#106: Will energy decide the AI race?

– Intro music of The Green Room begins followed by an introduction from our host. –

Oli Carpenter (Host)

We use AI in a digital world, but success will increasingly be decided in the physical one. Behind every prompt, agent and model, sits a physical system that has to be built, funded and powered. And as demand increases, the need becomes more urgent.

So, as countries race to adopt AI, a new question is emerging. Today we're asking: Will energy decide the AI race?

– The podcast transitions into showcasing highlight clips from the upcoming episode. –

Caroline Brown (Guest)

Everybody wants all of the benefits of AI. Nobody wants to look at it at the end of their garden.

Tom Cope (Guest)

AI and the growth and need for data centres, I think is happening at an interesting time. AI has almost come at the most imperfect time.

Caroline Brown (Guest)

You want a really quick answer. You don't want to be waiting while it sends the query to Iceland, and then comes all the way back.

Tom Cope (Guest)

We've come such a long way with renewable energy. It's almost a point of pride that they've got that within their community and they're supporting it.

– The podcast music ends and it transitions into the main episode, starting with our hosts introduction. –

Oli Carpenter (Host)

Hello and welcome to The Green Room by Deloitte. I'm Oli Carpenter, and I'm joined today by my co-host Hannah Gowen.

Over the past six episodes, we've explored the human and digital side of AI, from how it will shape the future to how we upskill a nation. And in our last episode, we explored trust in AI.

But beneath each of those conversations sits a less visible question. While AI may look digital, it's ultimately powered by physical infrastructure. The future of AI depends on compute. That depends on infrastructure, and that depends on energy.

So, rather than the best models or strongest ambitions, are countries and organisations focusing enough on the physical AI needs? Today we're asking the big question: Will energy decide the AI race?

Hannah Gowen (Host)

And to help us with that question we're joined by, Caroline Brown, who is a partner at Deloitte, working in AI for Infrastructure and Sustainability. And Tom Cope, who is also a partner at Deloitte working in the Infrastructure and Capital Projects team. Welcome both of you.

So to start, it's important to reframe our understanding of AI. As we mentioned, we often talk about the technology and the people that are using it. But AI is also just as reliant on physical elements as well.

So Caroline, let's start with you and go back to basics. It's a good place to start. What do we mean by AI compute? And ultimately what is it that AI relies on to function?

Caroline Brown (Guest)

If I used a laptop, as an example, right? Think of the AI as being equivalent to the applications Microsoft Word or Excel that you run on your laptop. You wouldn't think about using Word or Excel without thinking about the laptop that Word or Excel runs on. And AI is exactly the same.

So sitting under AI, you have a huge number of servers. That's the laptop analogy. You have to plug your laptop into the mains and the servers are the same. They require electricity to power them.

On top of that, they need a space that they sit within, and that has to be a highly secure space. It has to be very clean space. And if you've ever used your laptop during a warm weather day, last week was a great example, right? They kick out heat. And that heat has to be gotten rid of because if they get too hot, it's damaging to the hardware.

And so, what we have is a physical space we call a data centre that has massive amounts of cooling systems to take that heat away, that the servers generate. And then surrounding all of that, you have the other necessary infrastructure.

So all of the network connectivity, how does a person's mobile phone or a laptop or an iPad connect to that compute capability, to use the AI running in that environment? That's, another big volume of stuff that makes up that data centre structure.

So, if AI is the transformational capability, the data centres and all of the gubbins associated there, think of those as the factories that help the AI do the doing.

Hannah Gowen (Host)

And are those becoming more of a strategic resource now, rather than just a technical output? Is it becoming more strategic to have those sorts of bits of infrastructure in play?

Caroline Brown (Guest)

So some of those are common to what we call compute environments. So when we talk about stuff running in the cloud, it's not in the actual clouds, clearly. It's running in data centres. And so a lot of that infrastructure is common no matter the workload.

What we have with AI, are two new kinds of compute capabilities that are needed. One that does all of the model training. So when we think about things like ChatGPT, the thing that makes ChatGPT work really, really well is all the training that's done. And that has to be done over very, very specialised servers. So very specialised laptops; to continue to use the analogy.

