17. The data centre opportunity: Insights from the NZ DC Leaders Summit

Welcome back to the special conditions, your weekly podcast on construction and infrastructure insights in New Zealand, I'm Hamish Polland. And I'm Sam Holden. Today we've got a topic that I know is dear to your heart and also a real era of interest for me and that is data centres. Hamish I know you've been at the NZ Data Centre Leaders Summit so I'd love to hear a bit about that and perhaps we can cover some of the legal issues associated with data centre projects in New Zealand. So I'm really looking forward to getting into this one, why don't you tell us a bit about the summit? Absolutely and look I know we sort of try to keep these things to 20 to 25 minutes, we want to float to you Sam that will revisit this on a 4 and a half hour podcast, there's a lot to cover and we may go there on that. Anyway the takeaways, it's such an interesting and quite divisive topic globally at the moment, some enormous numbers, you know just to give you a bit of a snapshot and this sort of came out at the start, in 2024 there was 24 gigawatts of active data centres globally, now in 2026 there are 71 gigawatts active and the current development pipeline, you know people are expanding on that pipeline all the time but as that last week it's sitting at 280 gigawatts of further development out to 2030. So that's a massive amount of construction projects for people like you and me to get excited about. And what it pours down to in some other numbers, there's 11,800 DCs globally, just under half of those are in the US at 5,400, there's roughly 300 in Australia, and we actually have 81 data centres already here in New Zealand. So a lot of build out, governments are very interested in this topic because what it's starting to translate to is this idea of exporting electrons, forming a part of a country's GDP and really starting to make significant contributions to that GDP. There's other key topics in there like the data is a national asset, the DCs themselves are being described as quote critical infrastructure, you recall we discussed the trip to the Norwegian data centre where that built to withstand basically terror type attacks and requiring defences. So yeah over the last five years this has been an enormous industry, power consumption is transitioned from a steady sort of 5 to 8 annual enterprise growth to exponential double digit annual increases, and this is largely driven by a shift from those enterprise model data centres to generative AI model training, and global compute another statistic is increasing 18% per year, global institutional capital markets, earmarking, capex, this decade to exceed $1 trillion USD, this is a very very big business and look there in town here people are looking to build major data centres in New Zealand. Yeah and perhaps we talk about why New Zealand is a attractive place for these sorts of projects. I mean I think about what we have here, we have a clean grid up to 95% renewable, and this makes us quite a marketable location internationally for data centres. It allows some of these hyperscalers around the world to meet their emission targets, because of the amount of clean energy we have here, we've got a pretty stable democracy, we've got good privacy laws, and we're also kind of the gateway to Asia right, really well positioned there. It seems like New Zealand should be an attractive place for data centres, and in fact how much we do have a couple of major provincial mega projects in New Zealand at the moment. We are viewed favourably internationally, and in addition to that New Zealand landscape, there's some other observations you can make from looking overseas, so right now if you look at the US, I mean they have a massive grid crisis, those big data centre alleys North Virginia, Atlanta, Georgia, Texas, Phoenix, Arizona, the transmission substations are massively saturated, so there's big problems there with utility queues for 100 megawatt plus data centres, sort of unable to get permitting within a five to seven year period, and so the growth that those hyperscalers are looking for that capacity means that they're having to look abroad, that the US is sort of not an option. In addition to that, the social licence issue, which we'll touch on a couple of times in this pod, is massively coming to the fore in the US, and the stats I've got here that over 400 municipal and county jurisdictions across the US, 400 have introduced formal legislative moratoriums and zoning phrases, there's some high profile ones that you see like New York, so this is meaning that there's this international capital pivot of global hyperscalers and the institutional funds which are looking for some great returns here, are rapidly looking to deploy there, capital into a stable, call, renewable rich alternative jurisdiction, and New Zealand is putting its hand up at the moment and saying hey, we are shovel ready. So I guess that's led to why we're having this discussion now, and those two projects you mentioned, one was announced last week, that was the CDC, contact one, and the data grid one with Mercury has been in the news for some time, but there's some new business around this, it's worth noting though, that DCs are not new, as I said before, we have 80 in New Zealand, roughly, and today most of our compute has been used in New Zealand, so these are these kind of enterprise data centres, regional co-location facilities that are used by domestic talcos and IT providers like Datacom, Spark, 10 peaks, one New Zealand, you know, you and I were talking just before we came on air about Chapman trip and other professional services firms, we keep our data off site now, obviously, we're not using filing, and so we take up storage base in a data centre, we enter into a master services agreement with an operating, we pay a fee, they guarantee that we'll be able to go and retrieve our data at any time. These are sort of important for business, it's not new to New Zealand, we rely on them now in terms of the impacts of these types of facilities, I mean at the conference, Matt Neal, who is the head of data centres at Datacom, he talked about the Arrowsdale Centre up north, which uses