Environmental Audit Committee
Oral evidence: Risks and opportunities to the sustainability of data centres in the UK, HC 22
Wednesday 10 June 2026
Ordered by the House of Commons to be published on 10 June 2026.
Members present: Mr Toby Perkins (Chair); Jonathan Davies; Sarah Gibson; Alison Griffiths; Sojan Joseph; Manuela Perteghella; Dr Roz Savage; John Whitby; Sammy Wilson.
Questions 1-39
Witnesses
I: Christina Christopoulou, Senior Manager, Infrastructure, Energy and Environmental Policy, Amazon Web Services, Steen Stewart, Senior Advisory Consultant, Crown Hosting Framework Authority, and Michael Birtwistle, Associate Director of Law & Policy, Ada Lovelace Institute.
Written evidence from witnesses:
- Ada Lovelace Institute (DCU0075)
Witnesses: Christina Christopoulou, Steen Stewart and Michael Birtwistle.
Chair: Welcome, everybody, to the latest panel session of the Environmental Audit Committee. This is the first session of our new study on data centres. I am very pleased that we have two excellent panels in front of us today. I will start by inviting our panellists simply to introduce themselves and their organisations, and to give their most basic perspective on the need for data centres.
Michael Birtwistle: Hello, and thank you very much to the Committee. My name is Michael Birtwistle. I am associate director for law and policy at the Ada Lovelace Institute, which is a philanthropically funded independent research institute with a mission to make AI and data work for people and society.
Christina Christopoulou: My name is Christina Christopoulou. I am a senior manager of infrastructure, energy and environment policy at Amazon Web Services for Europe, the Middle East and Africa. As part of my role, I engage with policymakers on the energy, infrastructure and water dimensions of our data centre operations. Sustainably meeting the demand we see across our customer base is one of the key features of how we design and operate our infrastructure.
Steen Stewart: Good afternoon. I am Steen Stewart. I am the senior technical adviser working for GCA supporting the Crown Hosting Framework Authority in delivering that framework on behalf of the Cabinet Office.
Q1 Chair: Can you explain what the Crown hosting framework is, for those watching?
Steen Stewart: The Crown hosting framework basically allows public sector entities to buy data centre space through our supplier, Ark Data Centres. We sit in the middle, between the Cabinet Office, the supplier and the customer, to make sure that everyone does as they should do.
Q2 Chair: So you are a public sector body providing data centre services to the public sector.
Steen Stewart: Yes, absolutely.
Q3 Chair: Can I start with you, Ms Christopoulou? To get to brass tacks, for the benefit of those watching, it feels like data centres are something we have spoken about quite a bit and that have entered the national lexicon. I am not sure everyone necessarily understands whether we are talking about an increased version of the old server room that was in every company around the country, or something a bit more than that. Can you give us a sense of what a data centre is, and of the range of data centres—from the very largest to the smallest—that can be considered a data centre?
Christina Christopoulou: Absolutely. Let me start by saying that cloud computing is embedded in our everyday life. From the minute we swipe our card to make an online transaction, a hospital pulls up a patient’s records or a university shares the results of a final exam, all that data and all those transactions are happening in a physical building called a data centre. We are talking about computing devices hosted in a data centre: hardware, software and networking equipment used to share data.
From my perspective, representing a cloud computing operator, there are different parts of the sector. We are talking about on-premises data centres, with servers in rooms, where companies own the servers and use the IT equipment in-house. We are also talking about cloud service operators that have the facilities and infrastructure and can serve their customers through the cloud—the cloud computing feature that I just mentioned.
Q4 Chair: Can you give people watching our proceedings a sense of what someone would actually see if they were to visit an Amazon data centre?
Christina Christopoulou: You would see the computing devices, as I have said, and the hardware, which includes the racks within the data centre; they are the engine that is powering the IT services within the facility. Of course, we power those devices through electricity, but we also need to meet high security and resilience standards, for which we have back-up power available to make sure we can reliably service customer demand 24/7.
Q5 Chair: Some people’s fears about data centres, in principle, are that they effectively involve computers doing people out of work. To what extent are they providers of work, and to what extent are they thieves of jobs?
Christina Christopoulou: It is important to look at the broader benefits that cloud and AI can bring. To tell you a bit about our experience, we announced an £8 billion investment in 2024 for between 2024 and 2028. We looked at the data and saw that this investment in data centre infrastructure could bring up to £14 billion in GDP contribution. That is one of the figures from our own research. There are also third-party studies from 2023 showing that cloud computing at large brought a benefit of £42 billion, representing 1.6% of the country’s GDP in added value. To put that into perspective, at that time it was larger than the UK automotive manufacturing industry. That is one area.
Q6 Chair: Mr Stewart, turning to your own provision and thinking about the same fundamental value questions, does the environment that your organisation is responsible for, in physical terms, largely mirror what you would see in the private sector, or are there any key differences?
Steen Stewart: The key difference generally appears to be price. Whenever we have weighed it up against industry offerings, we have generally come out with a 50% benefit to the public sector.
Q7 Chair: Why is that?
Steen Stewart: Scale. The Cabinet Office buys in bulk, so it gets a good price. The data centres we utilise are all highly efficient, with a guaranteed PUE of 1.15, so they are some of the best out there. We have good terms as well: we have the ability to terminate for convenience, which you generally do not get in industry offerings and which is more flexible. Obviously, the public sector requires planning seven-plus years in advance, and that is difficult, shall we say.
Q8 Chair: We would love the public to be planning seven years in advance in some areas, but that is good to hear in this regard. Just so I understand what you are talking about, you are effectively buying these services for the Government from private providers—you are not running them all yourselves.