We then have the kinds of AI that consumers use in every day, but also that drive autonomous vehicles, that help with robots running in factories that we call inference models. So these are the kinds of queries that we might do over PairD or over Copilot. We want a fairly quick response. We're not training a massive model, we're using the model.

They require a specialist environment. Less specialist than the model training, which has to be like the most leading edge. And then you have all of the other workloads that we think about when we consider stuff running in the cloud.

Oli Carpenter (Host)

You don't necessarily think about all of the infrastructure that goes behind, you know, what you're typing in on your laptop or some of the inputs there. On that thread then Tom, for business leaders, should they now be thinking about compute in the same way they might think about supply chains or capital, for example?

Tom Cope (Guest)

Yeah, I think that's a really fair starting analogy, actually. And I think we're just at the start of that journey. Everyone targeting, you know, AI usage in their commercial strategies in the future, or their operational strategies. They're going to have to think about their requirements in the context of where are they, you know, where are they putting that compute? What type of laptop, to try and keep up with Caroline's analogy, you know, do they want? And you know, there will be a capacity trade off in that as well. So...

Higher spec, possibly higher cost data centres may have the capacity, but actually the proportional cost of that versus what your use case might be to run something a bit more simple, may not be appropriate. But then, you know, you’re then competing in a much more crowded, you know, cheaper end of the compute market.

So, I think we're going to see an evolution, I think, over the next three years around just how those capabilities are priced as we reach more of that capacity and it's a lot more constrained.

Hannah Gowen (Host)

And keeping things at that kind of broad level. Tom, at a national level, we've spoken lots about the country's AI ambition, what do you think could be the biggest constraint or risk to that?

Tom Cope (Guest)

So I think, AI and the growth and need for data centres, I think is happening at an interesting time. When actually, there's a whole host of needs and calls on our electrical system and our network. So, you know, AI has almost come at the, you know, most imperfect time because commodities are becoming a lot more constrained, nationalisation is becoming much more of a trend across global economies. And everyone's looking at their system and, you know, trying to think about how do they drive more security in that.

It’s driving a huge constraint on grid connections. And also, to your point on supply chain, getting materials to develop the grids that we need, just to drive our economy is, you know, currently a big challenge that we're facing.

So I think when you overlay the huge rush to AI and therefore compute and therefore data centres onto that, it creates a very interesting dynamic and I think one of the key constraints that we have in a UK context is actually power price.

It creates an interesting question actually. If you're in the south east of England actually building a data centre, in that geographical location versus building it over in the continent where there might be a price differential is a real consideration, I think. You then get into sovereign data requirements and, you know, how important is it to have the data on UK shore? But it's a very complex picture and kind of decision cascade that you have to go through.

Hannah Gowen (Host)

And you've described it as complex - so perhaps this question is a nightmare - but how do we overcome some of these physical constraints, whether it's the cost, for example, the wait times for grid connections, how do we overcome some of that? And potentially at the speed that it's needed in order to keep pace with change?

Tom Cope (Guest)

That's a very complex question. And if I had the answer in its entirety, I wouldn’t be here, I’d be on a desert island I can assure you.

But I think in the UK we do have some real national assets. You know, we've had... We are an island and we have enjoyed a rich history of some very impressive, complex, pieces of energy infrastructure which come with grid connections that have been decommissioned. So there's some really interesting projects. If we look at places like Cottam, the AI Growth Zones where, the government are looking at areas where there is huge AI potential in trying to unlock and drive that forward. So I think you have to start there as it's almost, speed to megawatts, gigawatts, depending on your ambition, where there's existing infrastructure, that unlocks a lot.

I think the broader debate around how we accelerate our grid system, there's been huge leaps forward from NESO and National Grid over the last few years, around reprioritisation of the grid networks. Historically, when we talked about energy transition renaissance, 4 or 5 years ago, there was a huge rush to grid connections and everybody was just putting applications in. The consequences of that is it did create a bit of a bottleneck.

And as the AI boom has come along and the datacentre boom was coming along, they were at the back of that queue. And it's not just datacentres that have been reprioritizing that, but you know, the whole grid connection system has been looked at and we're on the right trajectory.

