TRANSKRIPTIO 🔗
WEBVTT
00:00:01.439 --> 00:00:05.519
Hello and welcome to Talking Power with EDF Business and Wholesale.
00:00:05.679 --> 00:00:07.040
I'm David Edmonds.
00:00:07.200 --> 00:00:14.560
Today I'm joined by Jamie Allen from Iron Mountain Data Centres and David Ferguson from EDF's UK Innovation Team.
00:00:14.720 --> 00:00:22.719
Jamie is the head of site selection for Iron Mountain Data Centres, where he leads capital projects, real estate acquisition and site infrastructure.
00:00:22.879 --> 00:00:29.920
David is head of innovation at EDF, leading research and development into technologies and solutions that will shape the future energy system.
00:00:30.079 --> 00:00:35.679
We discuss the connection and cost challenges facing data centres looking to connect to the UK energy grid.
00:00:35.840 --> 00:00:42.640
In particular, it's interesting how there may need to be compromise between what the UK grid can offer and the needs of data centers.
00:00:42.799 --> 00:00:49.520
This episode is a great listen for you if you're keen to learn more about a critical new technology and what needs to be overcome to enable it.
00:00:49.840 --> 00:00:51.439
Enjoy the conversation.
00:00:54.640 --> 00:00:58.880
Hello and welcome to Talking Power with EDF Business and Wholesale.
00:00:59.039 --> 00:01:06.799
I'm joined today by Jamie Allen from Iron Mountains Data Centres and David Ferguson from our RD innovation department at EDF.
00:01:07.040 --> 00:01:09.359
Today we're going to talk about the data center challenge.
00:01:09.439 --> 00:01:11.439
So, Jamie, I'll jump straight in with you.
00:01:11.599 --> 00:01:14.000
What are the big energy challenges for data centers?
00:01:14.239 --> 00:01:15.599
Gosh, where do we start?
00:01:16.000 --> 00:01:16.879
There's so many.
00:01:17.040 --> 00:01:25.760
I mean, if just kind of work it back from who your typical customer is that is leasing the data center or self-building the data center.
00:01:26.239 --> 00:01:33.120
They're looking and have always looked for 2N and a reliable connection.
00:01:33.599 --> 00:01:36.959
The number of nines after the 99 is debatable.
00:01:37.519 --> 00:01:46.000
But a redundant and resilient connection is their bare minimum requirements, which is the hardest part to solve from a power perspective.
00:01:46.799 --> 00:02:11.919
We have a kind of maturity issue over the last several years across the EU, but specifically in the UK, where we are have essentially been flooded with speculative connection agreements because that's the only way to generate value on a potential uh piece of land is to try and get something close to a feasibility assessment for when you're going to get power and how much.
00:02:12.400 --> 00:02:23.120
So that's been an issue over the last several years, is that we can't actively prioritize which connections should go first based on that maturity perspective.
00:02:23.520 --> 00:02:24.800
And then the scale.
00:02:25.039 --> 00:02:36.240
So a typical kind of data center six or seven years ago, you wouldn't think much of um going beyond a kind of 12 to 20 megawatt data center.
00:02:36.319 --> 00:02:41.120
Like that would be very reasonable for Greater London in the typical slough.
00:02:41.439 --> 00:02:44.240
Most of the data centers are kind of around that size.
00:02:44.479 --> 00:02:58.960
Um and now it's just the order of magnitude has increased so much, it's unbelievable, which is great for the industry, like the growth that it's seen from a cloud storage, inference in AI training.
00:02:59.039 --> 00:03:19.599
Um all of this is picking up in such a huge way, which is kind of usually a mirror to lots of other industries and how well they're doing and how much they're using those services, um, as well as how much is coming off of on on-premise, um, which is what is still a huge amount of opportunity to move to the cloud now.
00:03:19.840 --> 00:03:30.719
So a very long answer for lots of challenges, but most of it is uh we need more power, uh, we need it to still be the same quality that our customers require.
00:03:30.879 --> 00:03:33.759
Um, and um we need it today.
00:03:34.479 --> 00:03:39.919
And when you say speculative grid connections, interesting phrase, what does that really mean?
00:03:40.080 --> 00:03:43.919
How how much capacity is there waiting and how real is that capacity?