Steen Stewart: No, we literally offer the environment only, so you get the building, the power, communications and stuff like that. For everything else, it is down to the individual entity to put their equipment in—to basically plug it in—and manage it themselves. They can employ a third-party provider to do that on their behalf if they so want. Larger-scale entities, such as the DWP, have their own engineers and they manage it in-house.
Q9 Chair: Before we move into some of the detail, we heard about the GDP growth potential offered by data centres. Mr Birtwistle, on a point of principle, does your organisation have any views about the charge towards public and private sector data centre usage?
Michael Birtwistle: I think we would say that building data centres on the scale envisioned involves large societal trade-offs, and it is important for policymakers and decision makers to have good information about those trade-offs. Taking both sides of the trade-off, we are concerned that the claims that are made about productivity, for example, and the information that policymakers have about energy usage, water usage and so on, are not accurate. It is very difficult to make good decisions in that information environment.
Q10 Chair: That is precisely what has driven this inquiry. As you are aware, the Committee is right at the start of it, but what key things would you want the inquiry to scrutinise to try to ensure that we are better informed on these matters?
Michael Birtwistle: I could probably talk at length about concerns and barriers around the use of data centres later. But in short, the claims about the benefits of data centres and AI generally—it is London Tech Week, and I was there on Monday watching the main stage—are very much driven by an intention to create a fear of missing out. One company said, “The next 18 months will determine how well the UK competes in the global economy.” Another said, “The UK mustn’t lose its head start.” It is sort of positioned as this being future productivity. We have published a report on the methodology that sits behind a lot of productivity claims, particularly in the public sector. It essentially says that there are recurrent faults in those methodologies that are potentially overestimating the benefits of these technologies.
The other thing I should mention briefly is jobs. We would like Parliament to take a view on where the benefits of the investment are actually accruing. You have one-off jobs that are created, such as in construction. You need only 50 to 100 people to run a full-scale data centre on a permanent basis. There are questionable statistics about the onward jobs that are created by data centres. On the other side, we need transparency around energy and water use, and around pollution and other effects that can occur.
Chair: That is helpful. It fits with some of what the Committee set out in our initial terms of reference.
Q11 Jonathan Davies: To move us on a little bit in terms of why this is such an important growth sector, we know that demand for data centres outstrips supply in the UK at the moment. There are probably a number of factors behind that, but it would be helpful if you could tell us what is leading that demand. Perhaps I can start with you, Mr Stewart, because you might be well placed to introduce that for us.
Steen Stewart: I can certainly comment on the public sector and what its demand looks like. The spend through our framework has been relatively even—about £64 million in the last two years—but there has been a slight change in what customers are commenting on. We have had engagement recently from a major education entity within the UK within the UK. They are looking at high-power compute and AI, and wanting water-cooling solutions. That is the route our suppliers are now looking towards.
Sovereignty is another one that I am sure has cropped up a number of times. A lot of customers are looking towards data sovereignty, and having things based in the UK—that is maybe a wider topic for another day.
Generally, the market is shifting slightly, from traditional air-cooled data centres to wanting denser racks for AI and the like. That seems to be the trend, and we are certainly seeing that in the public sector at the minute.
Q12 Jonathan Davies: Ms Christopoulou, do you have any reflections on this, perhaps from a more commercial, private sector perspective?
Christina Christopoulou: We are seeing demand growth across all our customer base—public sector, private sector, financial institutions and all sort of businesses. We did a study earlier this year where we tried to look at AI adoption across UK organisations, and we saw that between 2025 and 2026 there was an increase in AI adoption from 52% to 64%, which, interestingly, was 10% above the EU average.
There is also data available at the global level. The International Energy Agency looked into demand growth, and saw that in the next five years demand will double. It is important to contextualise that. Demand will double from 1.5% to 3%. When we are talking about demand, it is important to understand that there is general-purpose compute—the standard compute—but that we then also have AI.
We also see a demand need that might be coming out of the standard shift to cloud. If we look at today, 85% of the workloads are hosted on-premise, meaning the servers that we discussed before, which are potentially in a basement or next to a small business. When we look at that hosted capacity on-premise, we see that cloud is 4.1 times more efficient, so it uses less energy. There is also a discussion around where that demand and growth should happen.
Michael Birtwistle: In our view, there is one major technical reason driving demand, and it is the reason why we are talking about data centres today: the advent of general-purpose AI. The type of workload that is being demanded of a data centre is an order magnitude larger and more complex than maintaining cloud services, web and that kind of stuff.
If you look, for example, at DSIT’s projected figures by 2030, 6 GW of power will be asked of the UK grid by 2030. That is a significant scaling up that is due to precisely the kind of maths you are doing on these things. Accordingly, the types of data centres that are being proposed now are much larger. They are a different beast to the kind that, for example, Ireland saw a growth in over the 2010s.
There are a set of technical reasons there, but it is also helpful to talk about the political and strategic reasons why there is this growth. You have the tech sovereignty debate, which another Committee of this House opined on earlier in the week. It is essentially this idea that the UK will want its own, sovereign AI compute. The UK hardware plan released earlier in the week said that we “face increasing concentration of supply, intensifying geopolitical competition, and growing vulnerability in the critical technologies that power modern economies.” That is the Government’s view.
It is also about a bit of a strategic stake that the UK wants to have in the future AI economy. Minister Narayan said in Feb that, “We need greater British technology ownership before we can demand deeper British technology influence.” There is a bet being made about what kind of AI will be useful in future. The AI growth zones in the AI opportunities plan assume a strategy of investment in frontier AI, general-purpose systems and a future of agentic and multi-agentic systems.
