With degrees in Mechanical and Industrial Engineering and an MBA from Northwestern University, Govi’s career has been dedicated to unraveling the complex challenges posed by data center energy use. He has been at the forefront of developing innovative solutions to reduce environmental impacts while enhancing operational efficiency.
Govi sheds light on the evolving landscape of data center technology. From advancements in cooling and power management to the latest trends in renewable energy integration, he discusses the dynamic strategies being employed to make data centers more sustainable.
They then talk about the critical importance of optimizing data center energy consumption, particularly in the digital age. His extensive knowledge and forward-thinking ideas make this episode essential for those intrigued by the energy dynamics of data centers and their profound implications for our environment and technology-driven world.
Read the full transcript below:
Raymond Hawkins: All right. My name is Raymond Hawkins, welcome again to another edition of Not Your Father’s Data Center. Today, we are joined by my friend from Cummins. I want to make sure I get it right. Govi, you are the Executive Director of Global Data Center Business. So all things data center, you’re in charge. Did I get that right?
Govi Ramasamy: That is absolutely correct, Raymond, and thanks a lot for having me here. So all things data center within Cummins.
Raymond Hawkins: Excellent, all right. So I got to start off with a couple of really interesting facts. Your last name is Ramasamy, no W. Not to be confused with any presidential candidates, right?
Govi Ramasamy: Yeah, I don’t have any aspirations whatsoever.
Raymond Hawkins: Not looking? Okay, just want to make sure you weren’t making any announcements here today. You’re not running for office. Okay, good. All right, just want to make sure. Going to stick with the data center business. Let’s spend a few minutes on you. I know you’ve had experience all over the world. If you’ll give us a little bit of your background, how you got in the data center business, and some of the places you’ve visited, because when I think about data centers, you’ve hit a lot of the very interesting places in our world. So you want to give us a little bit of the Govi background, that would be great.
Govi Ramasamy: Absolutely, Raymond. Thanks a lot. It’s good to talk to you over the podcast. Been with Cummins for 16, 17 years, but my story is similar to any immigrant story coming in the US, grew up in India. Came over here to study in Auburn. I believe that’s your alma mater-
Raymond Hawkins: War Eagle?
Govi Ramasamy: We share that. War Eagle, yep.
Raymond Hawkins: War Eagle.
Govi Ramasamy: Yeah, War Eagle.
Raymond Hawkins: Love it.
Govi Ramasamy: 2002 to 2004. Background in supply chain operation, that’s where my focus has originally been-
Raymond Hawkins: Just some shameless promotion while you do your intro, that’s all.
Govi Ramasamy: We need more of that now.
Raymond Hawkins: That’s right.
Govi Ramasamy: Since then, supply chain operations, joined Cummins in 2006 in Minneapolis in the power generation sector, so it was a really awesome experience joining Cummins. Back when we were actually going through the initial phase of the growth driven by data centers, if you remember, I wouldn’t call it the badness, but actually the huge long lead times that we had in data centers generators back in 2006, ’07, ’08, really through the period of that, and then came the crash. Great experience just living through the supply chain and how we reacted and went through that experience. And since then moved around in different aspects within Cummins. Cummins’ been awesome in giving me opportunities.
Worked in China leading the power generation business for a few years when the data center was actually starting to pick up where it was originally dominated by the China Mobile, China Unicom, China Telecom, and Baidu. Alibaba was coming in right now, then Tencent. And then they call it IDCs, but essentially co-location players started actually picking up their business. So great experience setting the business up and actually growing, working with a phenomenal team that we have in China.
And then did some gas projects, EPC. So was fortunate actually to work in implementing a lot of EPC projects in Europe, in UK in particular, where the capacity market actually was super hard. So we delivered close to 500, 600 megawatt worth of projects in the UK. And then prior to this, I was leading our distribution business in Middle East and UAE.
Raymond Hawkins: Wow.
Govi Ramasamy: Saudi Arabia, Kuwait. And for the last two and a half years being in data center space, I thought I knew data centers before taking this role. But man, a lot of things have changed since in the last seven years and every single day I wake up and I learn something new. It’s been a phenomenal ride so far and I almost feel like actually we’re just getting started.