00:03:44.240 --> 00:03:57.439
Yeah, so I think um it it really ranges based on what are some of the uh plot options out there and how much is physically possible to even put on there from a test fit perspective.
00:03:57.680 --> 00:04:07.599
But you'll see a number of opportunities that might be classed as speculative because the land isn't owned at that point, may only be optioned.
00:04:07.759 --> 00:04:30.480
Um, you may not even have industrial zoning at that point, and it's it's so that's why it's just so speculative because um you you don't have an idea of who your off-taker is for that land, um, and there's land you know packaging essentially being undertaken to just try to drive any piece of value increase uh from that land.
00:04:30.800 --> 00:04:36.480
Um and and as we've seen, 50% of them have not materialized.
00:04:36.720 --> 00:04:48.560
Um, A, because you can't hold on to an option for a piece of land for 10 years, which is what some of the responses have been for power um coming back, which is a real issue.
00:04:48.800 --> 00:05:01.120
Um, and then B, um they it was so highly speculative that there's no um maturity funding being put into place to then progress zoning, permitting alongside of it.
00:05:01.279 --> 00:05:13.199
So you have one side of the coin be uh attempted to be speculatively progressed, which is the power, and then not the other side because it that there's no real money behind this opportunity.
00:05:13.360 --> 00:05:16.240
So there's quite a lot of those out there in general.
00:05:16.720 --> 00:05:16.959
Okay.
00:05:17.120 --> 00:05:21.600
And is this a general problem around
Load full transcript
the country or is there a locational element to it?
00:05:22.000 --> 00:05:27.120
Um it's a general problem around the country and just around the EU, um, I'd say.
00:05:27.360 --> 00:05:45.040
And there, you know, a number of countries have now um input this maturity process to try to essentially streamline uh what is mature and what isn't, um, and just get to what is the size of the prize that we're having to perform grid reinforcement works to.
00:05:45.439 --> 00:05:52.800
So I wouldn't necessarily say it's specific, but 90% of the data center industry in the UK is in Greater London.
00:05:52.959 --> 00:05:59.600
So it's going to always appear like it's happening around uh essentially around the M25 and beyond.
00:06:00.240 --> 00:06:00.560
Okay.
00:06:00.800 --> 00:06:03.600
And what sort of solutions, maybe David, this is a chance for you to come in.
00:06:03.680 --> 00:06:10.560
What sort of solutions are out there from uh an industry perspective, but also what can we help with as EDF with some of those challenges?
00:06:11.279 --> 00:06:12.959
Well, uh there's a lot we can do.
00:06:13.199 --> 00:06:15.120
I think we do need to be quite sensible about this.
00:06:15.279 --> 00:06:25.199
There's something like 50 gigawatts of data center uh connections in the queue, which is clearly not going to work when the total UK energy system capacity is 40 gigawatts.
00:06:25.600 --> 00:06:30.240
But even if only 10% of that is built, it's gonna be really quite significant.
00:06:30.800 --> 00:06:35.839
So, what we've been doing over many years is we supply power to data centers.
00:06:35.920 --> 00:06:39.759
So we've been a big power supplier to many different data center operators in the UK.
00:06:40.000 --> 00:06:46.160
Uh, we do facility management and mechanical and electrical engineering work for data centers.
00:06:46.480 --> 00:06:49.199
That's kind of our bread and butter, and we've been doing that for a long time.
00:06:49.360 --> 00:06:58.240
But what's increasingly happening is that developers and operators are coming to us saying, How can you help us get power faster?
00:06:58.480 --> 00:07:00.959
Um, what are the different ways of doing that?
00:07:01.199 --> 00:07:11.279
Uh, whether that's connecting to solar projects or battery projects uh that we're developing ourselves, whether that's uh developing a flexible grid connection.
00:07:11.600 --> 00:07:18.000
Um and then further beyond that, there are things like uh different models of power supply.
00:07:18.639 --> 00:07:35.199
So a number of our customers are asking for 24-7 power agreements, so where you're matching low-carbon power with consumption at a half-hourly or hourly level, which is really quite complicated to do um and uh complicated to deliver as well.
00:07:35.759 --> 00:07:40.319
And then there are additional things like um recovery of heat from data centers.
00:07:40.399 --> 00:07:50.000
So there's a whole load of things that our customers are asking us to do, but fundamentally it's addressing the problem that Jamie was talking about, which is getting getting access to power, grid connections as quickly as possible.