There is the desire to reform and improve public services; the AI opportunities action plan envisions quite a lot of tie-in between public sector transformation and investment in compute. There is also the desire to secure economic growth; data centres are powering a large amount of the growth that the US is currently seeing and reporting, and some of that is imagined to be beneficial here.
The last reason is this question of fear of missing out: one data centre builder claimed on Monday that its £8 billion investment into UK data centres will unlock £35 billion in productivity gains. Those are marketing claims, but they are materially affecting how UK policymakers are making the trade-offs that I talked about earlier.
Q13 Jonathan Davies: On why the UK is a desirable destination for data centres, I was interested to see that on many of the limiting factors—grid connections, energy sourcing, use of waste heat, sustainability, data protection or financial incentives—the UK appears to outperform Germany and the Netherlands. I think that is correct. Starting with you, Mr Stewart, what are the limiting factors on the growth of this industry?
Steen Stewart: It is difficult for me to say from the public sector. Our growth is limited by the public sector’s willingness to move, really. There is clear evidence that it is beneficial to move into our data centres, as has been shown by many customers. Some just do not have the desire to move, and unfortunately we have no influence on that. It is down to individual entities. Our market is kind of stuck.
Q14 Jonathan Davies: As yours is a commercially driven organisation, Ms Christopoulou, what would make the UK more attractive to the people who back your organisation?
Christina Christopoulou: The figures I mentioned before—the £8 billion investment that we announced from 2024 up until 2028—are evidence of our long-term commitment to the UK’s economy and growth. If there is one thing that is top of mind for a data centre operator such as AWS when we are expanding, it is, of course, having a stable and sustainable grid connection to be able to meet our customer demand and having available land and broadband network connectivity. Those are the factors we consider, but at a general level about the conditions we seek to find, when making decisions about our investments.
Q15 Jonathan Davies: That leads me on to my final question, which I will direct to Mr Birtwistle. I read with interest that there has been a planning decision in Slough about building a new data centre. The local authority initially rejected it—it was due to be in the green belt—but that was overruled by the Secretary of State for Housing, Communities and Local Government. I am guessing the reason for that is that it is such a lucrative opportunity for the UK.
I was recently discussing data centres and the economic opportunities that they offer the UK with a colleague from elsewhere in the country, and he said, perhaps a little bit flippantly, that he would rather they built an Asda because it would create more jobs for local people. Obviously, the country benefits from AI financially—it makes our public services more effective and efficient, and it drives commercial opportunities—but how do we take people with us on that journey at the very local level?
Michael Birtwistle: The “UK AI Compass” report, which was launched last night in the Palace, is a piece of work that looks very closely at the range of public attitudes about AI in the UK. A very powerful theme that comes out in that work, and in Ada’s own public attitudes research, is a sense of disempowerment among people and a sense that AI is not necessarily being developed in their interests.
Evidently, you may well have trade-offs between the national interest and the local interest. In many cases, that is a feature of the planning system, but I think it is very instructive to look at the examples of the US and Ireland and their experiences with data centre constructions, which have been much more a feature of their political landscapes than of ours. You are currently seeing organised resistance to the construction of data centres and concerns about how both the literal environmental impact and the system costs of data centres that are being pushed on to consumers—for example, through higher energy prices—are affecting support for those data centres.
Having clear, transparent decision making based on good data about those trade-offs, and having a link between that data centre and the community—for example, through waste heat management and purchase agreements that involve the data centre investing in the local community in some way, which is a feature of some of the discussions between some data centre bids in their areas—are ways of managing those trade-offs and making people feel included in that discussion and trusting that it will actually lead to some benefit for them.
Q16 Sarah Gibson: Very quickly, and picking up from what you just said, I want to ask Mr Stewart a question. A data centre has just been given planning permission in my constituency. It will clearly be used by the Crown hosting it next to a large MOD site, but during that process we were shocked by how little engagement they felt was necessary. As a major client of these data centres, you have an ability to influence how they work. Could you not use that influence to get them to do more than the minimum requirement, which is all they have done in my constituency? Could you try to persuade them to do more waste heat management and consider using systems for cooling such as earth tubes, rather than the highly expensive and energy-inefficient systems that they are currently using? There seems to be no pressure on them from the final user, who in my constituency case will be yourselves.
Steen Stewart: Yes and no. Sorry to be quite vague. If we are talking about data centres, we are roughly 18% to 20% of the business across Government. What we represent I am not completely sure, but what is going into those data centres—I know which ones you are on about—will be a very small part. They could be completely purchased by a hyperscaler, and we will not have any footprint. Certainly there is no schedule for us to look at taking them over, so our influence over the supplier is minimal at best. Obviously, it is currently a joint venture, but we are a 25% stakeholder in that joint venture, which means minimal influence, albeit we do have some influence. Should we engage more? I could not agree more strongly with you. Why wouldn’t we? In fact, I have come from a strategic meeting with the supplier this morning, and we had that discussion on how we engage with the public and the public sector to convince them that Crown Hosting and our data centres are good for us all.
Q17 Sarah Gibson: You will not convince them by just engaging with them further unless you actually show some tangible benefits and straightforward things such as waste heat management use and various other community benefits. It takes me back to thinking about when other countries built nuclear power stations and they had district heating systems; we have never managed that. As a large user, you should be in a position to have some influence over what they are doing. It seems worrying to me that even with a 25% stake, you still do not feel as if you have that influence.
Steen Stewart: It is because we have a minority stake that we have the ability to ask them to do stuff, but the contract is written, the Cabinet Office has signed it, and that is what we operate until 2029. After that, we’ll see. Maybe there will be a framework 3 that gives us more influence on new sites that are going to be for public sector use, or not. We are stuck with the framework we have today and the influence we have today, which is evidently not enough.