Raymond Hawkins: Yeah, boy. Man, talk about great global experience, great market experience. The Middle East, China, Europe, North America. I mean, you’ve been in a lot of places and you’re talking about that China Mobile, China Telecom when they were really building out their facilities and I think lots of people, the supply chain crisis that came along with COVID forgets that it wasn’t the first time we had a real bad supply chain issue. You’re bringing up ’06, right? I mean, you have to have been around a while to have remembered it, but man, it’s not the first time supply chain has hurt the delivery windows for all of us. All right, so we appreciate you hearing your professional background. I am going to be remiss, I’ve already held the helmet up. We got a highlight that you were at Auburn in 2004, it’s worth talking about how great that team was. You were new to Auburn, give us just two minutes. I know it’s a long way back, but you can probably remember two or three highlights about the ’04 season for you.
Govi Ramasamy: Where do I start? I mean sports has been a part of my life all through, my dad was a volleyball coach. I played volleyball all my life. My sisters played volleyball for the teams, college. So coming into us trying to figure out which game I can pick up because I can play basketball. But then Auburn, I think the first game that I went to was actually, is Alabama game.
Raymond Hawkins: Oh, boy.
Govi Ramasamy: It was Iron Bowl and it was madness, and I couldn’t actually relate to the euphoria around it and so on, but I was hooked completely. And the passion and the drive that everybody shows, the whole university town, I think it’s 24,000 people. During the weekend, it actually goes up to 40,000, 45,000. Sea of orange and red, and it’s absolutely crazy. And the great 2004, what I remember, I think at the be, if I’m not mistaken, 2003, Tommy Tuberville, he was almost going to be fired that year.
And then somehow he came back and we had a new offensive coordinator and then the defensive corner is pretty good actually, Gene. But then we went 13 zero. It was a rollicking start from the beginning all the way through. We had great players. I mean if you go back and look at it, you have Carnell Williams, right? It’s Cadillac Williams what we called. Ronnie Brown and then Carlos Rogers, Jason Campbell, I mean just name it, the best players in actually Auburn, and I happened to be there in every single game. One of those things-
Raymond Hawkins: Yes.
Govi Ramasamy: The only thing I would actually feel better about that is actually we didn’t get the national championship. That was the bummer because actually we didn’t get a chance to go play in the national championship. I think it was between USC and Oklahoma if I’m not mistaken.
Raymond Hawkins: You got it right.
Govi Ramasamy: We didn’t get a chance. But again, 2011. 2011 we won it so that’s all it matters at the end of the day.
Raymond Hawkins: So Govi, I love that you were at the Iron Bowl the year before, then you got to see the 13 and 0 season. And here’s the great thing about it, lots of people say that that season changed college football because you had an undefeated SEC champion not playing in the national championship game. And if you remember, that’s when everybody started going, “This doesn’t work.” And the BCS and then the playoffs came soon thereafter because people couldn’t believe that an undefeated SEC champion wouldn’t be in the game. And you’re right, USC plays Oklahoma and they blow Oklahoma out. It’s some ridiculous number 56 to seven or something horrible. And the great irony in that is after years later they find out that USC had violated a bunch of rules.
The NCAA vacates the national championship, they vacate Reggie Bush’s, who is their tailback, USC’s tailback, he vacates the Heisman. And to this day the NCAA record books show no national champion in 24. 2004, excuse me. Which is crazy because their Auburn was at 14 and 0 in number two position and they won’t recognize us as national champions. So I’ll just tell you, I have a personal letter writing campaign, I routinely write the NCAA and say you should recognize us. So you got to live through I think college football history because the reason we have a playoff today is that season.
Govi Ramasamy: I’m still waiting for the team to come back and actually regain the glory. I just keep checking the score. This time it doesn’t look like, but hopefully next year it’ll even much better.
Raymond Hawkins: Well, we’re certainly headed in the right direction, but yes, you are right. The glory days are behind us right now. But recruiting looks good. I’m fortunate I get to go to a few games and I love what Coach Freeze is doing. I think that our expectations are only heading up, but I think 2025 is before the next time we could get really crazy excited about things. There’s just a lot of rebuilding to do.