00:07:50.399 --> 00:07:50.560
Yeah.
00:07:50.800 --> 00:07:54.079
And you touched on a couple of other challenges by sounds there as well.
00:07:54.160 --> 00:07:57.120
So the sustainability angle and renewable power.
00:07:57.439 --> 00:08:03.839
I guess question one, how much of a challenge is that for data centers that they want to get hold of that renewable or low carbon power?
00:08:04.319 --> 00:08:07.519
And then another one, I suppose, related to the connections, is is cost.
00:08:08.000 --> 00:08:11.519
How much of a factor is energy costs for data centres?
00:08:11.680 --> 00:08:17.120
And is that uh maybe a lever that could be pulled in order to solve some of the connection challenges that are there?
00:08:17.680 --> 00:08:19.360
Yeah, they're they're really good points.
00:08:19.439 --> 00:08:23.279
And I think start in maybe with the electricity costs.
00:08:23.439 --> 00:08:44.159
Um these are things that you can potentially shield against when you're closer to a major metro, like closer to London in Greater London, where you're gonna have small to medium-sized data centers, um, you can theoretically um get away with a uh a much more expensive electricity cost compared to other parts of the EU.
00:08:44.559 --> 00:08:52.960
Um, because you have such strong industry driving inference in compute and storage, um, these things can be shielded slightly.
00:08:53.120 --> 00:09:20.399
But when you come out and you're you know 100 kilometres out from Greater London and you don't have those same customers to rely on, and the only thing you have to offer is AI training, which is just a tiny fraction of the electricity costs in, say, Finland, it's very difficult to get any customer to want to go to the middle of the UK and build a gigawatt data center when they can do it so much cheaper in other parts of the EU.
00:09:20.639 --> 00:09:30.559
So I think that total cost of ownership in electricity cost is huge once you're outside of Greater London, but you're very protected inside of it for the time being.
00:09:30.720 --> 00:09:32.639
Um so that that really is important.
00:09:32.720 --> 00:09:53.519
And from a sustainability perspective, um, there is there's no major customer out there today that doesn't have net zero goals and sustainability um pledges that have been made uh most of the time 2035, sometimes sooner, sometimes a little bit later, but they're imminent.
00:09:53.600 --> 00:09:55.519
Um so there's no way around those things.
00:09:55.600 --> 00:10:01.919
Like they all need to be reducing their carbon, uh decarbonizing by all all different methodologies.
00:10:02.080 --> 00:10:03.600
So that is really important.
00:10:03.919 --> 00:10:14.320
Uh there's I just don't see an there's no other way around it um than to think through those costs and to essentially absorb them into your
total cost of ownership.
00:10:14.399 --> 00:10:18.240
Aaron Powell Do you do you think that AI training data centers are going to be built in the UK?
00:10:18.799 --> 00:10:29.360
Uh I think there's lots of optimism and uh initial investment uh for purely the training element of AI data centers.
00:10:29.600 --> 00:10:35.759
My and and like we have lots of opportunity across the UK, like all these ex-industrial sites.
00:10:35.919 --> 00:10:42.559
Um, but I guess I go my simple thinking is that why are they ex-industrial sites?
00:10:42.720 --> 00:10:50.480
Well they've left because it's cheaper to go and manufacture things elsewhere, and why would that be any different to AI training data centers?
00:10:50.559 --> 00:10:59.279
Like they are always going to follow the cheapest path of total cost of ownership, which is led predominantly by electricity.
00:10:59.519 --> 00:11:09.919
In terms of um capex costs per megawatt, there's not a huge amount of difference wherever you go because switch gear costs the same, whether you're here or France, it doesn't make any difference.
00:11:10.159 --> 00:11:11.759
Labour is very similar.
00:11:11.840 --> 00:11:20.960
Uh, and you could argue if you're in the middle of Finland, for example, labor might be more because there's no human beings out there, so you need to bring them out there.
00:11:21.200 --> 00:11:38.320
Um so those things are fairly consistent, but it's only really the electricity cost that helps drive you where you should go for those huge, large um models that are being that are being created to train all this data in.
00:11:38.559 --> 00:11:56.879
Um so I think unless you're in the protection of the tier one region like London, where you've got all these other services that need to operate at one millisecond or less, and you there's a real premium associated to it for financial services and um lots of other industries.