Q18 Dr Savage: I will direct this question first to Ms Christopoulou and then open it up to other panellists. This follows on well from Sarah’s question. What are the principal environmental impacts of data centres, and which of them—water, carbon footprint, whatever—do you think is likely to grow the most rapidly in the years ahead?
Christina Christopoulou: I can talk from my perspective about what we do as a company to ensure that we are not putting any burden on the environment. We have a comprehensive sustainability strategy, which is guided by our north star goal to be net zero carbon by 2040. I can walk you through what carbon-free energy and energy efficiency are in terms of what we do. I will start with water.
It is important to talk about why we use water. We use it to cool our servers. In the UK, we do not rely on water for 97% of the year, so we only use water for 3% of the year, which is when we need to cool down the outside air so that our servers can be more energy-efficient. We have made a commitment to be water-positive by 2030. We are 53% along the way. Another metric showcases how efficiently an operator uses water, so-called water usage effectiveness, and across our European operations we have a 0.04 score. To put that into context, the industry has goals through the climate neutral data centre pact, and there the commitment is to meet 0.4 by 2030. Those are some of the features that we do on water.
Beyond our efficiency, we try to replenish water. We have a collaboration with the Rivers Trust whereby we will give back 800 million litres to the community to contribute to the River Kennet and to the River Thames and the Thames basin—we have a basin approach, where we look into it. Beyond that, whenever possible, we want to use industrial reuse water, which is also an area where we look forward to seeing a framework in the future in which utilities will be incentivised to provide and make such industrial water available.
On pivoting to carbon-free energy, we are Europe’s and the UK’s largest corporate buyer of carbon-free energy, and globally we are one of the leading buyers. As I said, we have a goal of net zero carbon by 2040. We have 700 projects and over 40 GW of capacity that we have contracted. In the UK, we have 40 projects, the equivalent of up to 850,000 UK households. To name just two of them, we have Moray West in Scotland, and Corlacky Hill in Northern Ireland. In those two projects, we are adding net new capacity into the UK grid, which benefits not only our data centres, but businesses, the public sector and so on.
Everything, however, comes down to the point of trying not to use much energy. The greenest electricity is what we do not use; that is one of the principles we abide by as a company. As I mentioned, we did a study that showcased that AWS data centres can be up to 4.1 times more efficient compared with on-premises data centres. A measure that showcases data centre efficiency is power usage effectiveness; our PUE globally is 1.15, and at the European level 0.11. To put that into context, the average of the industry is between 1.5 and 1.6.
The new data centre design that we announced—
Q19 Alison Griffiths: Sorry, to interrupt, but is that 1.5 to 1.6 hyperscalers and private cloud?
Christina Christopoulou: The whole industry—average industry PUE. Our new data centre design can achieve up to 1.08 power usage effectiveness, but we are not stopping there. We recognise that AI will need more power. This is why our new data centre components are designed to reduce mechanical energy by up to 46%. We also have the newest generation of AI chips, Trainium3 chips, which have four times greater energy efficiency and compute performance, to support the compute per watt produced in a data centre. We also operate in the context of a global energy management system, ISO 50001, which centralises and standardises our approach to energy efficiency across our operations. All those things are important, especially when we think about what cloud can bring in terms of efficiency.
There are two figures that I would like to share from the IEA report—one from 2026 and one from 2025. In April, the IEA found that, if scaled, AI technologies across the energy sector have the ability to reduce energy use by 13.5 exajoules. That is the annual consumption of a country like Indonesia, which has 280 million inhabitants. The second figure is from last year. The IEA estimated that AI technologies, if not scaled more than what we have available today, are integrated across the energy sector, they can make CO2 reductions up to three times larger than the total data centre emissions by 2035. When we are having this discussion, it is important to look at what data centres should be doing, as I have evidenced—
Q20 Alison Griffiths: Can I ask a quick question on that? Those are very seductive numbers but, in very simple terms, how would that happen? What is AI going to deliver to create that?
Christina Christopoulou: For instance, it can deliver power optimisation. If grid operators use AI technologies, they can reduce faults in their system. They can be more energy-efficient and can create flexibility in the system by way of creating specific tariffs. They can shift demand so that they are more efficient. A lot of technologies could be scaled across the energy sector—the energy sector has different ranges. This is not our data; it is data from the International Energy Agency.
Q21 Dr Savage: It would be very interesting to see the assumptions that underpin those figures. If those are available, could you send them through afterwards?
Christina Christopoulou: Yes. We are happy to share that through written evidence.
Q22 Dr Savage: Do Mr Stewart and Mr Birtwistle have anything that they would like to add on that?
Michael Birtwistle: There are three major areas of direct effect: power, water and direct pollution. I will address some of the claims being made. The scale of the pressure on the grid by the growth that is envisioned—a fourfold increase by 2030, which is only five years away—is such that, even if a single data centre developer has a net zero strategy, you push elastic demand that doesn’t care about whether it is green or not into fossil fuels and other non-renewables, so the net effect on the UK’s net zero targets is still damaging.
A second thing that is helpful to think about is exactly when demand happens. Data centres are not like other forms of energy grid consumer; they are much more likely to be constantly on. The workloads are scheduled across them, so if you are doing inference—running queries on AI—you need to be constantly available with low latency. If you are a training model, maybe you will do it at night when the energy is cheaper, but that still means that the data centre is constantly pulling on the grid.
On figures like the claim that water is used only 3% of the year, it matters when that happens. If that is happening at a time when that area is already experiencing water stress because of very hight temperatures—
Dr Savage: Which it presumably would, predominantly.
Michael Birtwistle: That is exactly when everything might be getting hot, so you will be increasing water stress at a time of critical water stress for that area.