All right, well, we’re glad to learn about you, your background, where you grew up and where you live now, I think we didn’t get that Minneapolis right? You’re largely in and out of Minneapolis as you serve all over the world, right?
Govi Ramasamy: Beautiful Minneapolis.
Raymond Hawkins: Yeah, lovely Minneapolis. A lovely six months out of the year. A little cold, the other six.
Govi Ramasamy: Well, with the climate change I would say it’s getting to be about seven to eight months.
Raymond Hawkins: Okay, very fair.
Govi Ramasamy: I hate to say that but seven months, eight months is okay.
Raymond Hawkins: I got it. All right. Seven months is fair to my friends in Minneapolis. All right, so let’s talk about some things that impact mine and your business every day. Let’s start with I think it’s one that’s on the top, everybody’s mind. Our industry is seen to be a big power eater and I think that comes with some assumptions about how concerned are we about the environment and how green are we and how do we handle the level of consumption required to do what we do? How do we handle it in a responsible way? We can go with that in many directions, but as a guy who handles the data center business for a diesel generation company, that’s probably a subject that’s high on your list and also maybe a little challenging to talk about. But I think we as an industry do a great job being good stewards of the environment. I’d love to hear how Cummins thinks about that.
Govi Ramasamy: Absolutely. Thank you, Raymond. I get asked this question all the time, and for the record actually, I don’t shy away from talking. And that’s part and parcel of actually who we are in the company as well because it’s an industry problem. Our generators, we provide backup source of power for data centers. When everything goes off, when the power goes out, then we have the last line of defense to come to bring it online. Cummins, Caterpillar, MTU, Kohler, those who are playing in the market. So we play a critical role of actually making sure the lights are on when the power goes up.
Our job is to actually make sure that we understand where the dynamics of the market head to, the concern about the decarbonization, the carbon, the climate change and everything and make sure that actually we come up with viable solutions that help the industry move along. So if you don’t come out and talk openly about the challenges actually that are facing in front of us in terms of grid constraints, climate change, and carbon greenhouse emissions and so on, we are doing a disservice. So I’m more than actually happy to actually come and talk about it and we are doing a lot to learn about the problem and see what kind of a solution that we can actually offer that can address that as well.
And it’s an industry-wide problem, it’s not actually specific to one particular component manufacturer or one particular sector in the industry. It is an industry-wide problem and I am glad to be in this place in this time to actually to work with the industry payers to find solutions that makes sense. And even if you take data center industry in general, I would probably say the data center industry has been one of the most forthcoming in terms of publishing how much energy is being consumed and also pushing the grid and the utility to actually incorporate and integrate more renewable solutions into the grid and down to actually tracking how much of their power consumption is meant by hourly matching of renewable energy and so on.
So the industry is doing a lot more. Of course there’s a lot more that needs to be done, but we must recognize where we come from as well so that we can understand where we need to go. The challenge that is in front of us is a bit different where because the significant growth in digital infrastructure, dependence of economy, and this AI driven demand that I’m pretty sure that everybody’s talking about the data center industry, and the grid constraints that are actually emerging.
We have a short term problem in the next several years that we need to address. And the solutions that would actually help us overcome the challenge in the next three to five years may not be the longer term solutions, but maybe a good bridge solution to get to the longer term solutions for which there’s a lot that has to happen in the broader economy. So those are the things that we have taken into consideration that keeps us actually awake at all times to understand where the industry is heading, where the economy is heading, and how as a company we can actually show up and deliver those solutions.
Raymond Hawkins: Yeah, Govi, I think you raised, first of all, I think that’s an excellent summary. I think our industry is very much in the forefront of being aware of, hey, where’s our power coming from? How are we impacting the planet? Are we consuming the power we are in a responsible way? I think it’s on the top of mind and on the tips of people’s lips all over our business. So I think we’re very aware and I think we’re good global stewards of what we consume and we’d agree with you wholeheartedly on that from the processor all the way out to generation and you guys being in that generation chain.