00:11:57.039 --> 00:12:08.559
Uh, it's hard to think why we would want to build gigawatts worth of capacity um in the UK just from that TCO perspective.
00:12:08.799 --> 00:12:20.159
Uh I think it will happen because there will be sovereign uh requirements that will be out there that will need AI training that needs to be air-gapped, you know, only to the UK and and things like that.
00:12:20.320 --> 00:12:24.799
But they're not a huge percentage of the overall workload request.
00:12:25.840 --> 00:12:26.159
Okay.
00:12:26.559 --> 00:12:34.559
Are there any sort of policy measures that could be taken to make the UK more attractive outside of energy costs that we come back to?
00:12:34.879 --> 00:12:41.200
But is there anything that can almost allow you to have a bit higher cost but still attract data centers into different regions?
00:12:41.600 --> 00:12:57.120
Yeah, I think um if you're purely bringing your own power and it's gas generation or some other form of it, there could be more support to um kind of catapult that to the front of uh front of minds at the moment.
00:12:57.440 --> 00:13:06.720
As it stands, there's no simplified method of making that go any faster than it is or any smoother from a community perspective.
00:13:06.960 --> 00:13:18.159
Um and uh the only other kind of policy change other than the electricity cost itself um would be that that you you there's more ability to bring your own power.
00:13:18.480 --> 00:13:36.879
Um so I I there's nothing that really jumps out at a moment other than getting through all of this the next 10 years of grid reinforcement works that have been stopping the ability to have firm dates that you can invest against, um, and then the cost of electricity.
00:13:37.039 --> 00:13:41.360
Like those those are really impacting the UK data center industry.
00:13:42.639 --> 00:13:42.960
Okay.
00:13:43.200 --> 00:13:46.399
So if we take take cost as a given, then then it's there.
00:13:46.799 --> 00:13:54.639
I guess David, more from a supplier side or energy partner side of things, what could we do to help with manage the cost of data centers?
00:13:54.799 --> 00:13:58.000
What technol technological solutions are out there that can help?
00:13:59.519 --> 00:14:02.639
Well, that's uh a a a difficult challenge.
00:14:02.960 --> 00:14:11.679
If if cost is if the cost of power is the dominant cost for a data center, then well, I guess it's our usual box of tricks.
00:14:11.759 --> 00:14:35.279
So having corporate PPAs that we offer to our um customers already, there are flexibility markets that allow you to uh monetize some of the inherent flexibility in the the way that you use power at that site, but you know, if you if you look at the overall package that a data center is uh package of costs that a data center is going to be incurring, these are probably gonna be relatively small, I imagine.
00:14:35.679 --> 00:14:39.120
Um so I think it in reality it's relatively limited.
00:14:40.240 --> 00:14:40.399
Okay.
00:14:40.720 --> 00:14:41.840
I don't know, what do you think?
00:14:42.639 --> 00:14:47.279
So I think there's big opportunities, certainly when you look at what makes up the energy bill.
00:14:47.519 --> 00:14:47.679
Yeah.
00:14:47.919 --> 00:14:54.639
In terms of non-energy costs, reducing your net consumption through having on-site generation self-supplying to some extent, that's there.
00:14:55.039 --> 00:14:58.960
If you can target that to the right times a day, there's more opportunities to do that.
00:14:59.519 --> 00:15:05.039
You say it depends on the total cost stack for a data centre, how much of an impact that makes for the energy bill.
00:15:05.360 --> 00:15:14.000
Um but then I guess if it's not making that big an impact, then arguably it's a levelizer across different locations anyway, if energy bills aren't such a big factor.
00:15:15.039 --> 00:15:15.279
Yep.
00:15:15.440 --> 00:15:30.559
And one of the things we have been looking at is effectively behind the meter supply from uh well, the the dream for many of our customers is behind the meter
supply from nuclear, because the production profile of nuclear matches very nicely the demand profile of of data centers.
00:15:30.799 --> 00:15:33.840
That's not always practical for various reasons.
00:15:34.080 --> 00:15:41.600
But behind the meter supply from renewables, so solar, uh, and maybe piggybacking on batteries as well could be really attractive.
00:15:41.840 --> 00:15:47.600
Um, but yeah, that's that's going to make a dent in the total cost, but it's not going to be transformative, I don't think.