There's a huge capital programme behind that now in the UK. And I think we're starting to see, you know, we're starting to see the benefits of that. We're also building supply chain. So we've got cable manufacturers coming to the UK that are, you know, the start of unlocking what is a big problem.

Oli Carpenter (Host)

And Caroline, the importance of all of this from both a business and a geographical country perspective is huge. And this isn't just about data centres and green energy, but it's about turning that AI demand into economic growth, right?

Caroline Brown (Guest)

Absolutely. Yeah. And I think, I think the world is acknowledging that AI is here and it's right now, it's no longer a future state. This is something that businesses and consumers are actively engaging with and seeing value from.

I think for me as a sustainability person, the thing that's most important when we look at our use of AI over the next 15, 20, 30 years, is to ensure that given its high energy needs, that AI doesn't compound the problems that we're experiencing as it relates to availability of power in the UK energy system. That it doesn't create tensions over how land gets used, that it doesn't create tension over how water gets used, and remembering that we need water for cooling. But the opportunities that AI presents, actually in solving some of those challenges to my view, completely outweighs some of the constraints and the concerns.

There was a study, it's about a year old now, I think, that estimated that from a sustainability perspective, using AI to optimise our energy systems, to help with accelerating the development of new low-carbon materials, particularly some of those really challenging materials like steel and cement that we know contribute significantly to greenhouse gas emissions. When we think about how we help electrification across big industrial processes that have traditionally used fossil fuels, we know because it's already doing it, that AI is going to deliver significant gains and benefits, and allow us to accelerate.

So the study that I referenced reckoned that use of AI could contribute over 5 billion tonnes of carbon emission reductions over the next 15 years, in contrast to the 1 billion tonnes of carbon emissions it generates through use of AI. So for me, in essence, emitting one to get five back, feels like a pretty good trade.

And what we're learning as we use AI more and more, is that there are certain roles that AI does actually way better than people. There is a limit to the human brain’s capacity to sift through information and make intelligent decisions, and actually, how electricity moves around the grid in the UK is one of those where NESO, as Tom mentioned, so the National Energy System Operator, are already using AI to make those split-second decisions to balance the movement of electricity between homes and EVs and businesses and data centres, far, far quicker - in a way that actually makes our use of electricity even more efficient.

One of the constant headlines in newspapers is around what we call renewable energy curtailment, right? This is where you'll see as you're driving past a wind farm, right? I make routine trips up the M1, Northamptonshire has got a wind farm, and probably three times out of five, the turbines will be stationary. Even though it's a windy day. And that's curtailment. So the turbines are in essence, being stopped rotating because our electricity grid can't actually take any more electricity at that moment in time. If we get really, really efficient, like ruthlessly efficient at how we balance both the forecasting and then the supply and demand, we need to turn those turbines off less. That feels like a really good thing.

If we're using more renewable energy, we're using less gas. Just that advantage that AI brings for that one example, I think, really just epitomises the advances that we can expect.

Hannah Gowen (Host)

That's really interesting. I think we've touched on there some of the, you know, the environmental impacts AI has, but a lot of those individual elements, whether it's water usage or the building of data centres, for example, a lot of that happens at a local level and can cause disruption at a local level.

So, how do you think we find the balance between having that disruption at a local level, but also working towards that sort of national picture that we want to create?

Caroline Brown (Guest)

It's a really, really tough one. And we talked, Tom and I, a little bit about this taxonomy of different types of AI, that drives different types of data centres. That also drives a... where's the best place to put a data centre.

And here's the interesting thing, the most advanced kind of AI that runs on the really leading edge, newly developed, compute power. You can put those anywhere, right? So model training, it happens in a data centre somewhere. You're not waiting on an immediate answer because the model is training the model. So it's going through cycles and cycles and cycles of training.

So those are the data centres that the really big tech companies, the hyperscalers like Microsoft and Google and AWS, they can put them anywhere, right? So they choose Iceland where there's loads of geothermal energy and where the climate is very much cooler, so you don't need to apply as much cooling or in the Nordics for the same reason.

But for what we call the inference model, so the stuff where you ask ChatGPT, a question of, “I've got this rash, what might it be?” One of the most common ChatGPT use cases is enquiries about your own health, did you know? Frightening! You want a really quick answer. You don't want to be waiting while it sends the query to Iceland, and then comes all the way back.