I’d also would love to hear your thoughts on this constraint that we see globally. And you alluded to the AI spike. We can talk about AI demand, but really when I look at the hole in the market or the pain point of the market, it’s really around is there enough power where we need it when we need it because of the spike in demand, and it’s been a spike in demand in all the big markets, right? Either we have a generation problem in a market or we have a distribution problem in a market, but we’ve stretched those markets to the breaking point and I agree with you, we’ve got to come up with some three to five year solutions because the demand’s not going away. So how are you guys at Cummins thinking about that? And I’ll use the word bridging power even though I know that could get misconstrued. But as we think about the global power demand by market, how are you guys thinking about helping bridge that gap?
Govi Ramasamy: So we look at it in two different ways, right? One is how do we provide solutions that can allow a data center operator to get up and running? So if there are good constraints and there are emission regulations that actually governs what kind of equipment that you can put on the site to allow you to get up and running so that’s the problem that we actually look at as one way. And the second one is actually how do you do it in a sustainable way? How do you slowly reduce the carbon intensity and the carbon actually operations of those equipment that you put on the ground? So we look at it in two different ways and our view is right now, if you go by strictly with the EPA regulations, most of the markets, even if you look at US, it requested Tier 2 emissions and it is way different than actually way over than compared to the world unregulated markets.
But what we are seeing is actually is when there are grid constraints. I mean take for example in Arizona or in Northern Virginia where there are grid constraints as you said, the power is available but may not be available 24/7, 365 days a year. So we are seeing that actually there are some emerging pockets where the role of the backup power source in a data center may be changing slowly and gradually from being purely in terms of backup to actually being called upon to use when there is a lot more constraint in the market from supply standpoint. It could be few hours a year or it could be few hours a month, but it could be actually more than just, “Hey, I don’t have a clear power, we can actually shut it down and you need to run the onsite power.” So that’s something that we are seeing.
And for that, there are solutions that are in the market that be coming up with especially Tier 4 emissions, right? We have products that the industry would actually come up with solutions that can meet regulations that can offer an existing diesel or any other solution that can meet the emission [inaudible 00:15:25], allows the data center industry to use the equipment not only when the power goes out, but also when there is a peak demand in the market and so on. So that is one solution that we are working on as an industry.
The second one is a lot more investment going in an alternate achievements as well, right? Because one thing is to actually get the fuel and how you burn the fuel that demands the criteria emissions or how much emissions you put out when you actually run the generators. The second is the well to work, right? For example, it goes all the way from what is the carbon intensity of the fuel when it actually goes all the way from the extraction point or generation point to when you actually put it to work. And their overview is that HVO has a lot more potential to it and especially in data center market because of the way the carbon intensity, which is 70% less than what we have considered compared to diesel towards in terms of full carbon content.
So those are the solution that we believe that actually can work. We are also exploring quite a bit in terms of natural gas applications as well because it does have a lower carbon intensity. But again, it all depends on the use case, right? Do you really want to use that many hours in a year and does it justify the economics of putting in the infrastructure and so on? What is the grid reliability of natural gas grid in the place where you’re putting in data centers? So a lot of those things actually goes into considerations and that is what we are looking at for the next several years is how do we continue to improve our product where we actually use less fuel, and at the same time we have aftertreatment solutions and we have higher efficiency products that actually consumes and emits less emissions on site.
And on the second hand, how do we actually work with our industry partners to explore alternate low carbon fuels that can reduce the carbon intensity of whatever solution that we put in place, and that’s in the near term. And then in the longer term definitely product to be working for a company that is actually investing a lot in hydrogen economy. So we are looking at solid oxide, we are looking at fuel cells, we’re looking at electrolyzers and also different technologies that can actually augment and come with the low carbon or zero carbon solutions on the ground.
Raymond Hawkins: Govi, you mentioned that the mission of your generators may not be solely as backup that you might be looked on to handle peak loads. I may use the wrong term here, but is that somewhat a microgrid idea of hey, I’m going to do generation on site and I’m going to augment the grid when there’s, let’s just say when it’s July in Phoenix for example. Is that what you were referring to there?