00:15:48.000 --> 00:16:00.320
And and that goes back to that very first thing of nothing has really changed for the predominant customer of data centers needing a 2N um reliable connection.
00:16:00.559 --> 00:16:16.399
Um and I think the uh the way round that at the moment is to tell a really good story and prove how you can uh re-engineer the exact same output but through a completely different method than what you've seen before.
00:16:16.639 --> 00:16:28.240
Um, and I think Bears is getting better year over year, like we're seeing the ability for the durations to get longer, which are which will support more customers.
00:16:28.399 --> 00:16:39.120
But it does unfortunately favor the non-traditional customers now, like what we call the neo cloud customers that are focusing more on GPU as a service.
00:16:39.279 --> 00:16:44.080
So essentially, they don't lease the facility, they lease the chips.
00:16:44.240 --> 00:17:01.519
Um, so it's much simpler for them to be able to have a theoretically less reliable or less proven reliable um piece of infrastructure behind the chips, um, because they're not penalized in the same way from their customers if anything happens.
00:17:01.759 --> 00:17:08.880
Whereas how all data centers have kind of been leased in the past is an is a facility perspective.
00:17:08.960 --> 00:17:17.920
That's the product that you're you're leasing, and that's why there's been such a focus on redundant, reliable uh power uh photos.
00:17:18.319 --> 00:17:33.039
So I think definitely that is a uh a huge uh pro for the industry, is that we now have these neo clouds that are able to make use of those different types of non-traditional connections.
00:17:33.279 --> 00:17:47.599
Um and uh very slowly uh other customers may come round uh to those uh if you can prove with the story that you're coming the output is the same, essentially, or close to the same.
00:17:47.839 --> 00:17:57.519
That's the hardest thing, I think, out there today is is proving that they will get the same quality of connection as they would to a traditional uh grid connection.
00:17:58.240 --> 00:18:01.920
And how suitable is that for different use cases of data centers?
00:18:02.079 --> 00:18:03.680
So not all data centers are the same.
00:18:04.000 --> 00:18:06.559
Some of them that might be a more appropriate solution for than others.
00:18:06.640 --> 00:18:08.559
Is is that something that can work for the majority?
00:18:09.039 --> 00:18:28.160
So for for the types of the three types of compute that you would typically have in Greater London, which is inference in, storage, and compute, um they they are not traditionally well known with any type of anything other than very reliant, very reliant, very redundant power.
00:18:28.319 --> 00:18:34.640
Um it's only really AI training that we have some flexibility in at the moment.
00:18:34.960 --> 00:18:43.839
Um and it's down to the technology mix to prove that you can use it for these latency-sensitive workloads.
00:18:44.079 --> 00:19:08.000
Um, and that I there's been some really good progress made in turning non-firm connections into firm connections, and um, there's there's some really good opportunity out there, but all of those opportunities uh requiring a big amount of CapEx and risk to develop them so that they're ready to offer to a customer.
00:19:08.160 --> 00:19:15.759
Um, and all the time you if you just have a theoretical ask, like, would you accept uh you know, a couple of non-firm connections?
00:19:15.839 --> 00:19:25.039
It's very difficult to get traction on those until it's real and someone's already invested the money to make it oh, it's actually going to be here next year, that connection.
00:19:25.119 --> 00:19:27.359
Uh, that changes a story drastically.
00:19:27.519 --> 00:19:36.960
They'll suddenly start considering that a lot more when it's near-term and someone's already invested most of the journey to make that connection a reality.
00:19:37.440 --> 00:19:40.000
But some of this is down to customer expectations, isn't it?
00:19:40.160 --> 00:19:48.400
So you're talking about the different types of of use here in London, and and there's always that assumption that you need to have your five nines of reliability.
00:19:48.640 --> 00:19:58.799
But then you look at what the uptime is for some of these AI inference services, like Claude, you know, their uptime is 99.5% of the time, which is not five nines.
00:19:59.039 --> 00:19:59.119
No.
00:19:59.519 --> 00:20:00.720
And we're all okay with that.
00:20:00.880 --> 00:20:06.079
And the latency is often a little bit of an issue because you know, the the AI is thinking.
00:20:06.240 --> 00:20:13.119
So, in a way, our our uh uh acceptable level of service has changed with some of these new tools.
