Data centres in the AI age: Power, Place and Cooling

Maria Rampa: Hi, I’m Maria Rampa and welcome to this episode of Engineering Reimagined.

Today, we’re taking a closer look at something you rely on every day – but rarely see: data centres. These are the unseen engines behind everything from your morning emails to late-night streaming, online banking, and now – power-hungry AI models.

Data centres are often called the factories of the digital age and Asia-Pacific leads global data centre investment, demonstrating the region’s growing strategic importance.

As their use expands rapidly around the globe, so too does their demand for energy and water. Balancing this growing demand with access to renewable resources, and efficient and cost-effective design and delivery, is one of today's critical engineering challenges.

How do we resource this digital backbone without overwhelming our grids or compromising our climate goals? And what role can smarter design, renewable energy and water, as well as regional planning play in striking that balance?

In today's episode, Bree Miechel, Partner in Projects and Energy Transition at Ashurst, interviews Aurecon’s Industry Director for Digital Infrastructure, Simon McFadden. This episode was a joint recording with Ashurst’s Business Agenda podcast and part of a longer-term partnership between Aurecon and Ashurst.

Simon and Bree discuss insights into the escalating power and water consumption of data centres, how renewable energy and grid infrastructure challenges are being addressed, and innovative design strategies to reduce their environmental impact.

+++++

Bree Miechel: Hello everyone, I'm Bree Miechel and I'll be your host for this episode where we'll be focusing on the topic of data centres. I'm pleased to be joined by Simon McFadden, Thanks for joining us Simon.

Simon McFadden: Thanks Bree, lovely to be here.

Bree Miechel: Data centres are often described as the factories of the digital age. Can you shed some light on why their energy consumption is escalating so rapidly?

Simon McFadden: I like this analogy. I suppose there are two points in it. Firstly, are they factories? And secondly, why are they using more electricity? So on the first point, the analogy of a factory is probably a helpful one in that the large language models, which most people mean when they're using, when they're talking about artificial intelligence now, those large language models produce tokens or just bits of words or small words. And the AI data centre's function is to produce tokens. People have often compared this to intelligence. So, the production of tokens is the same as producing artificial intelligence. And if you want more of that artificial intelligence product, you need more tokens. And if need more tokens, you need more processing, and therefore you need electricity to produce them.

Bree Miechel: Is the need to develop data centres at scale though being overestimated globally?

Simon McFadden: The answer is, it depends. There are a few things to consider here. Firstly, are the data centres being used right now? So the existing ones, are they being used? And then secondly, how much AI will be used over time and what will the market structure be? I know that data centres have been built and are leased and in some cases are not using their full allocated power. Sometimes a fraction of this, right? But on the other hand, if AI turns out to be as powerful as people think, we'll be using it all the time for everything. So we're probably not building anywhere near enough data centres right now. And if you take account of the market structure, for example, some organisations are building data centres for them to use themselves or a large cloud provider building it for themselves to use. But others are building data centres to rent, which you know is similar to property development really. I think that in practice there's an overlap between this demand, meaning that there will be some duplication there in terms of the sites that are planned, permitted, and so on but I would hope that the bulk of the sites that actually go through and are constructed are used. I would say if I summarise everything, I think there's more development to come.

Bree Miechel: And with this demand, many regions are facing challenges in providing sufficient grid power. We've seen countries like Ireland impose moratoriums on new data centre grid connections due to capacity concerns with requirements on operators to develop captive generation to meet their power needs. Are we witnessing similar trends elsewhere?

Simon McFadden: Singapore has had a well-known moratorium on data centre development for some time, with some recent, in the last couple of years, exceptions to that where small projects are being developed. There are also power constraints at local levels right around the world. So if you think about areas where many data centres are being developed, for example, the data centre alley in the US. There's very significant demand on power and some constraints there. The challenge for utilities is twofold in that having sufficient generation available but also getting it to the right place. And I think if I reflect on our challenge in Australia with the goal to get to 82 per cent renewables by 2030, our challenge will be getting enough power at the right availability to the right place to match up with the demand for digital infrastructure.