Govi Ramasamy: That is what I’m referring to, but I would tend to believe that actually it would be more driven by the local utility companies, the regulation that is mandating it than by choice.
Raymond Hawkins: Right.
Govi Ramasamy: Because any given chance I would, but always the grid power is going to be the cheapest power and in summer mostly likely the grid power is going to be the greenest power available in the market. The challenge is always going to be the peak periods when you really don’t have sufficient power to meet all the demand, and then the grid may want actually require the data center operators to have the flexibility to run onsite power generation or in the case of actually augmented, right? Take your data center part of the lower off the grid. So interconnected micro grid is where I see things are evolving, but again, our hypothesis actually is that the flexibility element of it will start to take growth slowly over the next five years.
Not something that’s going to happen immediately, but maybe by 2030 we think it’s going to be around 10 to 12% of the market. But then once it gets to 2040, 2050, we think it’s going to be about 50 to 60% of the market is going to be where flexibility would be absolutely required. Again, those are estimates and what we tend to focus on is actually is how do we put quantitative data behind trying to actually model it out? When does it make economic sense? When we look at any solutions, Raymond, we look for three things. One is technical feasibility, is it technically feasible? And second is supply chain feasibility, or commercial feasibility as you say, is it really commercially feasible? Because at the end of the day, economics matter. You can have a best solution out there, but if it doesn’t actually mean a cost-effective option for those who are using it, it doesn’t really actually drive adoption. So that part of it.
And the third part of it is more in terms of feasibility of application in data centers. The data centers now are not the data centers that if you go back 15, 20 years back. 15, 20 years back, a data center like 10 megawatt is a big data center. Now one data hall could be is actually is three megawatts, four megawatt of the campus could be 300 megawatt, 400 megawatt. So at that scale, the requirements change significantly. You cannot just rely on actually saying that hey, this solution would work for this a hundred global, 200 global, it has to work at scale. That means if it’s running, if the application’s actually being deployed, the supply chain all the way going up to the component, and also once installation supporting and fuel availability, everything could be scaling up to support the data center needs.
So that is what we look at in terms of hey, is it technically feasible? Is it commercially feasible or economically feasible? And third is the supply chain and the support infrastructure is why will enough to support a data centers. So that is what we actually take into consideration or modeling to understand how we see the adoption growing for different technologies, whether it’s in hydrogen, internal combustion engines, or in pump fuel cells, methanol, ammonia, natural gas, HVO, diesel, renewable, natural gas. So we look at all of those and we model it out to see is it available, is it feasible, and what are the probability that actually will get there? And that’s what we use to actually to guide ourself in terms of bidding on technologies that may not help immediately, but it may actually help us go faster towards a decarbonization journey five or seven years from now.
Raymond Hawkins: That is great insight into how you make decisions, I love that. The phrase, I might not get it exactly right, is it economically viable commercially will this work because how great the mousetrap is, if people can’t make it financially work in their business model, it doesn’t help any. And the fact that you guys have that perspective I think super, super interesting and super helpful because I used to, in the computer business, before I started doing data centers, I sold computers. Hey, I can get you as much reliability as you want. CICSPlex mainframes, I start at about $9 million. How many of them would you like? “Oh, I don’t need quite that much reliability.” Okay, I didn’t figure you did. And so what capabilities that are actually make financial sense for the application?
Govi, I know another subject that we hear a lot about in the data center business and the generator businesses is alternative fuels. Hey, we want to be super efficient, we want the generator to run really well. But there’s also talk about what you can power it with. It doesn’t have to be diesel, it can be natural gas, it can be HVO. I’d love to hear your thoughts and what Cummins experience has been. When we think about alternative fuels for these generators.
Govi Ramasamy: It’s one of the most hottest hot topic that we are debating internally as well. And our understanding of what solutions would work for data centers is emerging, it’s evolving, right? As we actually learn, test some units in different fuels and look at the economics and see whether it actually makes sense or not. Some of the things we looked at, for example, we talked about HVO, hydrotreated vegetable oil that we believe is the right product solution because it can store, it can be mixed in with diesel in any mix, any blend would actually work. We thought much laws in terms of the performance and the power. The well to work emission site is 70% of the diesel, that is really good. And it has shown that it can improve the particular damage matter emissions as well. There are some challenges. The real challenge is the cost premium of HVO compared to diesel, right?
