Google's head of energy and power on how AI data centers can help improve the grid

[00:00:00] Speaker 1: Amanda, you've been working on energy at Google for more than a decade now, and I just, this moment that we've been in since ChatGBT launched in 2022, I just, I wonder if you can walk me through how your responsibility, the scope of your role has changed as demand has just accelerated in this new AI hyperscaler build-out era. [00:00:18] Speaker 2: Yeah, it's really been an incredible ride. We've built out a really strong and big energy team at Google. As you mentioned, I've been there over a decade, and the step change that we saw around the time of AI was really the large power needs that AI uses, and we realized at that time that we've never been more critical to the business, and thankfully, we've been doing this, building data centers, powering our data centers for over 20 years, so we were right place, right time. But the industry was also not ready for this, and so what we did then is really start to partner with our utility partners, partner with the industry to say, hey, large loads are coming, larger than anything that really has been connected to the grid in the past, and we want to really partner with you. We're no longer a customer, we're a partner, and making sure we design and develop this the right way, and that's where we partnered with large utilities to design a model where large loads can interconnect and ensure that they pay their own way. Make sure that if there's new infrastructure that needs to be built on the grid, that we have a structure where we can build it, but also do it responsibly and think about how can we continue to meet our climate commitments by bringing generation on that's clean, diversified, and is really building that energy transition to the future. [00:01:33] Speaker 1: I almost feel like you have these three distinct models. So in a place like Michigan or Minnesota, you are actively working with utilities to fund bringing new sources of power, more often than not renewable sources of power onto the grid. And then you have a place like Nevada, we've got these utility tariffs, and it's almost like the consumer is underwriting this transaction to a new kind of financial engineering. And then finally, when I think about Georgia, of course, you're looking to optimize what you already are getting from nuclear power plants there. So let's start with the Michigan, Minnesota example. Walk me through perhaps the Excel relationship. And when you first started to have those conversations about like, hey, we'd like to fund bringing new sources of power onto the grid. [00:02:13] Speaker 2: Well, as you mentioned, we have a rich and diverse energy market here in the US. So no two markets are the same. As you mentioned, in Minnesota, we are working with Excel, who's a regulated utility provider. And there is where we are working with them to say, hey, none of your standard offerings work for a large, large customer like ourselves. And we really started by partnering with them. And we found a great partner who shares the same values that wants to make sure to protect rate payers, but also shares our decarbonization. Decarbonization goals to get creative with us, sit around the table. We know we're not the only player in the industry. So we wanted to work with them to make a standard that didn't just apply to us, but that was replicable across our peers that may not have the history or experience of building large data centers or connecting this much power in history. So we brought everyone to the table on our side. We brought many people from our team on their side, their system planners, their new tech development team, and really looked at what made sense for their grid. So we started by building a structure to make sure we had the right pathway in place that we paid for all the infrastructure costs needed to meet our growth. We committed to paying for that whether or not our data center load showed up over time. But the really cool thing on the generation side was looking at what made sense for that grid. And together we decided that that was adding another 1400 megawatts of new wind, 300 megawatts of solar. And what's really cool is deploying the world's largest battery by capacity and iron air battery that can run 100 hours of storage. And why that made sense in Excel, for example, is because Minnesota is blessed with a lot of wind, but a lot of that wind is either being underutilized because it's not being used at that exact time of day or there isn't transmission to get all that wind out to other parts of the state that needs it. And we really saw that this large battery was needed at this time because it can help to shift that wind to different times of the day or help be a conduit for transmission that isn't built out yet. And because of that, Excel was able to run the analysis and say that's going to save Minnesotans 1.1 billion dollars that they would have otherwise had to spend just because we were there and helping to improve those shared system costs. [00:04:33] Speaker 1: I definitely want to talk about that side of protecting the rate payer. But in terms of what Google is funding as part of a build out in this case with Excel or if it's DT in the case of Michigan, when you say they're covering electricity and infrastructure costs associated with the data centers, in practice, are we talking everything from generation to transmission, substations, backup capacity? And then how do you think through the risk of, and I can't imagine at least