So those data centres, they can be very much smaller because we don't need to do the big model training, but you need them to be closer. So generally they get put near cities. Now if I take that ChatGPT example and make it a business use case, help me understand how I can optimise my manufacturing line. You want an answer quickly. So what that drives is data centre build-out in quite well populated areas.

Now let me ask you, would you want somebody building a data centre at the end of your garden? I think all of us... if they look like distribution centres, big, massive warehouse buildings, nobody wants that in their line of view. And so, everybody wants all of the benefits of AI. Nobody wants to look at it at the end of their garden. So that's definitely a tension.

I would also offer that there's a trade-off between how land gets used. So would we want to surrender, highly productive, fertile agricultural land for putting a data centre on it? Probably not. So there's these different considerations that do naturally cause a tension.

I think there's definitely an opportunity for education, helping people to understand... it's a little bit like the where does my food come from? And the answer is not Tesco or Sainsbury's or other supermarkets that exist, but making the connection between when you use AI on your mobile phone to build a quirky image based on a couple of photos of your mates, understand that that drags a need for one of these data centre factories behind the scenes, that perhaps will help people overcome some of that resistance.

But there is definitely tension there, and what we're seeing in the US is a great example. There's been so much pushback on data centre development that it's really slowing down projects, simply from consumers that don't want, they don't want the land being used in that way, they don't want the impacts on water and the very real and justifiable concern that it might consume electricity that would otherwise be used to power businesses or homes.

Tom Cope (Guest)

It's a really good observation Caroline, around the you know, the starkness between the benefits that they bring and I suppose the current perception, and that's part of the challenge. This is happening so quickly. And I think as humans, we have a bit of reaction to things that we see appear and, and are so fundamentally changing, it’s just a natural inertia.

I think if you take it to the energy side of the equation as well, that just amplifies even further. And if you think about, so I suppose two examples, we've come such a long way with renewable energy. I think, you know, if you look at some of the programmes that Great British Energy are supporting now, you know, they're really trying to drive local community investment into power infrastructure. And it's almost a point of pride that they've got that within their community and they're supporting it. How amazing is that?

Nuclear is another example. Slightly different analogy. But if you go to anywhere in the UK where there's a large nuclear power station, which is one of the perceived, you know, absolute key enablers to driving the AI renaissance, there’s generational jobs, economic impact, and everybody is really on board with that. And, you know, very supportive of the technology generally.

Then you kind of introduce a data centre into the equation and say, “Well all this power that we're really proud of and we, you know, we own, as the people that are contributing to that, is being put into this box. And I can't quite see what that’s doing.” I think there's work to do on the license to operate.

But I think that the reassuring thing is, as Caroline rightly points out, if you look past everything that AI and, data centres are enabling the narrative is there, I think we're just going to take... it’s going to take time for people to make that connection.

Caroline Brown (Guest)

Yeah. I think there's also some benefits that are definitely being explored by the data centre, build out companies that are not stories that are being told well. And this is where proximity to cities could be super helpful.

So remembering that big heat output from data centres, there are examples where that heat is being used to provide very, very low cost affordable heating for homes and businesses, heating swimming pools, heating that's used for industrial processes. So instead of using electricity to generate heat for a manufacturing process, you use the heat that's generated by the data centres. So there's this reciprocity arrangement that benefits everybody.

Tom touched on the economic prosperity. So certainly during the construction phase, I don't think anybody would question that there are jobs that are created. But then once the data centre moves into operations, there are ongoing roles and employment for people in those areas. And I think looking at how data centres can make that contribution.

You know, there's been, a really interesting piece of research done. So remembering that comment I made previously about how it's the balance of electricity moving around the grid is really intricate and very, very carefully managed. There comes this situation, and let's use this evening's World Cup football match as the example, right? There's a hydration break. Everybody goes and puts kettle on, makes a cup of tea; that triggers a demand for power that is instantaneous and somewhat challenging to forecast because we don't have a World Cup match every night with a hydration break at exactly the same moment. How do you balance those changes in demand for electricity over the grid?