And data centers shouldn’t be much of an issue, but it is something that we’re tracking closely as well. Even if you take us for example, most of the refinery that are actually manufacturing HVO is the west of the Mississippi, but a lot of demand is actually on the east of the Mississippi. So there is a challenge when it comes to the cost of manufacturing HVO compared to the diesel, but also the cost of actually transportation. So if you add everything up, it’s not well set up. But we are hoping that as more data center companies adopt HVO, it becomes a very well adopted fuel and data center industry, and it becomes our primary goal of fuel in the market.
And other areas we looked at just like any other company would be is methanol. That is something that’s been discussed about as a really good fuel and then ammonia and then that is on the internal combustion side and also natural gas, right? Natural gas is known. So it is a good product. The challenge of natural gas is adopting that to the data center application because most of the national gas applications that we see in the market are mainly for designed towards efficiency rather than actually startup time, transient performance that are specific to data centers, methanol and ammonia or something that we are tracking.
It could be a good fuel but it won’t be a drop in fuel. So drop in fuel is that you have a diesel tank, you have a fuel tank, you have diesel, you can’t get HVO, so you’re selling diesel and then you come back and you found HVO, you fill in HVO. The product doesn’t actually care, right? It performs really robustly and so on. But other fuels you cannot actually mix in like that. So it’s not a drop in fuel. So you need to change the fuel handling system of the site, but also the product has to be capable of actually operating with multiple fuels without any compromise.
If you design a data center with a certain power output and so on, without making significant changes, you cannot just go and actually swap the fuels so the D rates are different, the performance is different and the client response is different, but we are looking at each one of them. We are positive in terms of longer term, but we are still not clear about application data centers when it comes to methanol and ammonia in internal combustion engines. But we are exploring those alternate fuels as good carriers for hydrogen let’s say. Because the biggest challenge with hydrogen right now is not just in the generation of hydrogen, it’s how do you transport hydrogen and how do you store hydrogen? We looked at certain application for a three megawatt data hall, three megawatt backup application with storage for 24 hours. The storage itself was more expensive than the hydrogen fuels and it’s going into multiple millions of dollars.
So there’s a lot of technology that still innovation still yet to be had in terms of developed in terms of how hydrogen is produced green at a much more cost-effective manner, which I think the latest announcement from the Biden administration is going to help in setting up these massive of hydrogen hubs around the country. But at the same time, how do you transport the hydrogen, and how do you store the hydrogen in a much more cost-effective manner, if not hydrogen, can you do it compressed without losing a lot of efficiency? Can you store it in a different methanol or some other actually carrier that you can convert industry side of use?
Those are some of the things that we’re looking at. And again, our hypothesis is actually is that for hydrogen or any other actually alternate fuel that will come in to take the place of the internal combustion engine, the entire broader economy has to start using those fuels in much more higher quality in real applications where you see the trucks, the trucks has to be working in fuel cells or hydrogen combustion engines and so on before we would see a broader adoption and broader availability of the fuel that can really come and actually display some of the equipments and the technologies that we currently use in data sales space.
Raymond Hawkins: Govi, how does natural gas, you mentioned it briefly, is it pipeline and going to be delivered that way or can you store it as well?
Govi Ramasamy: Both, right? Pipeline is the best way because you’re talking about the scale, right? If you’re talking about actually-
Raymond Hawkins: Right.
Govi Ramasamy: 10 megawatt, 12 megawatt, you can actually do a CNG, LNG station close by and they can actually come and expand and put an expansion tag and actually use an national gas from there. From the size of the campus you’re talking about, right, it could be 300 megawatt, 400 megawatt. It’s not a much more viable economic solution, right? Then you have to go to actually a pipeline gas infrastructure. And again, the challenge is the infrastructure worth the usage, right? If economics again comes to play. In a HVO or a diesel backup, you can actually store it and you use it. I mean I’ve gone to data centers where for five years you go and actually look at how many hours the generators have been on 30 hours. 30 hours, right? But the requirement is when it is called upon, it has come every single time.
