Powering the data center surge

Show ID (00:03):
Welcome to Power Onward, the podcast that pulls back the current on the technologies, people, and partnerships shaping the future of power.

Kim (00:13):
We're entering a phase where you can have land, capital, and demand, but if you don't have power, you don't have a data center. And right now, that's exactly what's happening across the industry. AI is accelerating demand faster than infrastructure can keep up. Rack densities are moving from 10 to 20 kilowatts to a hundred, even 300 kilowatts per rack. That's like charging dozens of EVs at once on a single rack. In markets like Virginia, Texas, and Phoenix, interconnection timelines are stretching from months to multiple years. So while everything still feels instant on the surface, behind the scenes, the systems enabling that experience are under real pressure, and that's why the conversation is shifting. It's not just about back-up power anymore, it's about how power is designed, integrated, and scaled as a core part of a data center itself. Because today, power isn't just about supporting growth.

(01:05):
Power is defining what gets built and where. Hi, I'm Kim.

Patricia (01:09):
Hi, I'm Patricia. I lead marketing for data centers.

Dawn (01:12):
And hi, I'm Dawn. I lead strategy for power systems here at Cummins.

Kim (01:15):
So in previous episodes, we've discussed how data centers are like the backbone of the digital economy. Everything that we know, we stream, every search, and every swipe is all powered by a data center, right? But we're seeing a tremendous growth in this industry. So let's take a step back a little bit. In the last five or 10 years, what has fundamentally changed about data center power?

Patricia (01:36):
Kim, a lot has changed. I think back to five years ago, we were just coming out of the pandemic. A couple of things were happening. One was because of the state of the world. There were challenges with supply chain, but even bigger was that the pandemic or that time drove a significant increase in the number of people that were going digital. So we saw increased digitization, but we also saw a huge shift to cloud. So a lot has changed as we look back. In fact, if I think about the forecasts that were put out in 2020 for where the market was going, I think if we all looked at them today, we would be shaking our heads like we could not see where we are today coming and where we're going coming. So a lot has changed. And I think the first one is obviously scale.

(02:29):
In the last, I'll say two years, we've seen quite an advancement in AI driven by advances or breakthroughs in large language models, generative AI. So that has changed a lot. It is driving a lot of demand to where we are today and some of what you talked about in the introduction. The second one is power availability. Because that growth has accelerated so fast, it has outpaced where the grid is today. And so that is why the conversation is changing. It is changing because we have to think about how we power data centers with such tremendous growth. One of the other areas that is changing is the architecture and evolution in architecture. So to think about several things, the first one is from an IT perspective. You talked about how racks are changing. We're going from what the average was 12 kilowatts to the hundreds. In fact, next generation AI, we're looking at up to 600.

(03:33):
I'm even hearing something like a meg. Wow. So significant changes in that. And then with that means that we have to think about, because now you're going into high efficiency, high power, so now we have to think about how to cool them. So advancements in cooling technology, liquid cooling. From a power perspective, just lots of changes all around. So a lot has changed over the last five years, and we have to think about how to respond to those changes and how to keep the industry growing.

Kim (04:03):
From a work-life standpoint, it's really interesting. I remember back in 2020, going back to like, "Oh wow, we're working remote. How lovely. Oh, we can rely on collaborative sharing through the cloud, all these Zoom meetings versus being in person, not even thinking, not even close to thinking about the strain that it was having on the grid." It's really cool to understand more about the demand. It's about the speed, the scale, and also how tightly connected that is to deployment. So Dawn, I feel like I hear it all the time in the news today. There's a lot in the news about data centers and what's happening. Break it down for me. What is the real problem we're trying to solve here?

Dawn (04:39):
The real problem is we need so much more power all of a sudden than anyone had planned for. So Patricia mentioned forecast in 2024. The forecast we had in 2024 didn't anticipate anything like this rate of growth in our data center business. You talk to the electricity grid operators. It's a similar thing. I had the opportunity to meet with one of the ISOs last week, and they talked about the fact that their peak power loads, they plan the entire Midwest have been flat since 2013. No growth at all. And now as they look ahead to 2030, there's a 25% growth in their peak load by 2030 and just four years. And if you think about what that means in practical terms, it means that they expect to actually add almost 20% more power to the overall grid than anybody had anticipated needing to add. And you can't build a power plant like that.