the next year or two that we'd run into this, but the risk of stranded assets if demand doesn't materialize in the way that it has been over the past 12 months? [00:05:05] Speaker 2: Yeah. So as I mentioned, we have a very diverse energy system. And so how we go about structuring these depends on the markets. What we structured with Excel looked at both the infrastructure and having them build and bring online the generation. In some markets, those two are decoupled, where we bring our own generation into a deregulated energy market structure, but also ensure that we're paying for all the infrastructure. And what that really is, is it's committed to the we're committing to pay for it over a number of years, right? So when the infrastructure is built, we commit to pay for it, even if our load doesn't show up. And that protects other customers against stranded asset costs, because those assets are going to be built, we're committing to pay for it. And so it's just making a more robust system. [00:05:52] Speaker 1: And then in the case of Michigan, the AG there is potentially looking at raising Google's minimum monthly payment from 80% to at least 90% of these projected demand build outs and potentially having stronger exit fees. Is that something that Google would accept those additional protections for the state? [00:06:10] Speaker 2: Yeah, when we saw this years ago, that we needed to create a standard that all large load customers could do, we went with the 80% because we actually wanted to leave room for flexibility. And we're really a leader in flexibility. We've signed over a gigawatt of data center flexibility at our data centers with five different utilities across the US. And so not having 100% where we're disincentivized to actually be a flexible load on our system and where we're continuing to evaluate our growth in any given market was really by design to enable the innovation that we need to actually be a better grid steward and be a good citizen for the grid by being able to deploy really interesting and cool flexibility products. [00:06:54] Speaker 1: And then on the rate payer side of this, I've flown in, I've gone to different, not just Google's data center sites, but gone to town halls where people are talking about how they feel about some of these new, especially the AI hyperscaler campuses that are coming up. But, you know, speaking to the consumer, how do you explain to them not only how their existing rates will be maintained, it won't be affected by having this new buyer on the grid, and then also like there's this conversation around how their rates might actually go down. So can you kind of explain that model in the case of either Michigan or Minnesota? [00:07:28] Speaker 2: Yeah, absolutely. So what we saw in Minnesota is by structuring our agreement for power the right way, rates will go down by 1.1, or anticipate to go down by $1.1 billion of savings for customers. And the way that that works is the grid is a shared system that we all connect into. And so there is a large degree of your bill or my bill, our energy bill is distributed of that fixed cost system that we all benefit from. And so when a large customer like ourselves comes in, we can actually help cover a larger portion of that fixed system cost that all rate payers actually receive the benefit from. And then it's critically important that as we need more power on the grid, we bring in our new generation, too, to meet our load and make sure that we pay for that cost. We also found that our grid is grossly underutilized. In a lot of cases, and this is why flexibility is so important, we find that our grid is really designed to the peak energy use of any given day. But for those non peak parts of the day, the grid goes underutilized. And so what we're trying to do is by bringing these large loads, we can help utilize assets that have already been built on the grid. And that just lowers the cost for all rate payers in addition to bringing new generation onto the grid and sharing in those fixed costs. [00:08:49] Speaker 1: You know, I thought about demand response a lot, first in the context of the Bitcoin miners in Texas, but how should we think about that in the context of a customer like Google? I can't imagine it's easy to spin that up and down and serving up Gemini. [00:09:04] Speaker 2: It's something that we are growing and learning every day. We first started figuring out how to shift our workloads actually to meet our climate ambitions, to move our workloads to when the sun is shining and the wind is blowing. And we took that and were able to use it actually in Europe during the European energy crisis in Ukraine, to really be helpful when the grid was under high stress, that we could actually reduce our workloads to be a good grid citizen. And from there, we've just grown our ability to think about how do we shift compute? How do we run workloads that aren't time critical at other points of the day and still provide the same service to our customers? We're also looking at how do we site batteries next to our data centers? So it actually doesn't look like it looks to the grid like we're reducing our workload, but we're actually just shifting our workload to an on-site battery and being able to provide flexibility through our own co-located generation in addition to what we do actually within our data centers. [00:10:09] Speaker 1: I do feel like the way that Google is approaching this is different to some of your hyperscaler peers. And there has been some pushback, though, in Pine Island. It's