00:20:13.279 --> 00:20:20.720
So maybe those old assumptions around five nines and you know, very very minimal latency, maybe that is changing.
00:20:21.039 --> 00:20:26.880
Yeah, I think it will have to, uh, because it's got no other choice but to change slightly.
00:20:27.039 --> 00:20:43.200
Um the I guess the problem is that over the last kind of 15 to 20 years of the major hyperscalers building up their products and what they offer their end customers has been that sort of uh bulletproof redundancy and resilience.
00:20:43.440 --> 00:20:47.200
And uh there the fact is that not all services need it.
00:20:47.279 --> 00:21:07.839
And I think that is uh certainly uh there are
you can see already it there are internal teams uh being created called um, you know, demand shifting and uh all these teams that are trying to be get creative around moving different types of workloads to different data centers for different purposes.
00:21:08.079 --> 00:21:16.079
Um, you know, each workload has a different density of rack, and each workload has a different reliability and redundancy that it's needed.
00:21:16.240 --> 00:21:18.960
Uh, and there are lots of workloads that will just never change.
00:21:19.039 --> 00:21:26.720
Um, if you're a bank and you're trading, you'll never ever be able to have anything less than the highest quality of power.
00:21:26.880 --> 00:21:36.720
And then there are others that you think, oh, maybe I don't need to pay as much for the service, and I can use uh a service that has slightly less redundancy uh in, for example.
00:21:36.880 --> 00:21:44.960
So I think that is being forced to be thought about, um, which it didn't need to be uh six or seven years ago.
00:21:45.200 --> 00:21:46.400
Absolutely didn't need to be.
00:21:46.640 --> 00:21:54.240
You could go and pick up uh a high quality connection and wait for maybe three years, um, you know, six, seven years ago.
00:21:54.400 --> 00:21:58.240
Now there's no tier one region that you could possibly do that in.
00:21:58.400 --> 00:21:59.759
The closest would be France, probably.
00:22:00.160 --> 00:22:01.279
Probably in in the EU.
00:22:02.319 --> 00:22:06.160
But even that is starting to kind of push further to the right all the time.
00:22:06.640 --> 00:22:06.799
Okay.
00:22:06.960 --> 00:22:11.200
So it sounds like it's driving compromise almost in into you can have exactly what you want.
00:22:11.599 --> 00:22:13.119
But you can have something close to it.
00:22:13.440 --> 00:22:24.559
I guess with that in mind, what what roles do you see that energy suppliers, energy partners can play to better facilitate some of that compromise and identify the best places to find some of that flexibility?
00:22:25.200 --> 00:22:40.400
That is a really interesting point because most utilities across the EU, you people will try to tell you they know where the stranded capacity is in substations, but no one has any certainty in it.
00:22:40.559 --> 00:22:51.200
So one easy role they can play is just by identifying where the capacity is, which you can then work backwards from that power and look for plots that are in a suitable distance.
00:22:51.759 --> 00:22:58.720
That that is without a doubt a really strong um option to support things.
00:22:58.960 --> 00:23:11.440
And then the going back to that kind of investment journey for bringing non-traditional supplies around, that is another area that the likes of EDF can definitely support.
00:23:11.599 --> 00:23:18.799
Um bringing those connections to available plots that we're not really going to know about.
00:23:18.960 --> 00:23:25.200
Like I I can't um go to any utility and just say, have you got any non-firm connections?
00:23:25.279 --> 00:23:32.400
or you know, you have to go and ask for exactly what you want first and then be told yes or no or how long it's going to come.
00:23:32.559 --> 00:23:37.359
No one's going to come to me and say, by the way, I do have something if you want it, you know.
00:23:37.680 --> 00:23:42.079
So it isn't quite what you want, but you know, you might be able to do something with it.
00:23:42.319 --> 00:23:44.480
That doesn't really happen uh ever.
00:23:44.720 --> 00:23:47.519
So that that would be a really strong opportunity.
00:23:47.680 --> 00:23:52.079
Is a more proactive use of available power out there.
00:23:52.559 --> 00:24:04.400
I think there's a really interesting dimension to this, which is that the the distribution or transmission system operator, they have a view of what is available at the kind of infrastructure level.