(05:37):
So when there is a peak load, there needs to be some way to actually meet that requirement. The US grid and grids anywhere in most places in the world were not designed in anticipation of that. And the grid here in the United States is pretty old. If you look back into the history of it, it was really built up starting in like the 1930s. And of course it's matured and been invested into over time, but a little bit haphazardly and not at anything like this rate of growth. So that's the number one thing that is a very real challenge. And if I want to build a data center, I'm adding out of almost nowhere from the standpoint of the grid, a massive incremental power demand on the grid.

Kim (06:16):
So building on that, Dawn, what's changing about how customers are powering their data center sites if they can't get the electricity from the grid that they're expecting?

Dawn (06:24):
So what is changing is we are shifting from, in the past, if a data center is built, they have grid power available, electricity power available. And what we are supplying is back-up power only. So if the grid is not available, they have more than enough energy available from their backup diesel gen sets. If they don't have enough power from the grid, they need to get it some other way. And so what we're seeing is two main alternatives. One is if they're solving for, "I don't have power from the grid today, but I think I'll get it in a few years." So I'm in the queue, I'm waiting, and we call that and talk about that as bridge to grid. And sometimes you'll talk about that and you'll hear about that in the press. There's a question about how long is that bridge. But if I'm solving for, I need to power this 100 megawatt data center for three years, I'm going to probably put a lot of natural ... I'm going to see if I can get natural gas turbines.

(07:14):
This is where you're starting to see some natural gas reciprocating engines put on site and you will still, there's some integration of battery storage systems to help with load fluctuations and often also still diesel backup because they're solving for redundancy. So that's the bridge. In some cases, we've seen builds where they say, "I don't expect the grid to ever get here. I need to be independent and I basically need to build my own power plant." And a lot of times what will happen is those builds will happen with very close proximity to natural gas pipelines. So if you can look at where these are situated, you need a steady flow of natural gas. And so they're standing up natural gas power plants on site.

Patricia (07:56):
Even where we have utility available today, the utility is being flexible enough to support the industry, but where you have peak periods, in some cases, the utility will need the data centers to help meet the need of when we have peak loads in the summer, everyone's running their AC. And so one of the other, I'll call it evolutions of the conventional use case today is that the assets that used to be stranded at data centers tied down to a use case for emergency power now can be dispatched to help support when there is that gap in peak demand. And so that is something else that we are thinking about and working through with some of our data center operators.

Kim (08:44):
You mentioned iterative changes. It almost reminds me of a home built back in the 1930s. You were talking about the grid. Way back then, we talk about those haphazard changes that previous owners may have had and how difficult is it to scale in that way, especially with all the regulations that continue to change. It's like layers of complications stacked on each other on themselves as well. Yes. So Patricia, I feel like you always have your ear close to the ground. You hear what the customers and investors are saying, right? How should they think about power moving forward? What are some critical things they have to understand? Yeah,

Patricia (09:18):
A number of things. One is you have to be agile. There's just so many changes coming at the industry. So it requires continuing to have that mindset of, I need to be agile, I need to be open to non-conventional solutions, I need to be partnering. I think partnering is really, really important because we are developing solutions for now, right? We are having to think about and partnering enables that co-creation. It enables us to think through, pilot, learn, and really move so much faster because we don't have time to develop solutions for three, five years ago, which is really how the industry has been in the past. So I think partnerships is really, really important. I would say the last one is project execution. When we look at the scale, one of the things that has really, really changed is the scale of data centers. I mean, we used to get excited just seeing scale at hundreds of megawatts for data centers.