less water than a four person household, so these are not particularly scary things and they're important for regional businesses and even in the big centres. I suppose stepping back for a moment, there will be listeners who, when they think about data centres, they think about facilities where you have, as you say, a range of customers who have their data stored there, but that's not really the only type of data centre, right? It's actually confusing and I think what's being sort of talked about in the media and what's getting all of that publicity is around those sort of AI super compute, big factories, there's an enormous one in career at the moment where they're literally moving mountains to build it, but it's not just those ones, we touched on that multi-tenanted facility I referred to before, where some like a law firm might take up some space, that's what you call a wholesale or a co-location facility, so the data centre operator will commercialise that through master services agreements or long term operational leases, which reserve capacity in that data centre, and they guarantee this uptime performance of what they call the 597, so 99.999, that's how your data will be able to be retrieved as the tenant in that facility. And these are the sorts of facilities, how much that have been in New Zealand for a long time, right? We're quite familiar with these. We're familiar with those and we'll touch on the construction models, but for those ones there, relatively simple, you know, you'll see entities, and you sort of move through into the hyperscalers, now these are looking at the notes and the word massive comes up a lot actually, but these tend to be a massive single tenant or purpose built campus that's designed to hold proprietary global data from a customer, and so these are typically backed by an investment grade tenant like an Amazon or a Microsoft Google Meta, and they enter into, it's often a long term triple net lease through hyperscalers centre like that. Then there's this other concept which we came across recently here, and this idea of NeoCloud. and so this is more of a person that's offering a special platform, these are companies like CoreWeave, Lambda Labs. So what these guys do is that they come in and they take up cloud storage in a data center and then they sort of on supply that storage to their own customer like an Oracle. Because it's not an Oracle, it tends to be slightly less creditworthy player and you've got to be a bit careful about who these NeoCloud operators are. But the attraction is is that you've built a data center rather than going out and having to manage multiple tenants. It's one person takes up the space and on supplies it. So that's the NeoClouds, but then I guess the bit that's gaining all the attention are these AI super compute or AI factories which is seeing the sort of radical engineering departure from the standard cloud data centers. So there's massive power requirements and you sort of going from an enterprise model. The wholesale model we took about earlier which will take up 5 to 10 kilowatts per rack. But these modern and video AI servers with a demanding 40 to 100 kilowatts per rack and this talk that the next generation architecture is actually looking to target 120 to 150 kilowatts per rack. So enormous power consumption and off that mass of power consumption there requires enormous cooling. So that's one of the next big things here. That's the asset class. You've got co-locos, co-lose hyperscale NeoClouds and then the big AI factories. And is it fair to say, I mean we were talking about the kind of wholesale co-location facilities. They are more for the hosting of data and a more conventional sense while some of these larger facilities, the hyperscale is the NeoClouds, obviously the super compute. These facilities are more directed towards AI compute, right? Whether it is training models which you might see in those massive AI factories or even just the sort of compute power to run existing models. They're all AI directed, really. And so you can see another social license issue weaving its way in there. You know, it's not just power and water consumption and also these things are very noisy. But if we are running at a million miles an hour spending a whole lot of money to get to that AI super computing and training these models, there's a nervousness in society right around what that means, you know, lots of jobs, etc. Yeah, and so let's get into how they're built. Because obviously these are, I imagine, quite complicated construction projects. These are not just simple boxes being built here, hey, Mish. So what are the sorts of procurement models that you're seeing for these data centre projects? As I touched on earlier, those smaller multi-tenanted co-lows, you know, we've done a bunch of those. They're relatively simple. It's an NZS 3910 and the data centre operator will often have the relationships with some of the specialised equipment and procure that directly and supply that in for a main contractor to install. And, you know, they look like a shed, you know, you can add some architecture around them to make them look a bit nicer. But relatively simple and even I've seen these built under a model that doesn't use NZS, it's like a construction management model where the owner sort of designs and then goes and procures all the equipment and effectively employs a construction manager to just then manage the installation. And part of those separate contracts which are engaged by the principal directly might be like a builder. And then you sort of have somebody that pulls it all together. So construction management is a model that we've seen. But as you come through into the hyper scale, you know, the first thing you might think is, well, this probably lends itself to an EPC. And we know that the size of these hyper scale projects, right, that the two that I talked about for New Zealand, these will be in the billions of dollars. And so it does start to call into question that traditional EPC model and whether or not an EPC contractor can actually wrap the risk of something that large. So in fact, what you are seeing overseas and particularly in Australia, which is probably the main