Raymond Hawkins: On every one of those 30 hours it got asked, it needs to be on. That’s right.
Govi Ramasamy: Exactly. So for natural gas, if you’re bringing in generators, again, the industry will evolve where the products will improve to meet the data center transient and the response requirements. So far that’s not been asked from the industry. I have complete faith in the industry and our industry partners that industry will step up. If natural gas is a solution that we are going to be using data centers, we will react altogether to actually get our products to a state where the requirements of data centers, the application from transient response, from responsiveness, startup time, everything can actually improve and meet the requirements. But that means that actually there has to be significant infrastructure investment and to get the fuel to the data center, to get the pipeline to the data center. And that depends a lot in terms of where the main trunk plane is and how far you can actually run the pipeline to the site and so on.
And the second question always comes in is actually what is the reliability of the gas grid? Is it going to be as good? And if you’re talking about Northern Virginia, you have what, two, three gigawatts worth of actually power. They follow them, switch to national gas, and if power goes on, every single one of them actually come online, can the grid really support the mode of the fuel one that is actually taking place in that, right? Those are some of the questions that has to be answered and our view on our experiences actually is that there is no one silver bullet.
Any progress is still a good progress. And that’s the way we approach it, right? Hey, let’s look at every one of them in its own merit. And it may not apply for every single site, every single place in the world when it comes to data centers, but for a given data center application, a different site where the economics are different, where the availability of fuel is different, then the solution could be different. At the end of the day, is it better than actually the ultimate, right? Even 10% reduction in carbon intensity is still a reduction, right? And if it’s economical, let’s go for it, and that’s the approach we are taking.
Raymond Hawkins: Govi, that’s very helpful, thank you. All right, let’s take up one more subject. So AI, and I’d love for you to talk about AI, Govi, from two perspectives. One is it’s in the news, everybody’s talking about it, NVIDIA’s going to own the world, GPUs are taking over, everything’s great. Generative AI is an incredible thing. So talk about it from a demand driver that you guys see inside Cummins. And then if you would, if you can give me two or three minutes on how you guys are using AI in your business, I bet there’s some interesting tales inside Cummins for how you guys are taking all the information that from all of the data points you have and all these gen sets out there to look at the data differently using AI. So can you give me both of those sides of the AI coin?
Govi Ramasamy: So if I go back and remember the conversation that we’re having with the customers back in 2022, even mid-2022, right? Most of the demand that you were seeing was driven mainly by the cloud adoption and just the general digitization of the world and coming off the pandemic and so on. And then the ChatGPT actually brought home and accelerated a whole bunch. So since then I would probably say that the general, what we read in the news is real in terms of what we hear from our customers as well. The demand and the magnitude of change now has gone up significantly, I would say probably two to 3X over the next few years. And I think the big wave is actually coming in terms of 2025 because of the supply chain, right? There is a lag in terms of initiating the project to working through the land permits and everything and getting all the occupants in and service in and everything.
So from 2025, 2026, we do see a significant jump in terms of number of data centers that are going up and vice versa is the demand on our side as well. And just like everyone in the industry, we are also ramping up our supply chain globally to meet that. By the nature of it, we are anticipating that most of the demand for AI driven, especially at the beginning would be in North America and then slowly spreading out to other areas including AsiaPac and to Europe.
And just with the nature of actually AI demand, there is demand driven by training models and the demand driven by inference models. The training models are the first initial builds that are going on. And what we are seeing in general is that most of the data center builds prior to the AI wave tend to be in the Tier 2, Tier 2 cities where there’s a lot of existing data centers, the zone availability and so on. But with the AI, we’re starting to see that that is getting spread out where the data center is actually being built where they have good connectivity and also good power and land available. It’s not so much the data gravity that is putting them towards the major hubs around the world. And that changes our perspective in terms of how we need to get really ourself to support the customers in the field because now it’s going to be out of someplace far away from where the main data center concentration is so we need to make sure that we have local capability there and we’re getting it to support the customers in there.