(10:24):
Now we're talking gigawatts. So when you think about the scale project execution and the ability to deploy these massive facilities just becomes so important, so real. And so we can't just think about the upfront, the CapEx and all of those things. We have to think about how to get it on the ground. And even once it's on the ground, how do we continue to support these data centers? Because uptime is still the most important thing for this industry. So I would say just it's a mindset shift, right? And so just getting in the ecosystem and working with partners and moving that piece forward.

Kim (11:02):
I think that's a great example of not just how we're a point of contact, but we are a trusted partner from design to installation and beyond. And I think that is what our customers are looking for from us to continue to be agile and stay on the leading edge so they are always in the know. That's kind of one of our differentiators, right?

Patricia (11:17):
Absolutely.

Kim (11:18):
So Dawn, I kind of want to shift this conversation beyond data centers for just a moment. What does all of this mean for the other industries and the communities that we support?

Dawn (11:26):
So in some ways, this is actually an interesting question, Kim, to think about just the dependencies here. Data centers don't operate for the most part, there are a few, but for the most part, they don't operate on an island. They're pulling power from an energy grid that we all share today. And so when they start to pull a lot more power, there are impacts across the entire system. That's showing up today in some communities and a lot higher electricity prices. And it's something that we are going to have to figure out how to solve. How do we add a lot more power to our overall energy ecosystem in a sustainable way? And when I say sustainable, I mean, how do we think about impacts on communities, both from clean air, noise, and affordability? I think here in the US, energy isn't something that we've thought about a lot.

(12:14):
Electricity pricing is something that has been pretty cheap and it's really noticeable if all of a sudden over a couple of years, your electricity bill goes up. And the reality of how the grid is constructed and how pricing mechanisms work today is that is going to happen in a lot of communities unless we make some changes. A lot of changes are being discussed and I think there will be a lot of moves in this space over the next few years, but the reality is, is these are interdependent decisions. And so any actions that data centers take have knock on effects on our communities, on other businesses, and all of us who use electricity.

Kim (12:51):
So recently I was at SEP or our Seymour engine plant, and I saw that some of our 95 liter gens were in the test cell and the power that they were generating, they mentioned they were selling it back to the grid. Is that a thing? Is that real?

Dawn (13:03):
Remember the big winter storms that swept through Midwest, East Coast back in January?

Kim (13:08):
Yes. How could I forget?

Dawn (13:10):
One of the announcements that went out from, I can't remember if it was the Secretary of Energy, was he basically made an announcement that data centers, diesel gen sets needed to be ready to run. Yep.

(13:22):
And his expectation was there was too much pressure on the grid or if they had lines down, he knows there's a lot of extra capacity installed, so run them. I don't know how many of those data centers are actually set up. You need to actually have the systems in place to actually sell that power back to the grid. You can't just flip a switch and be like, "Hey, now I'll run my gens and sell them back." So there is a lot of installed capacity, but there's a lot of retrofitting that would have to happen. And oh, by the way, those are diesel gen sets. The emissions are not insignificant. And so there's just a lot of work to be done, I think, to figure out, well, how do we feel about how often might they run and how are they going to be regulated? And a lot of them could be retrofitted with after treatment systems that could at least pull down the NOx.

(14:08):
It doesn't pull down carbon. Right, right.

Patricia (14:13):
Unless you go with VO. Just like to be able to receive or absorb power or get power from all these distributed sources is a challenge, right? And that is why in the UK, because they have a structure that is set up for this, they implemented grid codes because in the early days they started to see failures.

Dawn (14:31):
And you see lots of interruptions. So I don't know if you read about the power outage that happened across all of Spain and Portugal last May. There are aspects of just energy balancing. So like all the reports that come out is like Spain and Portugal will added ... Sorry, we're kind of off topic. They've added so many renewables and to your point on like the grid's ability to keep load shifts and balance, basically things got out of balance. Right, right. Yeah. And so this is a thing. Data centers, they risk really triggering pulling things out of balance on the overall grid, which can cause

Kim (15:02):
Blackouts. So we're operating in a dynamic, fast moving environment. Demand is accelerating and all of our decisions feel way more nuanced than ever before. And because of that, there's a lot of momentum, but also assumptions that are frequently around in the industry. So we're going to test a few of them, pressure test them, if you will. So first question, first statement, I should say, AI demand will outpace power availability for the foreseeable future.

