use these resources for demand response for peak shaving, or to harmonise the pitch, or to enhance sustainability. For example, management software enables operators to automatically charge backup batteries while running on renewable energy only, and then switch to using this stored renewable energy at times when the data centre switches to fossil fuels.
In the search for sustainability, distribution management software is going to be key for power management for utilities.
Accommodating AI
As AI applications become more integrated into business operations, and they start doing more with video inputs - HD facial recognition, traffic control, etc. – that data is going to need to be processed at a central location, and in real time.
This will drive the move to a local edge, although the rollout is going to take many years. These edge data centres will incorporate cloud applications, with one or even multiple clouds. We see that as a catalyst system driving deployment. How fast depends on how fast AI is integrated into business applications…
When it comes to deploying edge AI inference servers at scale, it’s not really the size of the data centre - which some people believe must be a single rack data centre to be considered edge – it’s about the central location.
Data centre design is already changing to accommodate AI. The servers are different, in fact they’re very similar to high performance computing type workloads where they’re just crunching as much data as possible.
With workloads like these, the servers run very close to their design capacity all the time. In a traditional data centre, you may design it to run a 12kW per rack, and normally it runs at 3kW per rack, and it spikes up occasionally. With AI workloads, if they’re designed for 100kW per rack, they’re going to run at 100kW per rack. Accordingly, the designs on the power and cooling systems must be accurate to support those workloads. We’re seeing higher power components in a small dense area, requiring some serious cooling systems designed for that density.
Hot topics for 2024
AI is front and centre for data centres and networks this year, but sustainability is still going to be important. The coordination of the power sources and supply is another big issue, particularly in the UK.
“Building the most sustainable facility means exploring efficiency, water use, and circularity.”
People are worried that AI is going to take over the world, take over jobs, and do so with dubious ethics. But realistically speaking, it’s going to take a decade of capacity building before we can enable some of the applications people are worrying about today. Just look at chatbots; AI is still very young, and we can all recognise it immediately. It won’t be taking over the world any time soon!