00:24:04.480 --> 00:24:10.079
So they will go down to the substation and see what capacity capacity has been allocated and if there's anything spare.
00:24:10.480 --> 00:24:15.920
What they may not see is how much of that capacity that's been allocated is actually being used.
00:24:16.400 --> 00:24:24.480
Because quite often, down at the you know, building level or customer level, they will have acquired or reserved a certain capacity and then just not be using it.
00:24:24.640 --> 00:24:30.240
So that is being effectively taken out of the system and not used, which is hindering you.
00:24:30.319 --> 00:24:38.000
So I think there's maybe something cunning that needs to be done to unlock that allocated but unused capacity that's out there.
00:24:38.079 --> 00:24:40.880
And our view is there's probably quite a lot of that.
00:24:41.359 --> 00:24:54.880
Yeah, I and I I I'd say that the data center, the legacy data center industry is probably quite responsible for a lot of that, of just uh there is inherently there'll be contingencies on contingencies.
00:24:55.119 --> 00:25:02.960
The worst thing that a hyperscaler or other cloud service provider could imagine is that they lose power somehow.
00:25:03.039 --> 00:25:13.279
So there's all this buffer that's added in to ensure that um as those racks kind of fluctuate throughout the day, there's no risk of them ever exceeding 100% of their utilization.
00:25:13.519 --> 00:25:35.440
And then exactly as David mentioned, if you just if that's just one rack that you've got that buffer on, you imagine a whole data hall, a whole data center, multiple data centers, there's all this buffer out there that theoretically you should be able to take a view on from a utilization factor perspective and start applying it to it and saying, I'm gonna bank some more out of this.
00:25:35.519 --> 00:25:51.200
You know, I feel comfortable that there is a a str uh a large amount of power that just isn't ever going to be used because of um buffer on buffer on buffer, um, which I think most hyperscalers are are certainly doing from my perspective.
00:25:51.359 -->
00:25:54.480
They're getting closer and closer to maximising it.
00:25:54.640 --> 00:26:13.039
Um, it takes time, and then there's also facilities that won't allow you to do that because you run out of floor loading or you run out of space in the facility, and uh your facility just wasn't designed for today's needs, and now we see the refurb cycle that starts up to try and utilize all of this power.
00:26:13.200 --> 00:26:27.200
Um, so there's lots of uh sort of secondary industries that are now being kind of born from basically um rack densities increasing and uh lack of fully utilized uh racks.
00:26:27.440 --> 00:26:36.880
But it's it this is just one industry, there's multiple industries across the UK where this is the same thought process, and there is definitely buffer out there.
00:26:37.279 --> 00:26:41.200
How how do you think about or do you ever look at peak shaving?
00:26:41.440 --> 00:26:51.759
So, my my understanding is that you know when you're building or designing a new data center, the cooling system will be designed for 3 pm on a Wednesday in July.
00:26:52.000 --> 00:26:52.160
Yeah.
00:26:52.400 --> 00:27:01.680
Because that's when you're gonna meet your absolute maximum cooling need, but you're gonna meet that once a year, but you're buying the capacity for that one particular moment.
00:27:02.000 --> 00:27:09.279
How what what do you do to lop the top of that peak in order to be able to have more capacity at your side?
00:27:09.680 --> 00:27:21.839
So if I was the end customer, the actual user uh of the compute, you you'd have many more options available to you if you were able to pick and choose when you use the workload.
00:27:22.000 --> 00:27:34.319
Um and you'd be able to plan things just like uh show my age, I suppose, when in primary school, when you needed to use the computer, you book time onto the computer so it knew exactly when everyone was using it.
00:27:34.480 --> 00:27:47.519
Um but then uh nowadays, I guess, the problem for colo, co-location companies and developers and people providing the uh kilowatts to customers in in the form of capacity.
00:27:47.920 --> 00:27:51.519
Um there owned there it goes back to that segmentation of customers.
00:27:51.680 --> 00:28:06.160
Like our typical customer um is only utilising us because of the high, the the ability to guarantee that they can just flip and go to 100% utilization if they need to at any point in time.
00:28:06.480 --> 00:28:10.319
They're they're paying a premium for the ability to do it.
00:28:10.480 --> 00:28:20.400
And then there's this new segmentation of customer that could absolutely um there's no reason that they couldn't choose when the workload is run.