Another thing that is worth disambiguating are some of these figures like water positivity and water neutrality. Those often do not measure exactly what you think they are measuring. When you hear claims, in particular in the US, about water neutrality, you have reports of communities saying, “This water is being taken from one place and being put back in another, and is still being counted as water neutral,” so it is not necessarily an accurate proxy of water stress avoided.
The other thing worth talking about is behaviour that happens behind the meter, as it is called. This is true across the industry; I am not making any claims about any particular developer. In order to serve the energy requirements and address the delays to getting on the grid, many developers are countenancing private energy generation, often using natural gas, which obviously has knock-on effects for the UK’s net zero targets.
There are all those concerns, and then there is a separate concern: will the protections that are supposed to kick in and take account of all those impacts actually kick in and bite? There are a number of reasons from, for example, the Buckinghamshire data centre—the Committee might be familiar with that example—showing that the policy levers that the Government are pulling to make this stuff happen quickly means that some of those impacts are being skipped or the information on which policy makers are basing decisions is not complete.
Dr Savage: Very interesting; thank you. Mr Stewart?
Steen Stewart: I do not have a lot to add, to be honest. To back up what Christina said, the fact is that data centres are built to be as efficient as possible. That is where they make their money. It is an industry—they have bills to pay. We just piggyback off it, really. I can only talk for our data centres—very similar to AWS. We do not use water very often, but you are right that, when we need it, inevitably it is at those peak times when there is not a lot. It is much the same as, when you want to water your garden, it is guaranteed that it will be when it is hot. It is inevitable, but they are investing. I can guarantee that Ark, certainly, are investing in rainwater harvesting to maximise the amount of water they can possibly store, so that when those times are hard, they do not drain from the grid and affect residents. They are doing their best. Is it enough? Who knows what weather there is going to be in the next fortnight.
Q23 Sarah Gibson: They are doing something; are they doing their best?
Steen Stewart: I would say that they are doing their best. If they can reduce the amount of water they take from the grid, it increases profits—every benefit they can get to put more pounds in their pocket. It sounds quite brutal, but they are a business. That is the fact of the matter. They need to pay their bills as well. The more effective and efficient they are, the better it is for them, and the better it is for us as well.
Q24 Alison Griffiths: When you are negotiating with private data centres, is there any consideration about which workloads require both the data sovereignty and the latency that would require you to be in the UK, versus workloads that could be in environmentally much better locations? I have just come back from Canada, where they use hydropower for pretty much most of their power. Much of the country is very cold. I am just thinking about how we place workloads more efficiently around the globe.
Steen Stewart: From a Crown Hosting perspective, we have no movement on it at all. We signed up to a single supplier for the next few years. They were UK-based only; they do have data centres now in Europe, but we have no customers with any interest in putting their data into Europe. They are quite happy with it being in the UK. Could we? Yes. Would we have any demand? Possibly not. Who knows? If the price is right, you never know. If the data owner has no issues with putting it in an overseas territory and taking the risk that that incurs, that is always possible.
Q25 Dr Savage: I assume that the water that is used for cooling remains clean water. It is not like fracking, where the water comes out polluted with lots of nasty chemicals. It is clean when it goes in and clean when it comes out. Is that right?
Christina Christopoulou: I can confirm that we do not use any industrial chemicals. We take the water as it is being supplied by the water supplier and then we use it for cooling purposes and discharge it afterwards.
Q26 Dr Savage: Is there any possibility of using something else to cool the servers? Is it possible to have a waterless system?
Christina Christopoulou: There are dry-cooling or waterless cooling technologies. When we are deploying a data centre, there is a suite of technology choices. We look at what is best from a water and energy nexus, so looking at both together and what we can achieve. As I said, in the UK, for 97% of the year, we do not rely on water. I also want to highlight the fact that we have a water-positive goal, so we are giving more water back to the communities. Our programme on replenishment is based at watershed level—the watersheds where we operate our data centres.
Michael Birtwistle: Something that it is important to understand at technology level is that there is a direct trade-off between water usage and energy usage. If you are not using water, you are using something else to cool your systems. You are probably using air cooling. There are some experimental technologies whereby you immerse the whole server in a non-conductive liquid, but I think those have not been developed at scale yet. So if someone is saying, “We are using less water,” that probably means they are using more energy.
The other thing I want to point to is the connection between efficiency and net usage. The more power efficient it becomes to make an AI query, the cheaper it becomes. There is a theory called the Jevons paradox, which the Committee will be familiar with and which is that as something becomes more efficient, net demand rises. So it is a good thing for data centres to be efficient, but we should not expect greater efficiency to mean less demand on the system in terms of energy and water.
Q27 Chair: What about wind cooling, Mr Birtwistle? I have been made aware of some data centres that are positioned in particularly windy areas and therefore have less demand for water. Is this something that you are conscious of?
Michael Birtwistle: I think there was an article this morning about a Chinese data centre that was perhaps using wind cooling. I am not a technical expert on that technology. From what I understand, it is not in mass-scale deployment yet. As Christina put it, there is a suite of technologies that can be chosen from at the time when a data centre is developed. The question we would raise is: what are the incentives to choose those different technologies and are there sufficiently strong incentives to choose energy-efficient and water-efficient technologies?
Chair: Obviously, that is something that our report may recommend and Government could at some point stipulate.
Q28 Sammy Wilson: Before I come to my main question, can I just ask this, out of pure personal ignorance on the matter? You have outlined the benefits of data centres in terms of productivity. I think the figure of £8 billion invested could give £35 billion-worth of productivity gains and so on. Given the controversy about whether they create a lot of jobs and whether they have impacts on local communities, how important is it for getting those benefits to have the data centres located here in the United Kingdom? Can they be geographically mobile, and can the benefits from them still be experienced even if they are located somewhere else?