reference point for us here where they've built fuel ready, is that those T1 contractors will just not take that single wrap over hundreds of millions of dollars of MEP packages and proprietary tech wraps, which, you know, they don't have those relationships necessarily with those suppliers. So they say, well, rather than us actually taking the risk in wrapping that all together, we will effectively be what they call an EPCM, which is where they like a design and construction manager, which pulls together all of those packages and hopefully does it on time. But naturally that model lends itself to significant interface risk. The other thing about that model is that when you enter into an EPC contract, you sort of saying to the EPC contractor, you're responsible for design and you run that design process because you are the expert like a process plant or a milk processing facility or a geo thermal power station. You're in the business, that's what you do, you know how to design and you know how to deliver that. But the difference here is that the end user of these facilities, the hyper scalars, the meters, the Amazon's, the Microsoft's, these are, I don't want to use the word Fussy, but they have extremely specific design requirements and over the course of constructing one of these things which take years, they want the flexibility to actually change the design. So it doesn't make sense for the EPC contractor for constantly throwing variations at them because the hyper scalars wanting to have this level of control. Again, that is one of the reasons why you have this EPC-M model sample. Frankly, the hyper scalars refuse to surrender control over their proprietary designs and direct chip calling loops and building management systems. It makes a lot of sense, right? I mean, you gave the example before of what we would traditionally see EPC contracts, process, plants, power plants. You know, the EPC contractors in those cases have been building those sorts of facilities for decades, right? They are the experts, they're very familiar with it. These data centers much less so, you know, this is a much more modern innovation. And the technology and the art is moving so quickly, I can completely understand that mid-project there might need to be some quite significant changes to accommodate those sorts of developments. So that flexibility would be really important. Probably one way to think about this is, rather than that traditional EPC contractor, which we all know, to think about this contractor coming in is the owner's professional project manager and design coordinator, that's what they do. And then the direct procurement and long lead items, that's then done directly by the principal. So they go out and get the high voltage transformers, switch gear, generators, the chillers, and all the OEM equipment. And another reason why you sort of can't say I'm going to run a process and point an EPC contractor and why you go please build it. Because those long lead items that I just talked about are exactly that. I mean, we are talking two to three years from placement of order to delivery of some of this equipment. So the project has to crack on and start placing those orders now, why you're headed for pointing your EPC in. And how are these projects being funded? Because again, these are very, very big, expensive projects where the art is moving very quickly. How are you seeing these being financed by the principals? Well, I mean, the ones that we've talked about data grid, we know that's in the market, and that would be project financed and now drawing off how that's done in Australia, so it's project financed. Now, what's interesting about that structure, and obviously we do a lot of that in the renewable energy space here. So in New Zealand, most of the solar farms that we are delivering at the moment, solar best, will be project financed. And those lenders are used to seeing a wrapped single point of responsibility for EPC. Challenge here is we obviously don't have that, right? We've got this EPC model, and yet they're still project financing it. And so there's a whole lot of different ways that lenders need to think about this, because there's this huge interface risk it needs to be dealt with. So suddenly you're needing tripartite agreements for the lenders to step into all of those critical suppliers and the long lead major equipment suppliers. We're having these highly sophisticated interface and scheduling matrixes, which are clearly defining who's responsible for what, when it comes to delivery and installation, and then allocated damages kind of designed to offset that interface risk. And in the end, In addition, with these projects being so big and with this interface risk not being wrapped, you see what they call "sponsors credit support". So this is where the sponsors that have formed ProJCO, potentially needing to put in "credit support to fund overruns" out of equity instead of calling on debt because the project finance lender will say there is no more debt, so it has to come from equity. And that's actually not an unusual thing to see in a major project finance transaction. My former life was in the downstream wall and gas sector and you would have a petrachemical facility. It might have 1220+ interfacing EPC contracts. The Sedara Petra Chemical Project in Saudi Arabia was a $25 billion US project and you've got like 20 EPC contracts, where the materiality factor was $500 million US, right? And they're all connecting and interfacing with each other. The sponsors are putting up these credit support arrangements to sort of cover the cost overrun, so expect that that could be a feature as a result of this model here. Yeah, that makes sense. And I mean, what are the sorts of challenges I suppose with these projects coming to market? I mean, is there a market for them? Well, I think New Zealand's trying to figure that out now, Sam. I mean, the number one risk that we talked about at the conference last week was the social license battleground. I mean, there were protestors outside waving signs, not happy, and so it possibly suggests have we engaged early enough, right? There'll be keen focus on this at the resource consenting stage. The Green