Data centres can have a role to play there, because they're used to managing and balancing surges of workload. When the computers are really, really busy, they use more electricity. When they're less busy, they use less. So they can add flexibility into that very careful balancing act.

So again, thinking about how we get much more efficient at using electricity in the UK, putting data centres as a contributing factor to that is another plus. So I think we need to do better at talking about the benefits of data centre build, not just the use of AI, but actually how they can make a contribution in that area next to that city at the end of your back garden.

Tom Cope (Guest)

I think we're slightly biased because of the sectors that we operate in. We are. So apologies for all the power analogies, but, you know, all of everyone's life, right? Healthcare, the fundamental change that it's going to drive and the benefit for everybody, I think it's hard because it's not one person's job to do that and amplify that. But it's so fundamental to unlocking the investment to unlocking the projects, to unlocking the energy infrastructure that's needed to power it all.

Oli Carpenter (Host)

Well, it's great to hear both your perspectives on it, because I think it's so easy to get caught in that bottleneck challenge. And admittedly, you know, I've read across the press there's a lot of negativity around the infrastructure piece, the impact on our energy and sustainability. But if we look at the flip side of that, and both of you have articulated some great points around some of the strengths that this can bring to the UK.

But Tom, I'd be interested to hear from you around this, this kind of flywheel idea. You know, where sort of cheap, clean power attracts investment in data centres and compute, more compute lowers the barrier to AI adoption and optimisation. Greater AI capability can support productivity, energy efficiency, wider economic growth. You know, the list goes on. But there's this flywheel idea. Can you talk to us a little bit about that idea?

Tom Cope (Guest)

Yeah, we might get on to one of my favourite topics, which is nuclear!

At the scale needed, you know, looking at that as an analogy, I think it's a really helpful one. So we've historically built nuclear power stations decades apart. Now everybody's looking and going, “God, you know, we can't get these fast enough.” Not only for our own grid purposes, but also the data centre challenge, because there aren't many comparable technologies where you can build them in quite a confined space.

You know, a SMR it's about the size of a football pitch. Confined space, co-locate, baseload power for the next 70, 80 years. The challenge is, we haven't done that at scale, ever.

How do you unlock lower cost of capital, repeatability, you know, you've got the proven data, people... A lot of it is down to, you know, humans and data, they can see that it's proven, they can see that it works. And then you can, you can apply that power.

Once you've unlocked that to data centres, you can throw that back into your construction methods so that there is a nice circle of life there, that everybody can see. But it's...

I think one of the biggest challenges is, AI and data centres are so prominent at the moment, and it's almost insatiable demand. And, you know, everything needed to happen yesterday. When you're talking about the scale of power infrastructure that's needed, there is a really long tail.

And I was at a conference in Austria, last month where we were talking about this tension. And, you know, everyone's trying to spec, design, race to get the power they need for the data centres of today. And that, you know, there is a huge struggle there.

But the problem is, by the time you've built that infrastructure, the technology would have moved on. We would have made the benefit of some of those gains that you're talking about in the flywheel already. And, you know, grid density would have improved. So actually, you've moved the dial again so your power demand has just gone up because, you know, your efficiency has gone up.

So I think it's really exciting from somebody that's passionate about energy infrastructure, and broader infrastructure, including data centres. And just the journey we’re about to go on.

But there's a real tension at the moment between the patience of people and the concept of a flywheel sounds really straightforward, you know? You do that and you do that and you get the benefit, it’s all great!

But the things that we need to do from an engineering, you know, regulatory processes will need overhauling to respond to that pace. You know, it's a very complex ecosystem that I think at the moment, you know, largely sits with the government or governments to help to unlock.

– The podcast transitions into an episode break for a Social Impact message from our host, with background music. –

Hannah Gowen (Host)

Access to digital services should be a given - not a barrier.

That’s why we’re proud to support Ability Net through Deloitte’s Social Impact programme. Ability Net is a leading charity and global expert in digital accessibility, helping organisations make their websites, apps and tools work for everyone. Because when technology’s accessible for all, everyone can thrive.

Visit AbilityNet.org.uk to learn more about their impact.