Michael Birtwistle: To be clear, I was referencing those productivity claims but referring to them as claims with questionable methodologies behind them. The link between a data centre being built in a location and the benefit of that data centre being felt in that local area or even nationally is not automatic and needs to be designed for. You can design for that with commercial incentives and requirements using procurement levers, to create a relationship between a data centre and its area or a data centre and the kind of work that happens on that or the efficiency of the work that needs to happen on that, if you are concerned about environmental outcomes. But there is not a clear theory of change that says that you plonk a data centre down in some bit of the UK and that necessarily generates value for UK businesses.
One sort of shorthand for that, for example, is that when you see a large announcement of a data centre investment, probably about half of that is going straight out of the door on chips, and unless those chips are being bought domestically, that benefit is not recirculating in the UK economy.
There were some messages in the AI hardware strategy earlier this week that suggested that the Government may be beginning to prioritise the domestic value chain, which would be very positive if realised. However, there is a tension between the geopolitical competition for having these data centres built in a particular location and the ability to insist on terms with the developers.
Q29 Sammy Wilson: On the environmental sustainability of the centres, how are environmental sustainability practices embedded in the planning and operation of data centres? Can you give examples of where and how it has been done?
Steen Stewart: Our data centres are designed and built in the UK. Some 95% of the materials are sourced from the UK to save on transportation costs and environmental impact. For example, the concrete pour is reduced to a minimum because of the environmental impact that has. Steel is sourced from Wales, and as much as can be, everything forming the data centre’s carcase is sourced from the UK to reduce our environmental impact. Inevitably these are large buildings. I do not know if you have had the pleasure of visiting one, but I would encourage you to. They are impressive, so please have a look. The structures are vast and obviously there is an impact, as there is to building houses or anything else, but there are mechanisms to reduce that to the minimum necessary.
Christina Christopoulou: I am happy to build on the points that I made before. I spoke before about water, energy and carbon-free energy, which are three of the areas that are always top of mind whenever we are planning, building and designing a data centre.
I can also talk to a few other areas around operational sustainability. Whenever we are planning to develop a data centre, we try to be a good citizen. We engage proactively with the community to understand their needs and ensure that there is ongoing feedback about how we can be a good citizen and neighbour in that community.
To give you a couple of examples on noise, there are design features such as acoustic enclosures and low-noise equipment. We are not managing those issues afterwards—we have it in the DNA of our design. There are also planning boundaries, periodic monitoring and ongoing compliance checks.
Biodiversity is the next area. We follow a mitigation hierarchy when it comes to biodiversity: avoid, minimise and restore biodiversity impacts. That applies across all our UK facilities. We are also compliant with the Environment Act 2021 to achieve a 10% biodiversity net gain. Where feasible and if there is a viable grid connection—because we need to meet customer demand—we also build data centres on brownfield sites, and we are trying to make sure that we are adding value to that site and area.
Circularity is another area, broadly speaking, that we are focusing on. We have the design principles: reuse, repair and recycle. We also design for as long a use as possible, along with repairing and then recycling. We have a European facility to test and repair equipment or then recycle it in Dublin. Recently we diverted 23.5 million components from landfill. Those were either sold in the secondary market or recycled.
There are a series of actions that we try to undertake, but our responsibility does not stop there. We are also trying to make sure that we are building a next generation of data centre professionals that can find a job in our community. Just to give you one example, last year we created the UK’s first data centre apprenticeship, where the next generation of people can learn what it means to work in data centre operations. We also committed to training 100,000 people in AI skills by 2030.
Michael Birtwistle: It might be useful to look at the example of Buckinghamshire. The Government have their AI growth zone initiative. Part of that initiative, as established in the AI opportunities action plan is to help fast-track data centre proposals through things such as environmental impact assessments and habitat planning. In that example, the Government called a decision in and admitted, I think, a serious logical misjudgment in the assumption that no environmental impact assessment was required. In practice that means that decision makers are not necessarily asking for, being given or have access to the detailed environmental information that they will need in order to make good decisions about those data centres.
There is that individual planning-level decision but there is also the system-level question. It was reported about a month back, I think in The Guardian, that there were serious disparities between DSIT’s estimations of future energy demand and DESNZ’s. What good planning would look like at the system level is all bits of Government having a single, accurate picture of projected demand on the system based on real-world proposals that are going through, not simply, say, the local authority doing that or the Minister looking at a single, called-in planning decision.
Q30 Sammy Wilson: You outlined, as far as the building and operation of the data centres, how you try to build in sustainability. Can I just take you to the point before you reach that decision? This is provoked by a visit I had during the recess. Obviously, Governments and authorities are keen to locate these data centres for all the reasons that you have given, and sometimes in doing that, maybe the sustainability or the environmental impacts are not totally thought through.
I am thinking of a current planning proposal for a wind farm in the Sperrins in the west of Northern Ireland. It is beautiful, untouched upland—peat bog—and an area of natural beauty. The wind farm will be designed to drive the power for data centres proposed in the north-west of Northern Ireland, and there is a big economic demand for them. It will involve a lot of disruption to transfer the power once it has been generated to where the data centres would be—a quarter of a million metric tonnes of peat will be removed from pristine bog.
When it comes to making an application, the authorities may well say, “We’ll provide the electricity.” Do the data centre providers, in their desire for sustainability, ever ask the questions behind that? Do you ask not just that you can provide the power for it, but how that power is going to be provided? Do you ask, “What are the environmental and social implications of that going to be in terms of the power that we will require in that particular locality?” You may argue that that is the responsibility of the authorities, which are going to give the planning permissions and so on. But do you ever ask those questions? Do you ask, “Even if the authorities are happy for us to come here, what are the consequences of our locating it here and making demands for either power or water?”