Party has come out, and before we've even got AI compute in New Zealand, I looking for a 12-month moratorium. So we're going to need to engage with people, Sam, and there's a number of social license subcategories. I'll just talk about a couple of them, so the obvious one is power, is very power intensive. And the fear from the public, not just here, but internationally, is that these hyperscale compute facilities will consume all of our domestic energy, and they're going to put up retail pricing and households, New Zealand maybe has a conversation, is it going to delay the turning off of fossil fuel facilities because we need this generation here? Then you've got job creation and skepticism there, so we're building these massive multi-billion dollar facilities, and look, the people that come in here in New Zealand say this is a great place to do it, will say there's probably the geographic conditions, the renewable energy environment, and the grid to support five to ten of these projects. Now ten's probably ambitious. We've got two on the go at the moment, and let's say it's very early days. But with all that money spent, this is an expectation that there's going to be hundreds of jobs created, and it's going to be good for New Zealand, but I think some of that social license piece is that you actually have very few people required, and a hyperscale facility might have 30 to 60 operational staff. The next point I'm going to say, we won't see it in our lifetime, but it sort of plays into the social license issue. These data centers need captive parcels, right? The conversation overseas is something called SNRs, and that sounds for small nuclear reactors. So the data center needs 24/7, zero emission, electricity supply, and some people might say what better way to cheap that than actually just attach a nuclear power plant to the data center. That's an interesting one. I don't think New Zealand is quite ready for that. You're sure we're quite ready for that one, but it certainly is an interesting idea. You can see the benefits. What about some of the other concerns you hear about with data centers, things like noise and water, to what extent are those social license issues, a barrier that might need to be addressed? Oh, I mean, data grid's got resource consent, right? So they've overcome that, and possibly it's because it's an invocable miles away, but yeah, absolutely. That's going to be a real focus of the consenting authorities, you know, acoustic and noise limits. This is this sort of continuous, low frequency hum that you hear from the hundreds of these rooftop chillers and cooling towers. So that's one of the things it needs to be thought about, and there'll be design pieces around that I saw today that data grid is starting early works on a six meter wall that blocks out noise, and that might be okay for one of these big centers way out on the countryside, but that example in Norway was an urban data center, so it's right in the middle of a city in Oslo, and I mean, I'm not sure how it's designed, but it is a complaint that you've got this 24/7 hum of this data center in the city. So that's noise, then there's the water. If you're going to use a evaporative cooling system, these things consume millions of letters of water daily, then the alternative might be this closed loop liquid system, but these require quite stringent resource consent conditions that govern sort of chemical treatment and the additives that are used and the runoff of the thermal waste water discharge. So water is a big, big thing, and that's obviously used to call the data center, and we haven't talked about it, but one of the reasons why New Zealand's attractive, maybe the further south you go, the better, is that it's freezing. A lot of us are especially at the start of the year, I mean, labour shortages is something that we'll probably experience here in New Zealand, building something highly experienced, certified electricians, MEP, commissioning agents, and all those sort of technicians that will need to come across from Australia and beyond, then you've got the grid connection. These are big projects as well, and the constraint will be, we'll can this part of the grid actually connect a 250 megawatt data center, it looks like there's some real positive steps already on the data grid project for grid connection. Couple of other ones, I mean, natural disaster resilience, we have earthquakes, floods, volcanoes, New Zealand, understanding where we build, it's going to be really important to keep these things resilient from natural disasters. I mean, I suppose there are a number of challenges, likely that these data center facilities will face, and a number of these things bite, I think, at that resource consent stage. But I mean, Hamish, as you mentioned, data grid are through that, all of these factors have been considered in relation to that project, and so clearly they are manageable. All right, well, thanks, Hamish, look, that's a really interesting discussion. There is a lot of interest at the moment, internationally, in data centers, huge amounts of capital around the world, pouring into these projects, very transformative, I mean, stepping back for a moment, I think these are critical for the development of AI, the training of these models, and also the compute required to run these models. And AI is an extremely exciting area, right? You can see a lot of value for society in the development and operation of these models. Of course, a range of social license issues, as you've mentioned, that will need to be managed quite carefully, and you mentioned at the top, you know, the green suggesting this moratorium, before we've even built hyperscale data center, so it'll be interesting to see whether that becomes a political issue in the coming election. But very exciting to see a couple of hyperscale projects in New Zealand, they are big infrastructure projects, and we do like seeing big infrastructure being built in New Zealand. So I suppose that is probably it from us, thanks very much for that, Hamish. Thanks very much Sam, good wrap up there, good sign off, for more insights from the special conditions, you can find us on Apple or Spotify or ChapmanTrip.com.