Catalyst with Shayle Kann - Frontier Forum: Future-proofing data center power infrastructure

Episode Date: April 7, 2025

Data centers face a critical timing problem. They need massive amounts of power immediately, but grid upgrades can take more than seven years to complete. Microgrids could be a solution to this growing power gap. "How do you future-proof your systems through adaptability of technologies? Microgrids are really a nice technology to address that," explains Adib Nasle, CEO of Xendee. In this episode, recorded as part of a live Frontier Forum, Latitude Media's Stephen Lacey speaks with Xendee co-founders Adib Nasle and Michael Stadler about how microgrids can help data centers get power quickly, while also preparing for future technology changes. The conversation explores how the combination of combined heat and power systems with supplemental technologies like batteries and renewables can create flexible, adaptable power solutions for mission-critical facilities. Their research across different energy markets shows striking cost reductions. "We saw cost reductions up to 80 percent in California and 60 percent in Virginia," notes Dr. Michael Stadler, CTO of Xendee. "If you really go with the utility, it's always the most expensive case." Using a stepped approach to microgrid implementation, data centers can achieve electricity prices significantly below grid rates — while maintaining complete control over their power infrastructure and preparing for future technologies like small modular reactors. "One way to think about microgrids is really as a Swiss army knife of energy infrastructure," says Nasle. "You're getting localized resilience, economics, flexibility, scalability, efficiency through controls, and the inherent benefits of a decentralized system." This is a partner episode, brought to you by Xendee. It was recorded live as part of Latitude Media's Frontier Forum series. Watch the full video to hear more details about microgrid design for data centers.

This is a Frontier Forum, brought to you by Latitude Studios. Adib Nassely knows mission-critical energy systems. Early in his career, he worked with power engineering teams on aircraft carriers and warships. He also designed a predictive energy management system for the Federal Aviation Administration. And this included, like, all of the air route traffic control centers, terminal radar approach control facilities, and many of the air traffic control towers. More recently, Adib is focused on another type of critical facility. data centers, and he found himself talking to many of the same people he worked with in aviation

in the military. I actually found data center facility managers as I visited some of them. They were former nuclear submarine officers. That's because the warehouse-scale computers that run our economy require extraordinary levels of reliability. You can't rely on a single energy source when you're thinking about mission-critical systems. And you also need intelligent energy management and balancing. And that background really mirrors much of what data center facility owner operators are planning, designing towards.

Around the same time that Adib was designing energy systems for U.S. aviation, Michael Stadler was running microgrid R&D for the Lawrence Berkeley National Lab, modeling different microgrid configurations from combined heat and power to renewables. We started microgrids as CHB microgrids. I mean, this is how we started. Of course, most recently, we focused more on renewables. But it seems like with the data centers, we are going a little bit back again to the roots of microgrids, which is around CHP systems. Today, Adib and Michael are co-founders of a company called Zendi.

It's a software platform for the design and operation of EV infrastructure and microgrids. Adib is CEO and Michael is CTO. And they've turned their attention to data centers, which are growing rapidly, but are increasingly constrained by the power grid. We are thinking that microgrids are really good solution for this, right? Because microgrids bring together on-site generations. other forms of flexibility so that you can respond in a better way to high-priced signals or even avoid having a bottleneck in the system. Data centers need massive amounts of power right now to feed the growth of AI,

but it can take more than seven years to upgrade infrastructure and clear interconnection cues, and that creates a critical timing gap. So there's this immediate gap, and then how do you future-proof your systems through adaptability of technology? And microgrids are really a nice technology to address that. I'm Stephen Lacey. And this week we're looking at how to future-proof data center power infrastructure. I recently sat down with Adib Nossley and Michael Stadler for a live conversation.

It was part of Latitude Media's Frontier Forum series. And we talked about how microgrids can help data centers bridge the growing power gap on the U.S. grid. The reality is doing nothing and waiting is probably the worst solution that you can do. and we should start fighting this problem now. What are we talking about when we're talking about microgrids? What are the configurations of microgrids that you're focused on? So one of the really well-suited capabilities of a microgrid is the flexibility it brings into the combination of technologies that could be accepted as part of an on-site cogeneration or tri-generation system. You need that firm power.