almost become a part of the campaign set that politicians are addressing. How do you navigate those relationships? [00:10:24] Speaker 2: I think, as we all know, you can't be given trust. It has to be earned. And there's a lot of confusion out there. There's a lot of misinformation and noise out there. And we're just committed to showing up in our community and listening. People want transparency and they want to be heard. And we're partnering with Excel. We're partnering with our local community. We're absolutely committed to building the most environmentally responsible data center we can. And this is why we partnered with Excel to do it the right way. [00:10:55] Speaker 1: How scalable is what you're doing in Georgia in terms of trying to drive greater efficiency from existing nuclear power reactors, get more, I don't know, juice for the squeeze? Yeah. How is it working out so far in Georgia? Yeah. How receptive have they been? And then where are you looking to potentially replicate that across the U.S.? [00:11:12] Speaker 2: So in Georgia, actually, we just announced this week that we are working with Georgia Power to what's called an uprate of the existing nuclear power plant to get another 95 megawatts out of the existing facility. Because of that, we're going to save ratepayers in Georgia $900 million. And we're really excited about that because that's capacity that can be brought online faster than building a whole new nuclear facility. And I think that when it comes to energy efficiency, there is so much that can be done. There's probably 100 gigawatts at least of capacity that can be better utilized on our grid. We just have to figure out the right structures to be able to access that. And again, that's only just going to continue to defray existing fixed costs on the system. But we'll also continue to need to build out our system. 70% of our transmission grid in the U.S. is over 25 years old, and it's starting to reach the end of its natural life. And that means we need to invest more capital as a country to update these grids. So the timing of this industrial revolution that we're really living through on the energy side that's being brought on by demand for AI couldn't be more perfect because it's right at the time when we needed capital to fix aging infrastructure in the United States. [00:12:28] Speaker 1: So on the nuclear piece of it, of the 100 gigawatts worth of power on our grid that's underutilized today, would you say, I mean, half of that is nuclear? And what percentage roughly is the nuclear piece of it? And then separately, how scalable is this to the U.S. nuclear fleet, what you're doing in Georgia today? How much can you first off just scale it within the state there? But are you looking are you having conversations right now about using the same method elsewhere? [00:12:53] Speaker 2: So the U.S. has today the largest nuclear fleet in the world. China is quickly building much, much more nuclear at a much faster pace than we are. I believe we have about 90 to 95 gigawatts of existing nuclear on our system. But we haven't been investing in nuclear for the past decade. And really that slow down and continuous build out of new nuclear means that there's only so many existing projects that can be uprated before you have to go build something new. We did this at the Duane Arnold nuclear plant with NextEra where we upgraded that system. We did it in Georgia. But there's only a handful of nuclear nuclear plants that are able to extend the life or uprate them because they're young enough that that makes financial sense. So we're also looking at investing in new nuclear. That, of course, takes longer. Anything we start today for new nuclear will be a decade at least before it shows up. And that's really why we realize this is going to be a critical part of our future and a diversified power grid of the future. And there are smarter ways that we can do it now than when it was built in the 70s. And we partnered with Westinghouse, for example, to bring our AI solutions. They're a partner using our Google Cloud to actually reduce the timeline, the construction timeline to build new nuclear by being more efficient. There's always stuff that happens in the development of any large scale project where a piece of equipment doesn't show up on time. And then you have, you know, a very large workforce sitting idle. What our AI capabilities have been able to help Westinghouse to do is figure out, OK, here's the facts on the ground today as I'm working through my construction project. How can I best optimize with these real time changes and what's happening on the ground so that we can constantly be building and working and moving on to the next task. And that's critically important, not just for nuclear, but building out all of this infrastructure to ensure on time delivery of the infrastructure and for our customers. [00:14:53] Speaker 1: One of the reasons why we're so interested in telling the story and having Google be a main character in it is because part of your approach is a pretty big departure from how other hyperscalers are treating this. I'm also thinking about the build out with Colossus and a lot of those natural gas turbines that were brought online pretty quickly. Some off-grid systems as well. Talk to me about how you made the decision, because what you're doing, it's not inexpensive. It takes a lot longer, I would imagine. So one, when that became a priority to scale this way. And then two, is this a model that other hyperscalers could replicate? Do you expect to