Bree Miechel: And how can the design of data centres influence energy consumption and efficiency?

Simon McFadden: There are different elements to it, of course. There's been quite a bit of work done on the building itself to have it as efficient as possible. And people often talk about PUE, or power utilisation efficiency. That has been improving over time to make sure that all of the things that are not computer chips or servers are as efficient as possible in the building. But also with the computer chips and servers, the way the software is used and the hardware itself can make a significant impact on power efficiency. So, we've seen that NVIDIA, for example, often talk about the increasing power efficiency that their chips have over time. So whilst the power demand of those chips is increasing because they're doing more work, they're becoming more efficient in the production of the tokens that we talked about earlier.

Bree Miechel: Just honing back in on the challenges for grid infrastructure across jurisdiction and the fact that most new energy generation is being established in regional areas where the best resource is located and land is available, could data centre siting offer a strategic solution to alleviate both the inadequate or no longer fit-for-purpose grid infrastructure and that curtailment risk that that poses for power producers?

Simon McFadden: That's a really interesting point, Bree, and one that I've been discussing with many clients over the last few years, right? So can you put a data centre or many data centres in a rural area, say in Australia, where there's a lot of, a very large renewable energy development? The answer is so far, if you observe the behaviour of the large cloud providers, the large hyperscale or cloud providers, they typically need to have their data centres close to a population centre. They're typically in what they call an availability zone. So there's usually two, three, four data centres close together to provide redundancy. And they often need skilled workforce nearby. And they also need to have supply chains that work effectively. So these buildings have a lot of computers or servers in them. They need to be able to get those in and out easily. And all of those factors together have meant that most data centres, at least the large ones, are close to population centres. I'm seeing conversations where people are talking about developing AI specific data centres in more regional or rural remote areas even, which would afford them the ability to have a behind-the-metre, so a co-located energy project, renewable energy project together with an AI specific data centre. Potentially even not using the data centre much at night when the solar energy isn't generating.

Bree Miechel: In our powering change survey, 77 per cent of respondents viewed renewable energy investment as essential to their strategic growth, underscoring the commitment of organisations to drive forward lower carbon intensity business operations. Can you explain how this has extended to the data centre sector?

Simon McFadden: If you think globally, some of the most progressive firms that we see on renewable energy and carbon emissions would be American hyperscalers like Microsoft, Google, and Amazon. They require their teams to focus on the use of renewable energy. They all have hard targets that they've either achieved or are close to achieving for 100 per cent renewable energy by certain dates. And they're often involved with the development of renewable projects or sometimes they lead them to source that energy. We have organised power purchase agreements and attest to their desire for renewable energy. They all want it. The challenge is that it's not available 24/7. And so often this energy is not time matched. But in terms of total megawatt hours going in and coming out, they have it covered. I think over time, as we see more storage, that balance will improve.

Bree Miechel: With the intimate nature of renewable sources and the need to ensure a stable power supply, are data centre operators turning to utility scale battery storage solutions?

Simon McFadden: Pretty much all data centres have batteries as part of their design to ensure they don't go offline. But these batteries are typically, their capacity really just covers short periods until the generators can be started and come online. We've looked at battery storage for data centres as part larger developments of campuses with renewable energy development. The challenge that the developers have is to have energy on all the time. But as you add more battery capacity, obviously the capital expenditure for that increases.

Bree Miechel: But power's not the only challenge in the design and siting of a data centre. Let's also talk about water. Data centres require significant amounts of energy and water for cooling. In regions where water scarcity is a concern, how are data centres managing this resource?