Dawn (15:30):
Agree. For the US and a few other markets, it is a lot more complex to invest into energy infrastructure than it is to start to build out data racks with chips in them. And so that by its very nature takes more time and that disconnect is going to persist for a while.

Patricia (15:52):
Yeah. Just to add to that, when we look at how fast the data center is growing, driven by AI training and AI inference, which is the application of AI in different scenarios, when we look at all of that, we look at the amount of the gigawatt growth for data centers. I think the latest forecast that I saw that was published by JLL was talking about a hundred gigawatts by 2030. Essentially, the market is doubling between now and 23rd. So now is already huge. We're talking about doubling and the grid infrastructure is just not yet set up to increase that fast to meet that demand. The challenge is not so much generation. The challenge is transmission, interconnection. And the permitting and the process of getting all that in place is years long, right? So when you talk to some of the data center operators, they're waiting for interconnection for two, three, five years.

(16:49):
Some of them don't even have that and they're just having to think about bringing their own power. So there's very few scenarios where we see the grid catching up in the short term.

Kim (16:59):
All right. Next up, natural gas will be the next generation of data centers.

Patricia (17:04):
Depends. I'll also go with depends. And the reason I say that is because where the grid is available and sufficient, the grid will always be the best option. However, like I said, we will see opportunity or a gap in what the grid can supply to the data center market. So natural gas is readily available both from a supply perspective, but also the technology is readily available today. So it becomes a really good, what we would refer to as a bridge solution, right? A solution between now and where we're going to get to as we see infrastructure investments. And the reason why natural gas as a technology is attractive is because the technology lends itself well to long run our applications versus say diesel generator sets. So lower emissions, both carbon, both carbon content as well as till pipe emissions. And so because of that, if I think about some of the technologies, we're talking about natural gas recip generators, we're talking about turbines, we're even talking about solid oxide fuel cells, which have come into the market and are seeing higher adoption because of the need.

(18:17):
So a lot of ready technologies that are there to meet the demand and the supply base for the natural gases there as well.

Kim (18:25):
Speaking of new technologies and adoption, do you think the new tech will scale fast enough to close the gap?

Dawn (18:31):
No. No. Reliability is absolutely the critical decision factor when power is installed to support a data center. Data center customers talk about five nines of reliability, which means 99.99% available. I've even heard them lately tell me six. And if you think about what that means in terms of number of hours this could be down over the course of the year, it means almost never. New technologies won't operate at that level of reliability. You see a lot of press about SMR. So SMR is a small modular reactor. It's nuclear. There are some people who would tell you it's not that new. It's in some ways repurposed technology. It's a different scale. It depends on the context. Regardless, commercialization is a long ways off. There are a lot of regulatory requirements. Installing nuclear is not something that most communities are comfortable with. Understandably so.

Kim (19:24):
And

Dawn (19:25):
So there are just lots of other hurdles to work through. So when people talk about new technologies, usually that means more time. And I talked about interdependencies earlier. You can move fast with something new if you do it all by yourself. There's very little about this space where people can do this all by themselves.

Kim (19:44):
All right. Next up, power constraints. Do they exist beyond the US?

Patricia (19:49):
Agreed. They do. So we're seeing the constraints in the US now. And the reason it's making headlines is because of how vast or how big the problem is, but there's actually precedence. Even before we saw constraints in the US, we saw constraints in Ireland several years ago, right? With large installations of natural gas that sort of helped bridge until the grid caught up and it did. We also saw similar changes in Singapore. Singapore actually has implemented and still has a moratorium because of grid insufficiency to supply the data center industry. This is something that has existed outside of the US, right? Now it's been in the US, but I think we will continue to see it persist and maybe even see a little bit more of the gap in grid versus demand outside of the US. We are seeing places in AsiaPac, Australia, for example.