And there is also some discussion going on about if it’s a pure play training data center, then what is the kind of reliability that is needed on the part? Because again, that is something that is being hotly discussed and debated. I don’t know whether it’s going to influence the longer term. I think the longer term forecast is about 10 to 12% of the data centers that are being built would be for training and the rest would still be an inference or it’ll be for mixed use. And in the case of training, do they really need actually the five nines liability or can they go with actually three nines or can they go with two nines? And what does it actually mean in terms of the backup power requirements and so on? And are there anything that we can glean from the AI itself?
But you can put in second points where you start and stop without losing all the work that has been done as a mission-critical, similar to what you see in other applications. So there’s a lot of things that are going on, so that demand, we still trying to understand how it’s going to play out, but we still believe the magnitude of what is coming in is huge because I mean second one is you ask how we are using within Cummins, and it’s a good classic example of how AI is starting to infuse and improve our productivity in real value.
And just take, for example, Zoom calls, the team calls, there is a summary button, there’s the AI button that is kicking in. And in Cummins, we are starting to use it quite regularly. And you start at the beginning and by the time you’re done, it can actually produce an amazing summary of the context of the conversation about who said what, not as a transcript, but in terms of summary, followed up with action items so nobody can miss, “Hey, I didn’t sign up for it,” right? So that is a good example of, I mean because Cummins, as mentioned, we are one of the largest diesel engine manufacturers in the world and our engines are used in very different applications, in on highway, off highway mining, marine and power generation, oil and gas, every single consumable application that we can.
And there is a lot of data coming in, and definitely mechanical engines, electrical engines, there is a lot of work going on to make sure that something goes wrong, we diagnose and fix it and so on. Just like any other company, all this data was sitting in different pockets in different information databases and so on. There are some work that is going on to say, “Hey, can we deploy AI to go and process that and try to bring it together?” And see, can we connect the dots that we may be missing because the same person is not looking at two or three different files? So those are real life examples.
And if you look at telematics for example, telematics is a good example where most of the applications that are coming up right now are not really meant for stationary power because if you look at the telematics and most of the investment is going in, how do we increase the efficiency of the product? How do we reduce the fuel consumption? How do we ensure that actually we can predict the reliability of when it’s going to fail so you can actually proactively interfere and so on.
But that is a design towards applications where the products are running all the time. In data centers, it doesn’t really matter. With all matters is actually is that, hey, when I really would like the product to come up and run, will it run or not? So the mindset itself is completely different. It’s not about efficiency, it’s about reliability to start. So we are leveraging quite a bit of our partners to actually to go and explore. Can we understand the application with the limited data that we have to predict when the next failure may be happening? Or when something that proactive intervention is required so that we can assure that it’s going to be starting when it’s called upon to start. So those are some of the areas where I see a lot of AI coming in, machine learning, coming to enhance how we are doing and the same thing that I think is actually everything a single industry is going through as well.
Raymond Hawkins: I really appreciate you reminding me. Obviously I’m in the data center business, I think of Cummins and I think of power generation in the data center space, but man, I forget you guys have marine applications and you have so many other uses for your generators. I appreciate getting reminded of that, that it’s a big business. It’s a big world out there and we in the data center space are just a slice of it, supporting it.
My friend from the Land of 10,000 Lakes and my fellow Auburn Tiger, Govi, I really, really appreciate you spending a few minutes with me here and letting us talk Cummins and talk in data centers and getting a little Auburn football in while we did it. So thank you for that, and I hope I can recruit you in my NCAA letter writing campaign to get our 2004 national champion finally recognized.
Govi Ramasamy: Oh, I will. Thanks a lot, Raymond.
Raymond Hawkins: Don’t thank-
Govi Ramasamy: I appreciate you having me here. It is a dynamic industry and super excited to be here and working with each of our data center customers in figuring out what we can do to improve the industry in general.
Raymond Hawkins: Yeah, we’re both lucky guys to get to be in this space at this time, that’s for sure.