So you need some sort of typically a gas-fueled prime power source. These can be gas turbines. And that provides a level of grid independence, steady, reliable power. And additionally, it opens up this opportunity for combined heat and power where the exhaust heat can be also leveraged as an additional capability and efficiency that further strengthens the business case. And then you need supplemental technologies. And when I talk about supplemental technologies, these are your lithium ion battery, solar PV, if you have the space, thermal storage. And these ability for integrating new technologies and new capabilities as they introduced himself or as the costs become attractive.

And this gives you, as a facility owner-operator, a level of flexibility that's simply relying on a single source of power, which would be the grid, inherently does not offer. So the combinations out there that typically we're running into our CHP systems, just because the thermal load is just as large as the electrical load within data centers, but also the need for this grid independence and this reliable prime power that you need to have. So those are the combination. And then thinking around some of the future technologies that are coming out and how can the system be designed in a way that these can plug in a very flexible, adaptable way so that as the

load grows and the load will grow. AI processing is just fundamentally a different workload than the traditional data center workload that we've seen to be able to have that ability to adapt without everything breaking apart. Yeah, I mean, you could say, well, I believe that maybe we get this discrete situation under control or we're waiting for new technology in five, ten years, maybe a small, module reactors. And you might think that installing just a small gas generator, diesel,

generator might be the solution. But it's actually a kind of stranded investment. Because, I mean, let's say the utility comes online and gives you the power three or four years, you have now this investment sitting there, which you're not using anymore. So this is why you need to start planning along a timeline. And if you install something right away, that technology should also help you five, ten years down the road. And this is why CHB generation may be in the future with small modular reactors is a good combination because they can augment each other, right? And that's the key functionality within a migrate.

You're plug in these technologies and they can help each other. You're not just thinking about one technology to solve a problem for the next two years. That's the key factor. So going back to the research, you studied two specific regions, California and Virginia,

very different energy landscapes. What were the common findings across these different markets and the differences in how this stepped approach as you're calling it could be implemented. So I think you're referring here to a white paper, which we did to specifically demonstrate the benefits of this approach.

And we picked these different regions for a reason because they have very different electricity rates, right? So California, very high-priced electricity, I mean, it can go up to, I mean, for data centers, 15, 16 cents per kilowatt hour, but the demand charges, the power costs are really high. I mean, we saw numbers around $40 per kilowatt electricity. that they're consuming. So very high priced. On the other hand, in Virginia, we have lower electricity rates than in California. And we were interested to see really if this approach and also

their technologies will be similar. And the reality is, the results are very similar because CHB systems and all these micro-technologies are so attractive for cost reduction that the price is not the utility price is not the most driving force here. And we have, have seen cost reductions up to 80% in California and 60% in Virginia, while the electricity rate was really almost half then in California. So the common finding is really, as alluded already, if you really go with the utility, it's always the most expensive case. If you wait that the utility is going to save you, and if the utility can provide the

electricity, in both cases, we ended up with the highest electricity rates for the next are 20 years that we analyzed. So we also had escalation factors on electricity rates and natural gas prices all building in this model, right? And then same outcome in terms of technologies and carbon savings are also very similar. So at the end of the day, you want to start doing something now, installing either a CHB system or a combination of renewables and preparing yourself maybe in five or ten years down the road to augment the whole system with small modular reactors.

and then these technologies will nicely work together and give you actually the cheapest solution. I mean, we saw electricity prices for data centers going down to three and a half to four cents in a microgrid if you use this approach that you start now with implementing something on the natural gas side or renewable side and then augmenting this. Adib, let's have you riff on that a little bit.

In thinking about that 10-year time horizon, you outlined the mix of potential on-site generation technologies like in the near term, but how does that change when you think about the five to 10-year time horizon? Well, I mean, we've seen a considerable amount of innovation in the energy space. I mean, SMRs are just one example of that. Battery technology continues to improve generator technologies continue to improve CHP technologies continue to improve.