(20:44):
We have seen examples of where interconnection is taking a significant amount of time, so there is that gap. So we will see it outside of the US.

Kim (20:53):
Dawn, talk to me about time to power. Is time to power now the most important factor?

Dawn (20:58):
It depends. It is certainly top of mind for many of us today. We hear it a lot from our customers. There's a recent quote from the head of energy at Amazon about grid access is the number one factor affecting their siding plants, but it's not the only factor. So access to land, access to water, access to permitting, access to fiber. All these things matter and it just, it's dependent on what an individual data center builder's requirements are and what they're solving for. And sometimes there's power available. Often in the US, there's not.

Patricia (21:30):
Yeah. In the US, it has suddenly risen to become the top priority. I had a conversation at one of the data center conferences with somebody whose role is to look for land for data centers, and that was his top criteria, is you have to go where the power is. So it has become the top criteria in certain markets in the US.

Kim (21:52):
And so we're seeing that this is not unique to just our company or one company, it's across the industry. Absolutely

Dawn (21:58):
Not unique to

Kim (21:59):
Us. Dawn, how are you thinking about the role of data centers in Cummins growth?

Dawn (22:03):
This is a really exciting space, Kim. If you think back a few years ago, data centers was one of many segments that we sold back-up power to. It was one among many. And now it is increasingly, I think last year it passed over half of our sales of power generation, and it is growing at such a rate that it will continue to be a larger and larger piece of what we sell. And the requirements are different, but we are thinking really hard about how should we invest going forward and what should we be investing into beyond just the product itself, the product we sell today. What we're finding and observing in this space is, I've heard it described as almost the wild west in terms of the builds. There is a market need and whenever there's a market need, the nice thing about markets is people rush to meet that market need.

(22:53):
So there are lots of startups. There are lots of pitches about, "I can solve this for you. " And we as Cummins have a long history of dependability. Depend on Cummins is one of the main reasons that a lot of people come work for Cummins. It's a lot of people, reasons our customers do business with Cummins, and we take that reputation for dependability very seriously. And I personally think there's a real opportunity in this space for us to take that value proposition into power and say, "Well, what does this mean when you really need power at speed?" How might people and our customers start to depend on Cummins for more than what they have in the past?

Kim (23:33):
Dependability is absolutely our middle name. If nothing else, I feel like that's one thing that echoes across all of the industries that we support. And also you mentioned fortifying those frontier technologies and making them dependable. That is such a unique value proposition that we're able to offer. Patricia, do you have anything to add to that?

Patricia (23:49):
No, I really liked the comment about dependability because that is what this industry needs. Uptime, dependability, all of those things, trust, the partnerships, all of those things. And I think that has been part of the reason why Cummins is-

Kim (24:09):
Positioned to

Patricia (24:10):
Win. ... positioned to win in this industry. Yeah.

Kim (24:11):
Absolutely. And these are definitely why Cummins is positioned to win. Can you tell me more about that?

Patricia (24:17):
Yeah. So one of the things Dawn talked about is that this is a significant part of our business, part of our strategy. We are making investments in this space. We're making investments in our manufacturing, in our capacity, our supply chain. We're making investments in our people. We're making investments in our products. So we're making a lot of investments and all of that is driven to meet the customer's needs, to meet what the market needs. That is what drives us. We're all about solving problems. So the other pieces of that are, when I think about Cummins, one of the things that really sets us apart is that we are able to learn from so many markets that we are in. We are able to learn from our vast experience. Our experience in data centers ... Dawn, I don't know if you've seen these numbers, but we hit 44 gigawatts of installed power this last quarter.

(25:14):
It's exciting. So we have a lot of experience in data centers. We have a lot of experience in other complementary markets and applications that allows us to learn fast, that allows us to scale, that allows us to support our customers. The other piece that sets Cummins apart is our channel. We've made a lot of investments in our channel. We have a global channel, which means that we have distributors in just about any part of the world that you go where there's a data center. We have global support, but it's local support. And because of that, we are able to meet customers where they need the support, not just from a sales perspective, but from an execution perspective. We're talking about that earlier, from a service perspective and all of those things that are really, really important when you think about reliability uptime.

