Sound Policy podcast: Building a better data center

Brian Amaral: First of all, can you introduce yourself and talk a little bit about what you do at FM?

Paul Heisel: Sure. I'm Paul Heisel. I'm the principal underwriter for Intellium and Digital and Semiconductors. So I'm a principal underwriter, which is in our underwriting department. So we track all these, I'll say, these specialty industry portfolios. And I've been with FM for coming up on 36 years, primarily in underwriting roles.

Narration: You mentioned FM Intellium. Can you tell us more about it? 

Paul Heisel: Yeah. Intellium combines all the different aspects of data centers, power generation and construction, led by a dedicated team of subject matter experts. And this is to help these clients and prospects because the space is growing so rapidly, where we have all this AI growth and cloud computing growth and power generation needs. We're here to guide all that and to help our clients and prospects navigate the intricacies of data center exposures, but also, power generation exposures.

Brian Amaral: I understand there's also an on demand resource unit, right? People can call up if they've got a problem anywhere in the world.

Paul Heisel: Yeah. The team, the strike force as, as we're called, or strike team, we're there to help people around the world navigate these issues that they may come up with. And a lot of it is because of the speed that we're moving at. So they may get a prospect in the door and where we normally would take six months to develop that prospect, do all these visits and everything where you might, they might have two weeks or less.

And so we use our brains and use our knowledge to help them navigate what they need to do from an engineering side and underwriting side.

Narration: And Adrian, what work do you do at FM?

Adrian Oxley: Yeah, so I'm the principal engineer for, semiconductor and digital. My primary role is to ensure that FM’s field engineers, account engineers, and anyone really involved with the occupancy is trained. Make sure that the resources are there that meet the needs of the industry and meet FM.

So we're talking about the operating standards, the operating requirements, but also primary role as well is to ensure that we're looking at the hazards of the future and not just the hazards of now. So really trying to track, what the industry's doing and trying to keep up to date with that.

And then if possible, try and make sure that we're ahead of the game, on that. So the advice that we have to our engineers and to our clients is relevant to the current industry.

Brian Amaral: Given how fast things are moving, is that pretty challenging these days?

Adrian Oxley: Yeah, it's a significant challenge. I mean we all understand the speed that the industry is changing, the speed that the industry is investing, and the speed that the industry is, building. So yeah, so it's a huge challenge. Every day there's a new article about a new product that's being used or a new technology that's coming out there.

The way that we're building data centers is changing. A data center five years ago, it doesn't look like anything like a data center does now. So yeah, it's a particular challenge within this industry.

Brian Amaral: What are some of those differences? What does it look like if you walked into a data center right now versus how it would've looked five years ago?

Adrian Oxley: Really, I'd say that one of the primary differences is the density of equipment within the data centers. Everything's been packed in. The biggest change that we're seeing at the moment really is the way that the data centers are cooled. Historically, it's always been air cooling and data centers, so effectively blowing cold air across the servers, which then, gets exhausted back out or recirculated back round once it's been cooled. What we're starting to see now, is more liquid cooling. So direct to chip, as servers become more power hungry and more dense, the more cooling that's required. 

Unfortunately air's not capable of doing that anymore. So we're moving more to a liquid cooled model. and that's a significant change to the infrastructure within the data center. I think the other thing as well that's significantly changing is just the sheer size and complexity. As data centers grow in size and power demand, the complexity of the infrastructure around the data center has, has increased substantially. So, we used to be going into data centers that maybe had 10 megawatts of power. Those are now in the hundreds of megawatts of power. So that's, I’d say, is probably the most significant change that we've seen.

Brian Amaral: Direct to chip cooling. What does that mean?

Adrian Oxley: Direct to chip cooling really, where you're taking the cooling medium and putting it directly onto the GPU. So effectively it takes away the heat, a lot more efficiently than blowing air across it. You know, so we're using a liquid medium basically to remove that heat from the chip.

Brian Amaral: Given how quickly these are growing, how quickly they're being built, the technology is changing, does that introduce new risks at all that maybe weren't there before this boom, 5, 10 years ago?