Christina Christopoulou: As I said before, before we invest in a new location, we always seek to be a good citizen and neighbour. We engage with the communities to understand their needs and demands. We have a very comprehensive sustainability strategy, and we are trying not to add any burden from an environmental perspective. But ultimately, we are also meeting customer demand, as evidenced by the figures that I shared with the Committee. That is what I can say to that question.
Q31 Sammy Wilson: But if the power is being supplied and the authorities say, “We can generate the power,” you do not really ask whether it is renewable or what the implications of providing that renewable energy may have been. Are you saying that that is not a decision for you—it is simply, “Are we getting renewable energy?”
Christina Christopoulou: We are grid connected but, as I have said before, we have invested in the UK with over 40 projects, which is the equivalent of 964 MW. We are adding net new additional carbon-free energy to the UK grid through wind and solar, which is the contribution that we are making.
Q32 Sammy Wilson: Would you expect that it is reasonable to dig into the implications of locating in a particular area and look at those environmental issues and the impact on local communities, who will probably not benefit from the data centre, but are nevertheless impacted?
Steen Stewart: In that case, I would expect that the data centre would be interested in knowing where that power is coming from. If it is a wind farm, it is obviously not a guarantee. It would not be ideal for them anyway, or I would be surprised if they were pushing for that solution. It sounds—this is the world according to Steen, not anyone else—like a really bad idea, but I obviously do not know the whole story. For a data centre, wind is not a guarantee, so I am sure that they would rather have another renewable source, and they would take that into account, I would hope.
Q33 Sammy Wilson: But data centres tell us that 50% of their energy at present is from renewable sources, which is not guaranteed, simply because the sun does not always shine and the wind does not always blow. This raises one of the issues of sustainability. What back-up systems are used? For example, would data centres rely on diesel generators in the case of an outage or whatever? How is that regarded as a sustainable provision of power?
Steen Stewart: I can talk about this one. Our data centres operate two power streams with two back-up generators. They are not diesel; they run on HVO—hydrotreated vegetable oil—which runs far cleaner and is far better for the environment. They are required only if our multiple feeds for mains power drop off. Could it happen? Yes, absolutely. We need to run those generators up to make sure that they are operating as expected and are highly resilient to make sure that your laptop stays operating and gathering the data that you require. There is an impact, but it is minimised as much as possible. Going back to that mains power, I am not a data centre operator, but I would say that that decision would be questionable.
Q34 Sammy Wilson: But given the increasing reliance on renewable power that data centres boast about, would it be common to have those kinds of back-up methods? How sustainable is that?
Christina Christopoulou: I can build on that. We have a plan to transition from fossil diesel to HVO. Hydrotreated vegetable oil is a renewable fuel from waste cooking oils and vegetable oils, and it can reduce up to 90% of greenhouse gas emissions over its life cycle versus fossil diesel. We started to do this in 2023; Ireland and Sweden started, and we have also started rolling out HVO at certain UK sites. Of course, there is a broader element that is important, which is the sustainable procurement of HVO. We are making sure that raw materials are traceable to their origin and that they are not derived from high biodiversity sources. Making sure that there is a sustainable supply chain is something that we are trying to support through our engagement with suppliers.
Michael Birtwistle: Fundamentally, at the bottom of the question was something about incentives. There is not an effective commercial or legal incentive for a company to take those things into account when it is being reassured that power will be there outside the security of that power supply and its reliability, other than passing its environmental impact assessment, which may be a process that is being completed through a fast-track process, to put it generously.
The wider question is about the lack of alignment on the incentives, or the ownership for outcomes that is held across other bits of the public sector. Across the local authority, the grid and the national planning in DESNZ, DSIT and the Treasury for the productivity gains, there is not really a vehicle for alignment to understand where benefit will be generated and whether it is the right trade-off to make.
One strange place where that lack of alignment in incentives planning manifests is the current disconnection between the priority in the grid queue and who can apply for that, and land ownership and planning permission. You have situations where some developers are sitting on land that is good for a data centre and a different developer is ahead of them in the queue for a grid connection for that land that they do not own. Again, that shows the misalignment of incentives. It is difficult to plan for good societal and local-level outcomes with those disconnections happening.
Q35 Sammy Wilson: Finally, in an attempt to reduce energy consumption in the longer run and be more sustainable, what is the potential for data storage clean-ups? A lot of data may be held that is no longer of any use. Is there big potential to reduce the power consumption of data centres, and is there any incentive to do so?
Steen Stewart: There certainly is for the public sector. Whether or not, in the entire trend, that will show a reduction over time, I think it will always increase, because technology is key to modern life, unfortunately. That is inevitable; it is not going to slow down in the near future.
You are right that we have seen public entities moving to our facilities. They mature their data, chop a lot of it because there is always duplication—I am sure we have all seen it—make it right, and then move it into the cloud. That is showing a significant reduction in the data storage. Unfortunately, not everyone has done it. It could happen with a push, and it would then reduce the data storage requirements and therefore the footprint in data centres, certainly in the public sector.
Christina Christopoulou: Just to clarify the question, did you ask about data storage or onsite storage?
Sammy Wilson: Data storage.
Christina Christopoulou: I am sorry; this is outside my area of expertise.
Michael Birtwistle: I would defer to Steen on this, but I would assume that, while it is right that people should push for efficiency there, the compute and power costs of data storage are likely to be dwarfed by the compute costs of running AI training and inference. It is an order of magnitude of difference in workload.
