Keeping the Lights On

(keyboard clicking) (upbeat music) - You seem like a pretty level-headed guy. Have you always been like that? - Yeah, I think so. Some people describe me as even-keeled. It's helpful, especially in urgent or sort of crisis situations to be able to bring a certain amount of calm. - Tim McAravy is the kind of person you want on your team in an emergency, especially in a power grid emergency. What does it feel like to be outside when it's 40 below with windchill? - Painful. Like when your fingertips begin to freeze or the end of your nose begins to freeze, it hurts. And that hurt can be second sometimes only to the burning when they fall out again. (upbeat music) - Tim is the VP of customer service at Omaha Public Power District. An electric utility in Nebraska with over 400,000 meter customers. - So that number includes all the way from an individual household up to someone the size of Google or other large customers. But on a population basis, we serve over 850,000 people that depend on energy for their lives. - He's the guy in charge of communicating with all the people who use power in Omaha, homes, hospitals, schools, supermarkets, everyone. - And back in January 2024, winter storm Jerry swept across the Midwest bringing sub-zero temperatures with it. - Feels like temperatures 20 to 30 below all day, Saturday, 25 to 35 below all day, Sunday. So just brutal cold outside. - Yeah, these next few days just buckle up and get ready for it. - And right as everyone across Omaha needed more heat, Tim got some bad news. What did you think when you heard that four thermal power plants were going offline? - I just thought that life was about to get really interesting. I was telling my wife that I probably wasn't going to see her for a few days. - Sometimes emergencies happen and only the people behind the scenes seem to know about them. Like a programmer who catches a bug in some code before it breaks the software. Or a wedding planner who finds the missing rings before the ceremony gets derailed. That's what happened in Omaha last January when four of the city's thermal power plants went offline. Tim knew that if the Omaha public power district didn't have enough energy to meet demand, they'd have to begin rolling outages. - So I was talking with and texting with my team. I was getting them prepared for what we knew was likely to come, which would be at a minimum. We were going to have to talk to key customers and keep them informed about what was going on. And quite potentially we were going to have to get engaged with them to explore options for them to help us reduce the load. - By reduce the load, Tim means asking customers to curb their energy use temporarily. It's something that utility people call demand response. Tim and his team began an all-out effort to enlist the help of their biggest customers, factories, supermarkets, schools, hospitals, distributors. And this time, Tim also reached out to one of the biggest energy consumers on the block, a Google data center. We've heard pretty consistently from our utility partners that demand response capabilities will provide tremendous local and regional value during extreme weather events. So data centers are really critical in providing resiliency for the grid. This is where the internet lives. A podcast from Google about the unseen world of data centers. I'm Stephanie Wong, and I'm your guide to the people and places that make up the internet. In our fourth season, we're exploring how data centers are enabling a more resilient world. In this episode, energy. We have three stories about how data centers are helping clean up the energy system and how to manage the growing energy needs of AI. First, back to Omaha, where Tim was dealing with frozen power plants during winter storm jerry. - The severe cold actually froze and constricted the water flow on the Missouri River. So we were unable to intake water into our coal generating facilities to be able to operate them. We need the water intake to be able to turn the steam, to turn the turbine to generate electricity. - Those units began going offline on Saturday, January 13th, right as homes were really cranking up the heat. Fortunately, the Omaha Public Power District had spent millions preparing for something like this. - It's often the thing we complain about when it's cold out, the wind, but with the Missouri River freezing over like this, OPPD says the wind helped keep lights and heat on. - They had added wind turbines and solar panels to diversify their sources of power. They had weatherized important equipment, and they were ready to import power from neighboring areas. But there was still a risk of outages. So the utility asked the public to turn down the heat by a degree or two. And Tim and his team began calling, texting, and emailing their biggest customers, asking them to reduce their power usage. Google's data center and nearby Pippilian Nebraska was on Tim's list. - Customers collectively saved upwards of 75 megawatts of load from our system. Google was one of the biggest contributors to that, saving many, many megawatts of reduction over those couple of days and were tremendously impactful in helping us balance that demand in the energy. - In the end, the power in Omaha stayed on. And most of the public never even knew how much Tim and his team scrambled behind the scenes to keep it that way. - People have come to expect and to be able to count on reliable energy. Most, I would say, average customer, my neighbors, my friends, or whatever. Most of them had no idea of the degree of planning and orchestration and reactiveness that was going on behind the scenes. And I think that's a testament to the US power grid. - So how does the data center that runs the internet around the clock shed energy demand to help the grid during an extreme event? Well, it starts with people who are as passionate about the power grid as Savannah Goodman. So I am a proud energy nerd. I have a background in analytics, modeling, product management. Savannah is one of the people behind the demand response tool Google used in Omaha. She leads the team that develops data and software solutions for climate goals. So basically, we make digital tools that help companies track, optimize, and transform their energy footprint. - When did you first realize you were a nerd? - Very early, I was in high school and we had a club called Green Software Initiative, which I joined right away. And we actually built out a biodiesel lab. We were taking the waste vegetable oil from