Paul Heisel: It does, and that's one of our challenges is to get in sync with the industry by engaging the larger companies in the world and saying, we need to know what you're doing, what you're putting in these data centers so we can help, at least understand the risks.

Brian Amaral: Can you walk me through the timeline of this data center boom? you've been involved in them for quite a bit. The company has been involved in data centers for quite a bit, but what changed in the last few years that it is in the newspapers every day, that it's such a big topic, that they're just sort of headline news now?

Adrian Oxley: The change really has come from, the application of the technology. I mean, if we look at it, AI has to be one of the biggest drivers. It's a point in global change that's not been seen really since the Industrial Revolution. I know that sounds a bit probably over the top to say, but all of a sudden we're on a new industrial revolution. I mean, we're able to start to monetize AI and what that means for us all. And really it basically means that we're able to improve productivity by using the tools that AI brings.

So I think that's really been the biggest driver for this jump that we've seen over the last few years. We all thought that the change will come when we talked about, sort of self-driving cars and that side of things, although that's going to be sort of like a significant change moving forward. I think it's the way that, the way that the data's presented to you and the way that that can enhance the way we do our jobs, I think that's going to make the biggest difference moving forward. 

And I think that's where this jump is—that everyone has to be in a position because they know that the implications of not being in the position to utilize AI to its full extent is it's going to put you behind. So whether you're an industry, whether you're a company, whether you're a country, whether you're an individual, we're all going to be relying on that technology in the future. And we've got to put the infrastructure in place to ensure, that everyone's able to do that or else you will be behind. Simple as.

Paul Heisel: I think you can consider it to be like almost critical infrastructure. You know, every think of everything that we do, can trace it back to a data center. 

Brian Amaral: Including podcast interviews, right?

Paul Heisel: Yeah. Podcast interviews – You go to a hospital, all the records are in some database somewhere. You go to the airport, that the air traffic controllers are using some sort of data centers, and you turn on your TV to watch Netflix. Well, that's using a data center as well. So it's been very, very integrated into our lives. And to get to a point where I'm not sure we really even think about it.

Brian Amaral: So you're an underwriter. Walk us through what a data center that's a good risk looks like.

Paul Heisel: Well, data centers, on the surface, are pretty good risks anyway. You know, think of their business model. They're selling uptime. So they want to make sure that they have redundancies. And that fits right into what FM preaches to our other clients and including data centers, is that having that resiliency, that redundancy. So what makes a good data center is going to be you have good construction, that nothing's going to burn. And then we know the occupancy. You look at a data center and we know what's inside, as opposed to looking at a warehouse. You look at a warehouse, you have no idea what's inside. It could be the most hazardous stuff in the world, but you don't know.

But when you’re a data center and you go, there's servers in there, we know it. So we know the occupancy and have good construction, because they don't want to have anything that burns. And they're generally protected because you want to make sure that these expensive assets don't burn.

So there's different types of protection as well, and you want to make sure that there's no external exposures or things that might, introduce downtime. And that's why site selection becomes really critical as well for a good data center is, you want to put it in a place where there's no or few natural hazards.

Brian Amaral: Can you talk about some of the research that FM is doing right now around data centers? I understand there's quite a few projects in flight right now.

Adrian Oxley: Yeah, I think the most exciting and the biggest project I think we're working on at the moment is lithium ion batteries. It's probably the biggest single hazard that's emerging within the data center industry at the moment. So not only looking at the infrastructure type.

Lithium ion batteries in the battery energy storage systems that are potentially used outside the data centers, but looking at UPS systems, which are needed for all data centers moving forward. Then also looking at what we call the battery backup units, which are located actually within the racks, within the data center itself.

So all of a sudden we're introducing a hazard into the data center that wasn't there before with lithium ion batteries. I think everyone's seen the videos of ev type vehicles going on fire. And then also we've seen the impact of even the smallest event with lithium-ion batteries, with, scooters , et cetera.