And you've got to have adaptability. these data centers with the AI workload that demand is going to continuously increase. And then you also have these market forces with new technologies coming online. So this approach we're discussing that Michael and his team worked on is really to bring in a level of adaptability, which I personally feel is a key factor in futureproofing against these technology shifts and market factors, while at the same time delivering on that business case. So it's a really neat set of two sides of the same coin where you want adaptability and at the same time you want to bring confidence

to the return on these assets for investors and really hard in the business case. When Michael was mentioning that cost being down to around three, four cents, that's something that's now under the data center owner's control, which is very different when you're signing contracts with outside suppliers where, you know, rates will change and input costs will change. But now you're sort of controlling that real emission critical commodity. And then you're taking advantage of the interdependency between generating that electricity

and the heat for the thermal load as well. So I feel that when we're looking also five, ten years and you have these new technologies coming in, being able to really leverage them in a way, understand the economics of it and also the operational opportunities from it. Because OPEX is kind of the name of the game in many ways around these things. And if you can manage and control the OPEX and bring a lot of confidence into the outcomes, then that's a pretty powerful and compelling strategy. I think one thing I want to add here is when we talk about this three or four cents

electricity, it's really not only the OPEX costs that we have here. it's also including the capital cost. So the point is we really get down to three, four cents, including all the capital that we installed there, and we model actually escalation on the electricity side by 10% every year, and natural gas also going up by 5% every year. But still, at the end of the analysis in 20 years, we are still at these low levels because, as Adip says,

we can control the cost through these technologies and we can hedge ourselves against this high electricity prices, demand charges and costs for upgrading the system. So just want to point out this three to four cents is not only because we're considering the operational cost. This includes also capital costs, and that's really an amazing number if you think about it. One way to think about microgrids is really as a Swiss Army knife of energy infrastructure. I mean, you're getting that localized resilience.

You're getting the economics that Michael just talked about. You're getting the flexibility that we just talked about with regard to sources and future technologies. You're getting scalability, which is kind of hard for. for a large infrastructure, you know, because you were talking about how long it takes for upgrades on the grid. Well, how quickly can they scale if just getting to point from 1 to 1.2 takes this many years. What is it going to take to go from 1.2 to 1.4? So scalability is real important as well, which microgrids provide, along with the efficiency through controls,

that normally you don't have with these large centralized bulk power systems. And then just the inherent benefits of a decentralized system. especially when it comes from a cybersecurity perspective. I'm going to start drawing in some questions here, and we have a question about whether these microgrids are going to be a bridge to eventual grid connection, or do you think that data center providers will be comfortable with being fully off-grid for the long term?

How are you envisioning the evolution of these microgrids? I think it will depend on the use case, on the region, the personal preference, right? I mean, I personally feel if you do all these investments and you have all these technologies installed, then why actually bothering about the grid connection? The analysis that we did for these two cases actually suggested you can run 100% grid independent, right? Then it just becomes a redundancy challenge for the generators. But depending on the cost structure, you might want to go with the utility to some extent because maybe they already have some of the power available,

but in some areas maybe you just go completely islanded. But that's something that you would get out from this multi-year stepwise approach to figure that out. Yeah, and it helps avoid stranded costs not only from the utility side. Because imagine you build out a microgrid and you have this level of energy independence and resilience. And then the grid upgrades. but those grid upgrades are now going to be underutilized.

So now you have that potential risk of that stranded asset. So this ability to really have this stepped approach, make sure that the assets that are being deployed on site from a cogeneration perspective can play a role if there's a decision to grid tie. Or as the grid upgrades occur, what does that look like? And what does the cost benefit analysis in that? and what are the enhancements that could be potentially achieved? And then obviously there's always we just run grid independent and what does that look like as well. So the flexibility in the approach and the flexibility within the framework of the microgrid help solve and answer a lot of these questions through a science-based method.

What are we talking about for optimal size of data center for this approach? There are a wide range of sizes from the 100 to 200 megawatt up to the gigawatt scale, and there can be relatively few gigawatt scale data centers. But certainly a lot of plans to build multi-hundred megawatt campuses. What is the optimal size of data center that you can serve with this approach? Yeah, I mean, this is where the scalability of microgrids comes into play, because you're not locked into a rigid framework. So if it's a small or medium,

load today, it can become a very large supply source as well for a future. So that adaptability includes the scalability because you can literally plug in and increase your generation and cogeneration capabilities in a very efficient, reliable way. So this sizing question that you have is exactly why the stepped approach is one that we are looking at and we've been investigating. because microgrids inherently bring with them the flexibility for a stepped scale where you're doing it not only based on what the load needs are, but also you want to do it in a way that enhances the resiliency, but also the financial returns and the return on assets and the market factors that are going to occur. Because it's not like it's going to happen all of a sudden within six months you need twice to load. but over the course of that project's lifetime, there's going to be inherently technologies and