00:28:20.559 --> 00:28:25.119
Um, and they are, I'd say, perfect candidates to to leverage this.
00:28:25.200 --> 00:28:27.200
And they should get something in return for it.
00:28:27.359 --> 00:28:34.880
And that thing in return is perhaps a cheaper power solution, but a more near-term uh power as
well.
00:28:35.119 --> 00:28:50.799
So everyone could be winners uh in that if you're an alternative customer that is able to dial up and dial down your workload, um, which a large majority, increasingly large majority of workloads could fit that bill.
00:28:51.359 --> 00:28:55.440
Which I think leads on to the last topic I wanted to really look into with you.
00:28:55.519 --> 00:29:00.160
Is is it feels like there are opportunities there if we can only drive the right incentives.
00:29:00.400 --> 00:29:09.119
So are there any policy or regulatory changes in particular that you'd want to see that could help drive that right behaviour and actually solve some of these challenges?
00:29:09.920 --> 00:29:29.680
I'm kind of biased, I suppose, and the thing that I always think about is the cost of electricity uh and how we can uh support the end customer who's the ultimate person paying for this, uh less so to the developer that's trying to uh you know bring uh bring their CapEx alive into an opportunity, I suppose.
00:29:29.920 --> 00:29:37.680
But um that is the one area that uh just absolutely cripples the UK um from what it could be.
00:29:37.839 --> 00:29:39.759
It it could be so much more than it is.
00:29:39.920 --> 00:29:44.880
So any sort of policy change that can drive that cost down is so welcome.
00:29:47.519 --> 00:29:53.039
I think potentially around um how we can accelerate grid connections.
00:29:53.119 --> 00:29:57.839
I mean, that is the big challenge for a lot of the developers is getting access to power as quickly as possible.
00:29:58.000 --> 00:30:11.920
Uh and so we've been talking with National Grid and and other partners about w what are the alternative models for offering uh connections uh instead of having a firm connection, having a flexible connection, but much, much faster.
00:30:12.079 --> 00:30:18.319
Uh and how do you then deliver the flexibility that's inherent in that promise from the developer?
00:30:18.720 --> 00:30:26.880
Um and and we think there's enormous potential there, actually, because that's what people are really crying out for, is just access to power as quickly as possible.
00:30:27.440 --> 00:30:29.279
Yeah, that message has definitely come through.
00:30:29.759 --> 00:30:37.440
Yeah, it feels like at the of course when it's theoretical, um, the answer more than likely might be maybe or no.
00:30:37.599 --> 00:30:43.200
When it's real, and it's oh, in 18 months you might be able to bring me something non-traditional.
00:30:43.440 --> 00:30:45.200
That might that might change my answer.
00:30:45.440 --> 00:30:51.119
We need more of those opportunities coming to the near term to be able to start to change minds.
00:30:51.519 --> 00:30:56.559
Yeah, we've been talking to a few different companies, and when you talk to the operations team, they say, absolutely no way.
00:30:56.720 --> 00:30:58.799
There's no way I'm going to operate flexibly.
00:30:58.960 --> 00:31:00.079
Why would I want to do that?
00:31:00.160 --> 00:31:04.240
And then you talk to the development team, the grid connection team, and they say, Well, yeah, maybe.
00:31:04.720 --> 00:31:05.839
Could be interested in that.
00:31:06.000 --> 00:31:08.000
If you can if you can bring it to me faster.
00:31:08.480 --> 00:31:08.799
Yeah.
00:31:09.039 --> 00:31:11.599
Yeah, all sounds like drive driving the right incentives.
00:31:12.160 --> 00:31:12.319
Yeah.
00:31:14.000 --> 00:31:14.880
Okay, excellent.
00:31:14.960 --> 00:31:16.480
Well, thanks both for for joining me today.
00:31:16.720 --> 00:31:17.039
Thanks for having me.
00:31:17.279 --> 00:31:17.839
Thank you, David.
00:31:17.920 --> 00:31:18.079
Yeah.
00:31:19.279 --> 00:31:19.599
Great.
00:31:19.920 --> 00:31:20.720
Thanks for listening.
00:31:20.880 --> 00:31:23.680
Uh and if you want to join in the conversation, we'd love to hear from you.
00:31:23.839 --> 00:31:28.079
So please reach out via LinkedIn or in the link in the description below.