– The podcast transitions back to the main episode as the music ends. –

Hannah Gowen (Host)

And a lot of these things, as they always do, they sound great in principle, and you've just touched on there all of the different elements that kind of have to go well to make this work. But there’s the M-word, which is money, which comes into most things, right? So, Tom, I guess there's a big question here of, who funds all of this? You touched on governments there, whose responsibility does it fall on to fund this large-scale change?

Tom Cope (Guest)

I think it depends which part you're talking about. I think, Caroline may want to come in and comment on this as well, but I think it's fair to say the companies that are looking to fund the data centre developments particularly are largely not short of cash. You know, very, very, you know, the large hyperscalers, which I'm sure everybody's heard the term, you know, are strategically incentivised and aligned to invest in these assets.

I think where it becomes a lot more complex is, you know, the value chain that you need to invest in to unlock, unlock the plug to go in and the electrons, is a lot more complex and actually isn't aligned with their normal, you know, with what they've been doing for the last decade with regards to tech, you know, operations, logistics, etc..

So if you take nuclear, for example, you know, there is not, there is not a... Sorry, there isn't a project being developed globally at the moment, largely in the West. If I just generalise a little bit, that hasn't got the hand of the state around it in some form, and that's important, you know, it's not a negative thing. It shows the importance of that technology.

But there is, there's a whole leap of kind of de-risking that needs to be done by government, through business models, funding models. A lot of the work that we advise our clients on, skills, just to... I know we are talking about financing, but, you know, all of these things need to come together to unlock that.

And it's going to require such a large degree of coordination to make those risk ratings drop down to an acceptable level to really underscore what are sensible levels of return, that ultimately, if you don't get that right, the cost of your power goes up. The cost of the... you know, your your compute then goes up and actually all of a sudden, you know, AI becomes an evil thing because it becomes a premium product that isn't accessible by everybody.

Oli Carpenter (Host)

I suppose let's look at the business context as well. And Caroline, interested to get your thoughts on this one. So what should businesses and business leaders be thinking about - because you used the example earlier of, you know, you might put a couple of funny photos onto ChatGPT, for example, get it to change something and then using AI in a very simple way. But for businesses, how is that going to change how they use AI and, you know, projects and usage? What should companies be thinking about with that? Because there's an element of responsible AI usage here as well isn't there?

Caroline Brown (Guest)

Yeah, absolutely. And I think there’s responsible as it relates to ethical use of AI, which means things like, can we make sure that we're not excluding groups of individuals because the model wasn't trained, to include them. Right. So there's a responsible AI, there's a how do we keep data private and secure in the right way when we use AI? That's what I mean by ethical use of AI. And some of the sustainability aspects sit in there too.

How do we make sure that we are designing models and we’re training models in a way that is respectful of the fact that AI has a sustainability footprint, and making sure that we're not training to get to 98% accuracy because we can, or we're not using AI for the sake of AI, that it's the right use case and the right tool for the job.

Then we have the, the other aspect of responsible AI, which is more commercial, this one, most businesses are very interested in. The commerciality of the AI itself. How much does it cost me to use it? That's a really great proxy, by the way, for the sustainability impact. If I'm not spending a lot of money, the emissions impact is less. That's a good rule of thumb.

But actually there's also the flip side of the how much is it costing me to use the AI? How much is the AI, either generating or saving for my business? And I think that commercial responsibility and that commercial opportunity is where most businesses look first. The ethical ones should definitely be looked at as well, don't overlook that. But what's it costing me?

And this is an interesting time. The suppliers of AI solutions, whether it's, Anthropic or Microsoft or Google or whoever your favourite, they're still working through what those business licensing models look like. And we're seeing some changes. You know, there's token usage is one measure. There's a per use, and usage is another measure. That's going to evolve. And, you know, all of the things that Tom was just describing that might have a bearing on what the price point looks like, it’ll be interesting to see how that unfolds over the next time period.

The commercial benefit of AI, how might AI help my business save money? How might AI help my business make money? Those are really easy business cases to explore and I think most organisations go there. So quite a spectrum of things to consider. And each one of them really has equal merit. I think most companies look at the commercial frame in the first instance.

Hannah Gowen (Host)

And will access to energy and compute for companies at some point become a competitive advantage, do you think?