load growth that you need to adapt to. And bulk power systems don't scale well, which is why you're talking about these seven to 15-year timelines. Microgrids scale very well. So they can really help that path occur in a very reliable, efficient. and I would say a financially sound way that minimizes risks both from execution risk technically and also from the business case on what the return on the assets need to look like. Yeah, bottom line is there's no real optimal size or typical size because, I mean,

microids, you can have migrants from, I mean, you can have residential microids which go a couple of 20, 30 kilowatts to hundreds. then you have commercial microgrids, larger bonds in the megad range. I mean, and this is actually the reason why this analysis is so important, because every microgrid, every load, depending on the structure of the load, depending on the electricity rates, depending on the technologies that are available in this area, depending on other factors, will deliver different optimal solutions. And you could have a microgrid, which is completely renewable-based,

if you have the space. and the money for this, or you could have a completely natural gas fired micro-wit, which maybe gets augmented by small modular reactors laid on if you have the appetite for this. So that's exactly why we are saying you need to do a careful analysis and use a guidance system or approach like we are doing, which guides you through time and all these upgrades that you maybe probably need to achieve really the best outcome for here. Michael, you touched on the other.

ability to build renewables only microgrids. And we have a question in here about how realistic it is to think about 100% renewables in a microgrid or renewables plus storage thoughts on optimal mix. We have seen the last couple of years, we have seen that, of course, in certain areas in the country, people are very focused on renewable-based micro-wits, right? And you can achieve renewable-based microgrids if you have the space and the money for this, right? I mean, renewables play an important role. But if you want to be 100% focused on renewable and also when you have outages or problems with the utility, then you need also expensive technologies like batteries and all that stuff.

And this can really drive up to costs. And the reality is this, if you're focusing on costs and carbon emissions, you will always see a combination between renewable technologies and some gas-fart engines for backup reasons or for resiliency reasons, right? It's a pretty bad idea from an economic perspective to have a very large battery PV system renewables to carry you through an outage event because this really adds up. And every analysis that we did was like this, I mean, the whole country. And also the AI workload is fundamentally a different workload than maybe the traditional data centers that have been more around communication speed. So the application of some of the renewable technologies for the AI workload could be a challenge.

Yeah, that actually brings me to another question which we got from a listener who asked about load swings and data centers. I mean, these are unique facilities where you can see massive load changes in seconds. So how do you optimize generation batteries to handle that variation in load? Yeah, I mean, it's absolutely true, especially now with the AI. I mean, we can go to 100% very quickly and then drop down to 25% when we wait for the next job. and I mean, naturally, the data cent industry is very used to backup power, which can handle this. And generators and batteries can also handle this if we make sure that we have to write chemistry and technology around it.

So you need a fast-moving technology, and this can be the battery or the generator. We will see if the small module reactors can do this, because historically, a reactors were only for base load. Now what we're seeing also with the research work that we're doing with the University of Illinois at Champaign is that this new reactors can be bad in load falling. If they will be able to really follow this crazy spikes that you're mentioning, we will see.

But for now, I think it's all on battery technologies and generators. Michael, can you talk a little bit more about data integration and scenario modeling and how Zendee's software, where can simplify that complex planning process? Yeah, I realized we're talking a lot about this approach, but we haven't really explained what we're really doing here. So, I mean, as we started doing all this, we realized, I mean, it's a very fragmented process

to actually figure out how big a technology should be, a battery, a generator should be to give you a certain saving cost, carbon, whatever is, right? As I did this the first time, I mean, more than 10 years ago, it was half a million dollar five engineers working a full year to figure out how big a battery a PV system should be for demand church management, all this stuff, right? And that's not sustainable. And this is why we decided we need to standardize this approach in a sense that the data

collection comes through databases and through API so that the user doesn't have to actually figure out what is the solaration at a certain location. What is the wind speed so that I know if there's a potential for these renewables, right? databases, what are the technology customer? Why is it that the user should basically go through all this vendor documents and figure out what this stuff costs, right? So you need to automate all this

and bring it into a platform which then allows you to define an objective. What's your objective? I want to do 100% if this is possible 100% resilient microgrid at a data center at minimum cost, right? That's my objective. But then my

technologies could be gas turbines CHB, maybe a small model reactor in the future, renewables, wind, PV, I could do interaction with the utility. So all these options need to be analyzed. And there was no real approach for this. I mean, what is the value stream of all these things? And this is why we set out in 2018 to actually standardize this math approach, which goes back to my research time as I was in Berkeley, partially at least, and standardize this in a way that a person without an academic background can actually do this, right? So at the end of the day, now we are down to actually analyzing a microgrid in terms of what technologies should be in a microgrid within a couple of days if you have the data.