So we know the hazards and effectively lithium-ion holds a great deal of energy that if not managed properly, can result in a fairly significant event. What we're looking at on the lithium-ion space is trying to develop an FM Approval standard, where we have effectively a fire safe type lithium ion environment. So looking at how we prevent thermal runaway spreading between the modules within a UPS system and preventing the fire spread between BBUs within the data center. Again, also associated with that is looking at how we can limit the possibility of explosion, due to thermal runaway and looking at all the different technologies that are out there to come together to mitigate that risk. So that's a really big part of our research at the moment.

Another area which I think is necessary, is around non-thermal damage. So should that event happen, what happens to those servers? We all know that it's fairly sensitive equipment, but what impact does smoke give? What impact does high humidity give, should there be a fire and how much damage has actually happens to the server equipment within the data center, should that happen?

And then again, looking at construction materials, data centers are changing the way that they are constructed and looking at the materials that are used to make sure that those materials are fire safe. And again, trying to work towards this idea that if we limit the amount of combustibles within the data center, that limits the size of the event, and therefore it limits the exposure to both our clients and ourselves, within the data center.

Brian Amaral: What do you think the main bottleneck is for the data center industry? Is it power generation? Is it water scarcity? Is it even insurance capacity availability?

Adrian Oxley: I think it's all of them.

Supply chain, it's going to be a huge issue moving forward. Is that, with the scale of the investment and the scale of the growth, can the supply chain keep up with it? Can the amount of generators, can the amount of transformers, can the amount of electrical switchgear keep up?

You know, power's going to be a real challenge moving forward. We're, as I said to you earlier, we're moving away from, small data centers to campuses that are, that are absolutely huge, which have, have got the power needs of a small city.

Paul Heisel: I think the biggest constraint is power and where you're going to get your power. That's why you see a lot of expansion into, I'll say some head-scratching places. Like you know, people are putting data centers in Wisconsin. In Iowa, Wyoming. Who would've thought Wyoming, and New Mexico? So these areas that are, don't have the huge populations, but they have the power, and the land. So there's power and land and you see a lot of growth here in Texas where I am. Because of the availability of land, but also the availability of power, I'll say your conventional power. But also we have, there's many renewable power sources here in Texas. So I think power is the biggest challenge that they have, and we're seeing clients that are building their own power stations at their locations, which they have to do, if they can't get a grid connection fast enough.

Brian Amaral: How do you look at that as an underwriter? All of a sudden, they're not just plugging in, but they're actually building power plants at their site. How does that affect how you underwrite that?

Paul Heisel: Yeah, it's certainly a different risk because you have the– they're mainly combustion turbines– so you're going to have different risks with that.

And the good thing is they're building a power station, but they're thinking like a data center. So if they need four gas turbines, they build eight. So they have definite redundancy in that, which is great. So they can have a schedule of running them and maintaining them, which is great.

But it certainly changes the underwriting. We have to look at it because it's big pieces of equipment that need specific deductibles and we need to understand. You know what they're doing with that power. If they're planning to sell it to the grid or just use it for themselves, or what's their plan?

We need to understand all that and we need to understand who's operating it. Are you hiring a third party to operate it? Which we would encourage. Or is it the local utilities going to provide people to operate it? So there's many different questions that we would, as underwriters, that we need to ask to understand, to make sure that we need to understand the exposures, but we need to make sure that the clients understand the exposures. because they're in the business of data, not in the business of power. So that's where we can help them. Because we can say, look, we know this equipment. We know the technology. We can help you keep these things running. 

Brian Amaral: What are some of the big loss drivers right now for data centers?

Adrian Oxley: It's a difficult one because ultimately the industry is very resilient. It's always been very resilient. However, the new hazards associated with lithium-ion batteries, we've all seen in the news incidents associated with those, in various data centers throughout the world.

So, trying to mitigate the lithium-ion risk. Things like power outages and, and cooling outages are going to be something that, potentially in the future could drive significant loss in the industry. But ultimately, it's managing the whole ecosystem around resilience. It's the way people work in data centers, the maintenance that you carry out within the data centers. It's the human element programs that data centers have looking at ensuring that just the basics, are carried out. 