Caroline Brown (Guest)

I think it already does. If we think about really simple things, the ubiquitous use of Amazon for shopping, and then we look at what companies like the big retailers are doing, they're doing it in a way to compete with Amazon. So, you know, the really simple examples, Amazon was one of the first retailers to come up with the, customers who bought this also bought the following. So you buy a 54 inch TV, you might need an HDMI cable. Here's a suggestion. And so we're seeing all the retailers respond similarly. Now it's expanding into other industries. And I think AI already brings very real competitive advantage. I don't think it's a when will it be, I think it's there right now.

Oli Carpenter (Host)

And we're very quickly coming to the end of the pod; I think we could spend a lot more time talking about this, it’s been a fascinating conversation. But as we do come to a close, it would be good to look at some of the takeaway advice for our listeners and think about some of those concluding thoughts. So, Tom, if I turn to you first, perhaps the biggest thought at the end of this is where does the responsibility for delivering this all lie?

Tom Cope (Guest)

Wow.

Hannah Gowen (Host)

Not with you. I mean, individually.

Tom Cope (Guest)

I'd love to help. Wherever we build infrastructure, people benefit and citizens benefit. Ultimately, so the short answer is there's a need for a coordinating and enabling role with government. And from government. And I think there is lots going on at the moment to do that.

The challenge is, because I can hear people listening, people always say that's not quick enough or not enough. But, you know, we live in a very complicated world at the moment, where there are lots of trade-offs which governments are trying to manage.

Where does really a lot of the commercial benefit sit? In the hyperscale with the hyperscalers, with some of the, you know, AI businesses that Caroline just referenced. I think they've been quite busy for the last couple of, well, for the last decade. And actually, with the resources that they have built up, I think there's a real opportunity to broaden out and move towards government on a lot of this stuff. With appropriate investment and, you know, appropriate commercial structures and, you know, everything else that comes with that.

I think there's a tension there that we haven't quite got our heads around some of it. I think it's inevitable that government has to be the enabling, driving force. But, I'd hope and expect to see the expansion of, roles, responsibilities, investment and also risk appetite for those to enable all that stuff that needs to be done either side of, you know, either side of that focus value chain, which is the compute and the data centre to enable AI.

Hannah Gowen (Host)

And Caroline, there's clearly no silver bullet to any of this. It's not easy. But what's one thing that business leaders should understand about the relationship between AI, energy, and infrastructure that should influence their thinking?

Caroline Brown (Guest)

I'll repeat what I said a moment ago, that the cost that you pay is a very good proxy for everything else. Cost has a very direct and immediate impact on a company's bottom line. Keep an eye on your costs. Help your employees understand the implications of the choices that they make when they're using the technology. Because it has a direct cost impact. And that in turn, I think is the best possible proxy for that impact on the energy system and the underlining infrastructure.

Oli Carpenter (Host)

Fantastic. And last but certainly not least, we always like to bring it back to our big question. So, finally, let's do that. Will energy decide the AI race?

Caroline Brown (Guest)

Yes, I think it will. I think it ultimately it will come down to that, but I don't think it's that in isolation.

Oli Carpenter (Host)

Tom, any other thoughts?

Tom Cope (Guest)

It's inevitable. It already is in my point of view. I think, you know, it's also driving massive acceleration in the policies, the investments that's needed to unlock it. So I think we're already seeing it.

Oli Carpenter (Host)

Well, Tom, Caroline, thank you so much for your insights today! It's been a brilliant conversation. Lots to talk about. Lots that's going to happen, I'm sure, over the next few months and, few years. But thank you very much for joining us in The Green Room.

Tom Cope (Guest)

Thanks for having us.

Caroline Brown (Guest)

Thank you.

-- Outro music of The Green Room begins followed by our host talking –

Oli Carpenter (Host)

Thanks for listening to this episode of The Green Room by Deloitte.

We release a new episode every other Tuesday with another big question so don’t forget to hit follow or subscribe to this podcast wherever you’re listening or watching – and make sure your notifications are on so that way you’ll be alerted whenever a new episode drops.

This podcast is produced by our very own pod squad. Original music by Ali Barrett.

-- The podcast music fades up and the episode ends --