see some of your peers follow suit? [00:15:30] Speaker 2: Absolutely. We've been we are blessed that we've been building data centers for over 20 years. That has helped us have really strategic relationships with utilities all across the U.S. So when we go to a utility and we talk about, hey, we want to go build a much larger data center. Number one, we have a history with them. They know who we are. They know we are going to do what we say. And they want to partner with us because they know we have the right long term approach to this. Any infrastructure you put in the ground, whether it's a data center, a new power plan or a transmission line, is an asset that's going to be there for decades. And so when we think about how we build, we don't just think about what do we need next year. We actually start decades looking decades in advance because that's how long this infrastructure will be there. What's right for the system today? What's right for the system tomorrow? And then we work backwards to decide what does that mean we need to do today? And I actually think that when you do that and you do it thoughtfully with partners, you actually can meet your short term delivery needs because you're sophisticated in the space. You have those good relationships. They can work with you to get you the capacity you need in the short term because they know you're going to be there for the long run and you're going to be committing to building out with them what the system needs. [00:16:47] Speaker 1: And that's exactly what I've been so curious about. When you think about scaling, I mean, Google this year has disclosed that they're looking at up to $205 billion worth of CapEx. So much of that going towards this hyperscaler build out. Does the equation start with where can we just get the power? We'll figure out the plot of land later. We'll figure out the regulators and even the community reception to the idea of this later. We have to start with where there is sufficient power. Like, are you step one to all of this? [00:17:14] Speaker 2: Very. Who's step one has always changed. There's a lot that goes into deciding where you site a data center. Of course, we want to be in a community that wants us. But power is a critical, critical part of it given the increased power demands of AI. So we work with our partners in land and network with our community development to say, where does it make sense for us to grow? We also think about where our customers are located to make sure that we can serve inference, for example, closer to the edge where our customers are located. But power absolutely is a critical component of it. [00:17:46] Speaker 1: And where does it make sense to build this new power data center where we can access large power. In terms of optimizing for co-location, does that mean that you have like a hub in Silicon Valley or I don't know, like something as close to Silicon Valley as you can get, but also as close to Hong Kong as you can get or Beijing, where we have so many of these open source models running, which require massive amounts of compute. Yeah. So we have data centers of various sizes all over the world. [00:18:10] Speaker 2: And so when we think about our larger training data centers, these are the ones you hear about that are a gigawatt plus in size. Those we have in locations that absolutely are very thoughtful about power, power cost, power availability. And then we also need to serve inference and just our regular cloud customers closer to where they operate. And that often looks like much smaller facilities in cities across the U.S. and around the globe, as you mentioned. And those could be a Google owned and operated data center or a data center that we lease with a third party provider. [00:18:44] Speaker 1: Is there any sort of internal mandate in terms of prioritizing the U.S. build out versus any one of these data centers internationally that you just described? [00:18:53] Speaker 2: We have customers all over the globe. So we're really being driven by where our customers want us, where the demand is, sovereignty issues of where we can actually host the compute for different customers around the globe. The vast majority of our growth still remains in the U.S. and is predicted to be in the U.S., but we serve customers all over the world. [00:19:13] Speaker 1: There's obviously this race underway to get as much compute online as possible. I mean, to the point that Google's renting capacity from SpaceX to serve all of the demand that there is for Gemini. It's quicker to get data centers spun up potentially with off grid power. How do you think about prioritizing the near term goals against, to your point, if you get more capacity built onto a grid, you establish these multi-decade contracts. That's for the long haul, but it delays it. And we're just in a race where weeks matter, where days matter, it seems. [00:19:45] Speaker 2: Yeah, absolutely. I mean, in the case of SpaceX, that's a very short term contract because our demand from Gemini Enterprise customers was huge. And we are working in West Memphis, Arkansas, to build our Google owned and operated data center where we are excited to really do the same thing we did in Minnesota, which is working with our utility in Arkansas, Entergy, to bring customers over a billion dollars of energy savings by paying our own way, bringing 600 megawatts of new solar and 350 megawatts of new battery storage onto the system to bring our own generation. But that takes time. And when you have customer demand like we've seen for a