So I think everything together, we've got to look at and take a very holistic view, going forward. But I think some of the challenges are going to come around the size and complexity of the systems that we're now using. I think everyone used to look at a data center as a sort of fairly benign concrete box, but data centers now are very, very complex industrial machines. And, you know, that complexity drives risk and as soon as it drives risk, we've got to mitigate that.

Brian Amaral: Yeah. What do data center clients and prospects most need right now?

Paul Heisel: Well, from the underwriting side, I would say it's capacity.

They have all this growth and which is their total insured value, their property damage values, their time element values, it's all growing and they need this capacity to provide adequate insurance. And that's on the underwriting side. But the other, the other, I think there's two other parts to this. Like one is they need responsiveness. They move fast. 

So we have that responsiveness where internally we can respond to people around the world within a couple days to help them solve their problems, or at least give them some advice or direction on what they need to be doing. And the third part is I think the prospects and clients, they need engineering expertise. It's just to build in that additional resilience. Data centers are already resilient, but I think we can add, even though it might be that an extra 5%, but that gives them an edge  So that's, I think, what the clients and prospects, that's what they need right now.

Brian Amaral: I understand that FM recently developed a new construction policy. Can you explain what that's all about?

Paul Heisel: We have a new construction policy for data centers. It's a master builder's risk, and what it allows us to do is provide certainty for our clients that are building data centers where we can outline everything that's going to be happening in a year for construction for a client, and basically list out and pre-underwrite look at these projects ahead of time. So the client isn't scrambling at the last moment trying to find insurance here or there. There's some certainty behind it when they placed the builder's risk with FM.

Brian Amaral: What are the big risks to a data center when it is being constructed that they need to protect against? 

Paul Heisel: Well, would say a lot of going to be when you get to substantial completion. It just depends on what type of combustibles might be around there. And our loss history has shown that during construction you do have some vulnerabilities.

If you do have combustible insulation or something, that might cause a fire. Also during testing and commissioning, you could have some electrical failures of that type of equipment when they're getting it ready to be for use by the tenants.

Brian Amaral: Adrian, I understand you're on the road a lot visiting these, these sites. You must get a lot of frequent flyer miles in. What does a resilient data center look like for folks out there who might be thinking about investing in or building a new data center? What's Adrian Oxley think about what a really good one looks like?

Adrian Oxley:  Yes. I mean, the answer to that is yes, I'm on the road a lot and yes, I do have a lot of air miles. It's getting the time to use them, unfortunately that's the challenge. 

Yeah, if I looked at the, I suppose rose colored spectacles view of the data center of the future, the data center of the future would be, a non-combustible box, it would be of the size that limits the overall exposures to both the client and to us as an insurance company. It would have, very resilient systems built into it. So, looking at a philosophy of n plus one, n plus two, on key systems to ensure that should one, be down, then there's another one there to take over.

And then also baking into that, the ability to maintain the systems while keeping the resilience going. If I was to look at the actual white space, where the server was in, what would I, what would I be looking for? I'd be looking at a very limited amount of combustibles, limited amount of plastics, within them. So any incident that we did have within the data center wouldn't propagate and the fire wouldn't propagate. We'd be looking at making sure that the lithium ion battery hazard is well controlled. So making sure that we understand the risks and challenges of putting them in the locations that they need to be put in.

It’d be resilient from climate perspective. We would make sure that it's out of a flood zone, we'd make sure it was built to the relevant wind speeds, hail performance, of the area that it's in. So I think, looking at it, there's a lot of critical factors there all coming together to give a positive view of what good looks like

I think being cognizant of the risks and challenges going forward is absolutely critical for the industry as you know, it rushes to build the infrastructure of the future.

Brian Amaral: Great. Thanks for being on the podcast.

Paul Heisel: Oh, thanks for having me. I really appreciate it.

Adrian Oxley: Thank you.