Q36 Alison Griffiths: Looking at the growth of data centres, there are two things I want to cover. First, why do estimates for the number of jobs created by data centres range as much as they do? Can each of you provide clarity on the figures that you are able to provide?
Secondly, I was interested in the differences between public sector and commercial growth pressures. What will be the different constraints in those two sectors?
Steen Stewart: On the employment issue, I struggle to comment because I do not run a data centre. I can only comment on what I have seen. I visit our two or three sites and there are a lot of people working there, such as cleaning and security staff. There is an environment there that requires people and jobs—some very skilled jobs. There is employment. Whether that competes with something like an Asda, as I think someone said earlier, I am not sure, but it certainly is in an employer in that community and there are highly skilled jobs that the UK requires.
What demand looks like is difficult to say. The public sector has been constant for us, but—again, this is only for our framework—it is all dependent on the Government’s drive at the time. When Crown Hosting was brought into being, there was a big push with the “Cloud First” policy and all that good stuff—we were named in that policy. There was a drive, a big take-up and the trend towards us climbed massively.
I would not say that has died off—it has just plateaued a little; yet, when we look at the data that the likes of DEFRA gather, it is questionable whether it is right. That is not DEFRA’s fault; it is given information by other entities. A lot of public sector entities have those broom cupboard data centres, which are highly inefficient. They burn power like there is no requirement to. If they were to move into a highly efficient data centre, such as those offered by multiple providers, there would be a vast saving on efficiency, power costs, carbon, water usage and all that good stuff. We are driven by, I guess, yourselves. If you push it, entities come to us and we can make things better, hopefully.
Q37 Alison Griffiths: One thing I have heard from all three of you is the massive, exponential increase in compute power that is needed as a result of AI versus the way that we were using compute previously. Is that built into your expectations? I guess that your contract with Ark was signed at a time when AI was nascent and that you now have to adapt massively to that.
Steen Stewart: The advantage we have is that our supplier wants to and has to adapt to stay current, so it is adapting to support those types of technologies and we can house them through our framework. Is it perfect? No, but hopefully when we rewrite it again it will be even better. We are kind of stuck with what we have because, as I am sure you appreciate, the framework is regenerated every so many years. It does not really move; it is set. We work with it for a number of years; it is set—
Alison Griffiths: I am conscious of time—I suspect the Chair is chivvying me a little.
Steen Stewart: We are kind of stuck. We are limited by our framework. I am sure the private sector is different.
Q38 Alison Griffiths: Christina, do you want to build on the point about the difference between the private and the public sectors then talk about the jobs point? That would be helpful.
Christina Christopoulou: I can talk from an AWS perspective; I’m afraid I cannot comment on public sector data centres. We serve a wide range of customers in the public sector and the commercial sector. We see demand growth, as evidenced, but also an opportunity as a lot of today’s cloud computing is still on-premise. We service customers such as the NHS, the Department for Work and Pensions, Octopus, Sainsbury’s and AstraZeneca, just to give you a few names.
You asked about jobs; I think we need to look at three layers. One is construction, with jobs in the thousands. We have data centre operations, with hundreds of very highly skilled, permanent, safe jobs. The more important part is the study that I mentioned before, which we conducted. It talked about the £14 billion GDP contribution and also stated that that investment can add up to 14,000 permanent roles into the UK economy. We are supporting that through the measures I mentioned before, such as the data centre apprenticeship programme, which trains young professionals to make sure that they are equipped for the new and modern work market and to make sure that we have the skills that are needed.
Alison Griffiths: If you have not already, it would be helpful if you could share that report.
Christina Christopoulou: We will provide written evidence.
Q39 Alison Griffiths: There is one thing that I, and I suspect my colleagues, would love to know. A local community that has a data centre nearby potentially runs the risk of there not being a local job uplift. Do you have anything to add on that? Are there any specific roles within the data centre itself that you think we should highlight?
Christina Christopoulou: I mentioned the data centre operational jobs—engineers, for instance, and security staff—but if you need more information, I am happy to provide it through written evidence.
Michael Birtwistle: I would point to two places. On the question of jobs and growth, as I said before, there are a bunch of construction jobs, which are significant but temporary. Your average data centre needs 50 to 100 people to run it on a permanent basis. They are often quite highly skilled jobs, so may or may not be sourced locally. There are a lot of imputed or inferred jobs, which are created further down the value chain because the data centre exists. I encourage the Committee to test the assumptions that sit under those. They often use fairly simplistic calculations, such as multipliers that essentially say, “We are going to guess that this is going to create six times as many jobs as the data centre runs”, and that is the number that is probably worth pressing on.
Secondly, you were asking about AI in public services and what is driving demand, use and growth there. It is worth distinguishing between narrow AI and general-purpose AI. Narrow AI is something that looks at cancer scans; it is very specialised and it uses an order of magnitude less compute than speaking to a chatbot. That is often the type of benefit that you see talked about, but it is not what is driving the big system demand here. The strategy to date has been to give everyone across central Government a chatbot and see whether they become more productive. Then there is some organic adoption at the coalface. Ambient voice translation is very popular with doctors at the moment and is seeing wide uptake across the NHS. That again is narrow AI—it is a small model.
Then you see a frontier of system transformation. The NHS, for example, is very actively looking at whether it can build in agents and multi-agent systems into the way it works and whether that can improve how the system works. That involves polling a large language model to say, “We have had this text in. What does it mean? What’s the response?” That creates that kind of system load. As the frontline of public services and the stuff that happens behind the scenes starts to adopt agentic, in particular, you can expect to see the demand and dependence go up.
Chair: Mr Birtwistle, Ms Christopoulou and Mr Stewart, thank you very much for your evidence. We will bring this first panel to a close and switch over to the second panel.