Gemini product, we had 400 million users in May, and that grew to a billion in August. That's our fastest growing Google product ever. We will sometimes have to use short term solutions until we get the data center fully connected. But what's important is that in the long run, we have a stable energy supply that's reliable for our customers and it's built the right way. [00:20:51] Speaker 1: I mean, we talk all the time about the power shortage that exists and really the main bottleneck to AI scaling is just that. But I wonder, and to your point earlier, part of this is just the fact that the, I mean, the interconnectivity lines are 25 plus years old. And so just the transmission lines aren't as capable, even if you do have grids that have been spun up with new power sources. So in terms of thinking about this, is this a problem of where data centers are being cited, slow interconnection queues, utilities not using the grid efficiently or just some perfect storm? [00:21:23] Speaker 2: The challenges with our grid are not new. And I think that for the first time in over two decades, we're seeing load growth on the customer side. Historically, demand growth from customers has been less than 1%. For all of my 25 year career in the power sector and for most of those in the power sector today, they have never seen demand growth like that. And so I think that the challenges with interconnecting new projects with how we study generation and loads have always been there for permitting continues to be an issue. And what's happening now is it's just bringing to light that these challenges really need solutions and answers. The incredible thing about this demand that is bringing to the power sector is really being able to find solutions to age old problems in the infrastructure community. One of the cool things that we're doing is think by co-locating generation closer to the data center. We're not thinking that that necessarily needs to be the long term solution just to be co-located off grid. But actually, how can we build a virtual grid where we have islanded or load plus generation of places along the grid that can act as a flexible resource to really support where transmission isn't being built out today and use more efficiently use our existing infrastructure? [00:22:47] Speaker 1: Why is it do you think that gas powered plants are becoming so widely used for AI data centers kind of across the board? And I mean, this is something where Google is thinking, you know, let's help build more solar wind batteries at your sites in Texas instead of how are you able to how are you able to index away from this? [00:23:05] Speaker 2: Yeah, I think that there are a lot. We've seen a lot of new entrants to the space in everybody. There's a lot of new players working to serve, meet this growing demand and build data centers. We've been doing this for over 20 years. And so when we said, how do we co-locate a data center next to power? We know that that having a diverse resource mix of a micro grid is actually a more reliable product. These machines and equipment need to go down for maintenance on on an annual schedule and a multi annual schedule. And so the best way to have high reliability is to look at using multiple generation sources, whether that's a combination of wind, solar storage in Texas and using gases as a peaking product. The other thing to remember and we're experiencing here in the world today is gas is a commodity and you're subject to the commodity prices of gas. By having a diversified portfolio of generation assets, you're actually reducing your volatility risk and dependence on on both gas supply to run a plant, but also the price volatility that comes with it. [00:24:14] Speaker 1: In terms of the demand response, if you are a buyer who's able to curtail your energy use, is that something where you're given optimized rates for that? Is that something that's a part of these contracts? [00:24:25] Speaker 2: The benefit that we see the most in being able to be flexible is that we can connect to the grid faster. So it allows us to have what we call a non-firm interconnect, meaning most customers today, you connect and the expectation of service is it's always on when you need it on. And when we talk about when we talk to invent this with our utilities and create new standards, it's saying, how can I connect to the grid today? But if you need me to drop off, I will drop off. And that is a much smarter and more durable solution while they build out the infrastructure where we can eventually get to a firm interconnection than just going behind the meter where we don't have the full force and reliability of the grid. [00:25:06] Speaker 1: And that's what I was going to ask you. When you do this behind the meter power consumption, what do you lose in the process? Yes, you can spin up a site faster. But what's the trade off? [00:25:14] Speaker 2: The grid was designed for a reason. Actually, in the 1900s or 1800s, everyone had their own behind the meter generator that was powering a small part of their community. And they realized that was highly inefficient and highly expensive. And also, if one generator went down, you didn't have power. And so the grid was developed by recognizing by pooling these resources on the generation side, we can have much higher efficiency, lower costs and have greater reliability of our systems. And so I think the trade off there is that you are dependent on a single source risk versus being able to be connected to a much more robust network. [00:25:57] Speaker 1: And there's also this question of obviously Google has stepped up in several of these states to fund the infrastructure build out themselves. But the utilities also prefer to do it sometimes. So help me think through that pricing dynamic. When is a situation where Google would want to fund it versus utility? Why would a utility push to be the ones behind the financing? Does it change the economics in terms of the back end when you actually price out power that you're serving up to customers through the grid? [00:26:24] Speaker 2: Yeah. So our system was designed and it's different in every utility in different places for the utility really to be the one to build and bear that upfront fixed cost and share it and recoup it back through rates through their customers. What we saw though is that model breaks down when we talk about really, really large customer loads. And that's why we went in and partnered with our utility partners to build a structure that says at a certain point for a very large customer, that model breaks down and let's build a new model where we are accountable. We'll pay for 100% of the infrastructure costs. We'll bring our own generation to you to make sure your other customers are not bearing any cost of our growth. [00:27:08] Speaker 1: In terms of this clean transition tariff, what you're doing with Nevada Energy with some of their geothermal power. Can you walk me through the financial engineering behind this and what really works about this model in terms of thinking through new ways of financing the build out? [00:27:22] Speaker 2: Yeah, what really works about this model is that it enables us to bring new technologies like advanced geothermal through FURBO online that utilities may not have been able or incentivized to build because the first of a kind of it is a little more expensive than building a gas plant that's been operating for decades. We know that we're going to need more technologies in the future to meet our growing demand and we also want to meet our decarbonization goals and build a smart grid of the future. So by structuring this in a way where we commit to paying for that additional geothermal plant and no other customer has to pay a premium for that. But we also get the capacity from that plant, which is credibly important for us to grow. We have confidence that the capacity will be there. The utility has confidence that that costs aren't going to be borne by others. We wear the risk if it doesn't get built or show up, then we can't grow. But we're very comfortable with that risk because we have an incredible team that works in partnership with our developers to make sure these are developed and built on time and online. [00:28:33] Speaker 1: You mentioned that your team started with five people when you first joined over a decade ago. How many people are on the power team now? [00:28:39] Speaker 2: I think I was the fifth. We have well over 150 people working globally around the world, all purpose driven to do this the right way. We're incredibly excited about this energy transition and all the amazing build out that can come because of this demand. And really, really, this is a moment in history for anyone that's worked in the power sector to really design the grid that we know we needed in the future. We're building it today. [00:29:08] Speaker 1: I know that you said so much of this requires years of lead time to get set up for future data centers. When you look at the goals that have been set internally for the company. And when you see this, I mean, to your point, a precipitous rise of Gemini inference and all these needs and the Gemini Enterprise customers who are just devouring this new tech. Are you concerned when you look at the power that's available, whether that's in the U.S. or, you know, on a wider scale, like kind of looking out to your global locations. Are you confident that you're going to be able to meet those demands in a year from now and five years from now? Or is there just a serious dearth of energy and you can't build fast enough against that? [00:29:46] Speaker 2: Everything that we're doing today is to meet needs five years from now and everything we're meeting real time today as we as we deploy new customers or support new customers. We built three years ago, four years ago. We've really, really scaled up our team, our capabilities. We recognize that how we approach this problem in this industry means we have to think incredibly differently. This is why you see us making what we call bets today or planting the seeds today for new technologies and infrastructure that we know aren't going to come online until next decade. But it's because of the the always thinking five, ten years ahead and doing those things today that when that day comes, we're confident that it will be there. So starting from scratch would be really, really hard. Thankfully, we're not starting from scratch. We've been building to this moment for over 20 years and we just know the step up and change and how we approach it has to be different to meet this higher scale. [00:30:42] Speaker 1: I mean, obviously, you have your finger to the pulse of the build out and how it's going. Are you confident that by January of twenty seven, the bandaid of using SpaceX compute will be gone and you won't need any sort of. [00:30:52] Speaker 2: Our work is to build our data centers the right way with our communities as fast as possible to to to have on time delivery. And I am confident that everyone in our infrastructure team is working incredibly, incredibly hard to