Kim (26:03):
And it's that global support and the local expertise that really just takes that whole thing home.

Patricia (26:08):
So one of the other things, Kim, is when we think about where our customers are, our customers are not experts in power, right? They're building data centers and we are the experts in power. So we are doing a lot of consultation. We're doing a lot of partnership, a lot of advisory. You started by talking about how fast moving the industry is. Just thinking about how fast regulations in this space are changing with regards to power permitting and all of those pieces. So working with customers to think through their evolving data center designs. Data center designs have evolved so much in the last five years, and we'll continue to do that because now we're talking about AI, we're talking about higher kilowatt racks and things like that. So we are bringing that expertise to our customers, to the market, and helping think through solutions with them and playing that advisory role.

Kim (27:00):
So we started this episode with a look back, but let's look ahead now. Dawn, over the next five years, what do you expect to see play out in the data center industry?

Dawn (27:08):
The first thing I expect is continued innovation and surprises. I think things will continue to move in ways that are really difficult for us to anticipate sitting here today. And as I think about, well, what does that mean as we think about what investment choices we make and what strategy we pursue, I think we need to plan for continued agility and adaptability. And one theme that Patricia and I have talked about a fair bit too is system level design. So I mentioned all the interdependencies across the power system. Designs need to move towards more interoperability, more flexibility in terms of how power is delivered to customers, how our power is charged, how we pay for power. We need more flexibility in all these systems to make sure that we have access to the power we need and that it is the right people are paying for it, if you will, that we're matching basically investment and risk with the people who are incurring that risk and realizing the benefits from it too.

(28:12):
And I expect, when I mentioned innovation, so I expect innovation in products and solutions, I expect innovation in business models, and I think that AI will actually contribute. This is an interesting thing where AI is both the driver of all this power demand and also can be a part of the solution. So I'm really curious and excited to see what the next few years bring.

Patricia (28:37):
We've seen the industry sort of move from a little bit of decentralized. We talked about the shift to cloud, right? A little bit of centralized. But as AI gets to moves from predominantly training to where we start to see AI and application inference, then we're talking about a lot of data being generated and needing to compute real time closer to where the end user is, right? So interested to see how that could drive a shift in where data is processed, data is stored. The concept of edge data centers, for example, is something that we've talked about a lot, which is moving that processing closer to the end user. So interested to see whether that is something that plays out in that way. I really like where Dawn went with the adaptability because we started by talking about five years ago in 2020, we couldn't have fathomed where we are today in 2025.

(29:35):
And I think it's the same thing. I think we have to keep taking a step and then just looking at where we are and seeing where do we go, how do we adapt, how do we change being flexible and we'll get to 2030. And I think we'll look back and we'll smile, but we'll get there and we'll get there strong.

Kim (29:54):
I like that we're just leaning into knowing that there will be surprises, perhaps some obstacles, but the agility Quality that we have is what's going to play into our strength. Dawn and Patricia, I really appreciate you guys making time to join us on the podcast and bringing some clarity into a space that is anything but static. Thanks for having us.

Patricia (30:12):
Thanks, Kim.

Kim (30:13):
So what's the big idea for this episode? As the data center landscape becomes more complex and more demanding, success isn't about chasing trends. It's about designing power systems that can keep up with the world that never slows down, systems that are reliable, resilient, and ready for what's next, because at the end of the day, it's about delivering today, adapting tomorrow, and earning trust at scale.

Show ID (30:35):
Thank you for listening to Power Onward. Your support means the world to us. If you enjoyed this episode, please don't forget to subscribe, rate, and review on your favorite podcast platform. If you want to dig deeper on what we covered today, check out the show notes for additional links or go to Cummins.com/podcast. Until next time, power onward. We're from Cummins, the company that's been innovating toward the future for more than a hundred years. We're no strangers to rapid change, global shifts, and economic uncertainty. We want to be your constant during the energy transition. Whenever, wherever, forever. Power onward.