Simon McFadden: It's become more and more relevant in the last few years. I do reflect as an engineer, we have plenty of water around Australia. And most people are living near the coast. We've plenty of solar energy available to turn it into drinking water. So, you could argue it's a cost and movement issue in the long run. On the whole, the data centres need to get rid of heat. You can do that simply though to the air using free cooling or air cooling, or you can use liquids. And there's really a trade-off between how much electricity you use and how much water you use. It's interesting that we're seeing some clients that are considering water that is treated but not to potable or drinking water levels. And this can be more economic and isn't taking supply from drinking water. When considering the water supply, we consult with local water authorities and understand whether water is needed for drinking water and they consider that when giving permits out for water and so on.

Bree Miechel: And in Europe, data centres are being pressured to reduce water consumption to a maximum of 400 mL per kilowatt hour of computer power by 2040. Would similar efforts likely take place in the Asia-Pacific region?

Simon McFadden: I worry about specific rules like this that have one metric because I often find they have unintended consequences in other areas. So, you know, it might, in practice, if we limit water use, it might increase energy use and maybe has another consequence. But I recognise that's a challenge for regulators.

Bree Miechel: APAC is becoming an increasingly attractive region for data centre investment and operations, evidenced by the rapid expansion throughout Asia and the recent sale of AirTrunk to Blackstone. What are some of the unique challenges facing data centres in this region?

Simon McFadden: The thing that comes to mind, Bree with that question, is that people outside Asia-Pacific sometimes talk about it like it's one country and it operates like a country. Obviously, Asia-Pacific isn't one place. It's a vast region with very different topology, climates, cultures, economies. And from a practical infrastructure perspective, different needs. So, if you want to develop a data centre in Johor that's different from Sydney and again from Jakarta. We're operating in 11 countries around Asia-Pacific and even inside those, there are different permitting requirements, climatic conditions, populations, and therefore different data centre designs. And I think that's different in some respects to other locations outside APAC.

Bree Miechel: What lessons could data centres learn from other sectors such as energy and resources that have been focusing on decarbonisation for some time?

Simon McFadden: I think there's an opportunity for the data centre developers because they're really at the early stage of their large-scale infrastructure development phase that they can use more renewable energy than historically our energy and resources sectors have been able to. So they can have a lower carbon set of infrastructure assets than industries that developed 20, 30, 50 years ago.

Bree Miechel: It's certainly clear Simon, that while there are significant challenges, there are also innovative solutions on the horizon. Before we draw this episode to a close, would you like to share some final thoughts with our audience?

Simon McFadden: On a personal level, I'm excited that I've been lucky enough to work in an area that combines the main pieces of experience that I've had in power generation and telecoms and now data centres, and also that it's enabling the next generation of technology with AI, which I find very interesting. And to anybody listening out there, I'd encourage you to look at the sector and it's thriving, so get involved.

Bree Miechel: Thanks very much, Simon. And thank you for sharing those thought-provoking insights with us.

Simon McFadden: Pleasure, thank you.

+++++

Maria Rampa: We hope you enjoyed this episode of Engineering Reimagined.

Don’t forget to hit subscribe on Apple Podcasts, Spotify, or wherever you get your podcasts, and follow Aurecon on your favourite social media platform to stay up to date and join the conversation.

Until next time, thanks for listening.

Data centre design for a data-hungry world

Data centres are often called the factories of the digital age, powering everything from cloud computing to artificial intelligence.

Aurecon’s Industry Director for Digital Infrastructure, Simon McFadden, joins Bree Miechel from Ashurst to explore the vital role of data centres, the growing demands they place on energy and water resources, and the critical engineering strategies being employed to enhance efficiency and sustainability.

“Data centres need to get rid of heat. You can do that simply though to the air using free cooling or air cooling, or you can use liquids. And there's really a trade-off between how much electricity you use and how much water you use. Our challenge will be getting enough power at the right availability to the right place to match up with the demand for digital infrastructure.”

Simon discusses how innovative approaches in data centre design can help address data centres’ increasing energy and water demands, current grid infrastructure limitations, the role of renewable energy and much more. This episode covers exciting opportunities in the digital infrastructure sector, highlighting its critical role in supporting rapidly expanding technologies like AI.