deliver that on time in the right way. [00:31:09] Speaker 1: So much of the build out across the U.S., there are communities that are welcoming of these new data centers. And then there is a lot of pushback. I know that we talked about it in the context of Pine Island, but this is something that you're figuring out. And Arkansas as well and Minnesota. Talk to me about how you're working with communities on the ground. [00:31:26] Speaker 2: Yeah, we've been building data centers for over 20 years and we are first and foremost is showing up responsibly. And that is just as important today as it was 20 years ago. And what does that look like? Number one, it's knowing that no two communities are the same and they just want to understand, learn and we show up and we listen. We figure out what that community needs and how we can benefit them. And we're also ensuring that we pay for all of our 100% of our energy costs and protect affordability. One thing that we've been building out for a number of years is our energy impact fund, where everywhere that we show up in addition to rate payer savings that they can recognize from us being on the system is actually investing and working with local NGOs in the community to help direct community members better weatherize their home. Getting a new air conditioner is expensive, but an energy efficient air conditioner or even just a new roof really helps you use less power on the system. What we found are that these programs in North and South Carolina have saved individuals $700 on average a year on their energy bills. We've seen in Omaha up to $1,200 a year in energy savings just through working with our communities to invest to make them better, safer homes that use less energy. [00:32:56] Speaker 1: And even five years from now, do you think that the American public generally is going to look at this AI data center boom as something that really strained the energy grid or actually helped finance a much needed upgrade? [00:33:08] Speaker 2: I think that one of the challenges that we're facing now is that there is a lot of new build out that's being discussed, what we're talking about, what we're doing, but it hasn't actually been built yet. And I think once they see it built, once they see it built the right way, once they recognize the benefits that that brings, it actually is going to be a net benefit to the system. It's just hard until you see it and it's actually been delivered to recognize it. [00:33:36] Speaker 1: The only other piece to this is obviously there's been this huge emphasis on renewables that you're bringing onto the grid in some of these states, whether it's, you know, wind, solar, some of the geothermal efforts in Nevada, but like bringing renewables onto a grid isn't inherently a good thing. You need to have the base load to counteract it. There's just a little bit more volatility that has to be dealt with. So how do you think through that equation? Obviously, like you came up with the power industry, you understand the nuances of this tech. [00:34:01] Speaker 2: Yeah, we we look at we look at the grid and we look at the resources, right? The resources of wind or solar are not evenly distributed across the world and definitely across the U.S. So you look at where are their resources, you look at where their resources are already built. If there's tons of solar already built in a particular place, then it doesn't make sense to necessarily build one more. And actually in that location on the grid, it makes more sense to build storage, to actually help shift that solar from the peak of the day to 5:00 PM when everybody, everybody comes home and turns on their lights and starts making dinner. And so what that takes is really an understanding of how our grid works, how our system works and putting the right resource in the right place that also can work to decarbonize our electricity footprint. [00:34:50] Speaker 1: How closely do you work with the Gemini team as they get new data in terms of adoption, in terms of upcoming contracts that they're signing with these external enterprise customers? [00:35:00] Speaker 2: We have a full process that looks at the demands of our customers, the energy efficiency improvements of our TPU chips and what does that mean for us respectively on how much power we need from the grid, how our models are running and updating and how that looks in terms of the down effects of the energy use. So we're in active conversations to always be innovating because the growth and the evolution of this is happening every day. [00:35:28] Speaker 1: Do you find that a data center site runs more smoothly or better optimizes for the power available if it's primarily TPU based versus using NVIDIA's equipment? [00:35:37] Speaker 2: We're able to optimize our data center for any chip. We are constantly evolving and putting new chips in our grids and we have tons of engineers that are both working with our utilities and then working within our data centers to manage that. [00:35:50] Speaker 1: I do also though think of an Amazon or Microsoft that also has similar longstanding relationships and we just haven't seen similar build up programs like what Google is doing and I wonder why? [00:36:00] Speaker 2: I just came back from climate week in New York and the feedback that we've heard from all of our industry partners is Google has the most sophisticated energy and power team in the tech sector and I take that with real pride because we are constantly have been thinking about this and building this out for a very long time.