Because if you're serious about doing this on a large scale, we don't have enough people actually doing this or not enough companies. And this is why it needs to be standard. in a sense that it's easy to do. It can be replicated. So you build your base case, you build your scenarios. And if something changes in your assumption, you just change the assumption, you rerun it and compare it to your baseline. And this all needs to be automated to some extent with automated data integration so

that this is actually less expensive and can be done in a shorter period of time, because that's what we need here. And, Steve, in addition to that speed, there's a level of sophistication that we felt the market needed as well. So it's not just sizing the assets, but it's also where should they be installed? And then more importantly, what is the physics of the system? You know, power flows, voltage, voltage drops, transformer loading, cable loading, ampacities, impedances.

All these are important from a balance of system perspective that are critical drivers, both from a efficiency perspective, but also from operations and safety perspective as well. because you want to make sure that onside energy system is able to get the energy from where it's being generated to where it's being consumed in a safe and reliable manner. And that means you've got to understand the power flows. You've got to understand the loading on the cables, the transformers, and what you can avoid as an upgrade or how you can avoid it by placing technologies in certain locations. And an algorithmic science-based approach where it generates these optimal solutions from this enormous problem space. I mean, Michael kind of touched on it. all these different things that have to come in.

And just a burden on a user or a planner or a strategist to try to bring that all together and understand it is way beyond the realm of a spreadsheet or something that you can potentially cobble together. So we felt that there was a need if we are truly going to be looking at something that needs to happen at a large number of sites at scale, while at the same time bringing the in and aligning the interest of the investors, the owner-operators, and the end users, that it needed a science-based streamlined approach. As we round the conversation out, Adi, let's put you in the shoes of a technology leader,

a CIO at a major tech company today. What would be some of your top recommendations for how they'd approach power infrastructure planning in the next decade? I would encourage the leaders. to implement a energy system plan that hardens a business case, brings confidence for investors into their return on assets, and adapts to the technological shifts and market factors without breaking down. And just as adaptability is a factor in future-proofing technology, organizations, even ourselves, right?

From a personal skills perspective, it's important for critical energy infrastructure. And microgrids are a keystone piece, I feel personally. when it comes to adaptability and energy systems, especially for future-proofing. As I mentioned earlier, they're kind of like the Swiss Army knife of power infrastructure, and the level of adaptability and flexibility

that is gained through the implementation of microgrids. It's hard to beat. Michael, any final thoughts on what your recommendations would be? I mean, I think it's less technical. It's more around the uncertainty that we are facing, and this big back and forth between, well, should we do decentralized or micro-rid? I mean, at the end of the day, it's a given that micro is really providing high efficiency,

lower costs, but at the same time, they're also creating a challenge for the utility system. And I think this is something which we haven't really tackled in a good way. And if we want to be successful on both sides, I think we have to really tackle this and figure that out because we cannot just kill the utility system because we're taking everything away because it's making it too expensive for the people on the utility system. But at the same time, we need to get out of an approach, which is always, trying to avoid migrants because they are more efficient. So I think it's more like a political or a societal discussion than a technical one.

Because from a technical perspective, I mean, there's no doubt that migrants will do a better job here. Adib Nassely, thank you so much. It was really good to chat with you. Thank you, Stephen. Appreciate it. Great to chat with you as well. And Michael Stadler, good to see you. Thanks.

Thank you. Thanks for having us. This conversation was recorded live as part of Latitude Media's Frontier Forum with Zendi. And there is so much more. This is an edited version of the conversation. So if you want to watch the full video with tons of listener questions and more details on microgrid design, head on over to latitudemedia.com slash events and just click watch recording. You'll also find a bunch of other frontier forums that we've done,

and you've got lots of good content there to comb through. Zendi helps customers model and control 25 plus technologies, enabling you to quickly screen design and operate your distributed energy project of any size. helping achieve predictable outcomes and prepare for future changes in technologies. Learn more about how Zendi is serving data centers at Zendee.com. That's X-E-N-D-E-E-Zendie.com.