our cafeteria and making biodiesel. And we were using that biodiesel to power our concession stands for our football games. Instead of going to the mall or doing something whatever normal high schoolers do, I would go get the sticky dirty waste vegetable oil and make biodiesel out of it. After biodiesel, Savannah moved on to the power grid, studying all the tools that help reduce strain on the grid and cut emissions, things like microgrid, solar, and batteries. In 2020, she joined a team at Google that had been building a tool called Carbon Intelligent Computing. The carbon intensity of a power grid depends on the time of day. When there's more wind or sun, the electricity system can run on cleaner energy. Google developed software to match the computing loads inside data centers to the times of day when the grid is using more renewable energy. We actually covered the origins of this project all the way back in our first season with the technical lead, Anna Radavanovich. Wow, we have compute load. We can directly know because we log everything at Google. So why couldn't we actually map it to power and map it to carbon footprint in real time? We started asking the question, what else could we do with these capabilities? Could we do something not just for Google's footprint, but could we also help the grid? And we thought what types of demand response would be most helpful for the grids that we specifically operate on? How could we accommodate those load shapes without impacting our customers or users? And how could we, again, start testing this on a small scale first before we roll out something bigger? So back to the original question. How can a data center that runs the internet 24/7 drop its energy use on demand? You can imagine there's very inflexible workloads like a search query, right? If you go to Google search, you type in a question, like what's the latest Taylor Swift album? You expect to get that answer right away. I know I do, being a big swifty. And that is not a workload, right? That we want to touch or mess with. That's like super critical, super important. But there's other workloads that are more kind of behind the scenes that do have some flexibility. So for example, when a YouTube creator uploads a video or when they're doing some editing and they have some processing, they may not need that to happen right away. And so that kind of workload may be has more opportunity to shift in time or even potentially in space. After refining carbon intelligent computing for internal use, Savannah and her team started working with utilities all across the world with Google data centers connected to their grids. Places like the Netherlands, Belgium, Ireland, Taiwan, Oregon, and Nebraska. So I had a notification that would-- wake me up at like four or five a.m. on call so that I could set off the DR curves. You know, those are some of the fun challenges you do as you're innovating something new. But I, you know, it was worth it, right? To say like, hey, like, this is a first of a kind thing. We're literally doing demand response to data centers. So setting off DR curves at four a.m., please tell me you call one of those big red emergency telephones. The page, the page, no, we didn't have that, but you know, we got Google chat on our phones and that that works pretty well. After some initial success with pilots, Google started relying on data center demand response in increasingly extreme circumstances. Like when Russia cut off the flow of gas pipelines to Europe in late 2022. There's a lot of concern that there would be major gas shortages and therefore power outages right across Europe. Russian energy giant gas from has said that it will once again drastically cut gas supplies to EU countries through its main pipeline, Nord Stream 1. And this comes as fears of an acute gas crunch continues to group Europe. So we, you know, thought, hey, like we've been piloting our demand response capabilities. We should leverage them here. You know, it's the least we can do. And so what we did was we ended up reducing our data center demand using our demand response tool during peak hours. So from five to nine p.m. local time every single day over the course of the 2022 2023 winter. And so this helped alleviate grid stress and basically prove that our data centers could be, you know, a reliable resource during this time of need. So given what you've been able to do in shifting data center demand, what role do you see these facilities playing in resilience of the grid? I mean, I think we've heard pretty consistently from our utility partners that deban our response capabilities will provide tremendous and has provided tremendous local and regional value during extreme weather events. And large energy users are going to be a key piece of that. And data centers are a large energy user. And I think similarly a lot of the data center operators have these ambitious goals like 24/7 carbon free energy, you know, not just demand response, but we can really leave the way with these ambitious goals to drive the investment in diverse technologies, develop the ecosystem to enable system-wide decarbonization. With the help of AI, data centers can assist the grid in times of crisis. And as we've heard so far in the season, AI is also helping accelerate decarbonization efforts and climate resiliency. But over the last year, many experts and commentators have turned their attention to the other side of AI, the energy demands. So what you're seeing is that demand is growing at a significantly robust clip. Yet new supply is challenged right now and those challenges are stemming from this power availability issue. So it's worth backing up to ask. What are those long-term impacts of AI on the energy system? To answer that, we need to revisit the history of hyper-scale computing. Back in 1999, some researchers predicted the internet would consume half of U.S. electricity within two decades. Today, data centers consume 4 percent, according to the International Energy Agency. Those wild figures in the 90s made sense when the growth potential for the internet was so vast. So a lot of times when these forecasts are done that come up with alarming numbers, which is actually happening right now with AI as well. They assume that technology remains constant and that demand is increasing exponentially. Boris Holtzley was one of the architects of Google's early data centers. He later became the senior vice president for engineering and stayed in that position for 20 years until 2023. He's now a Google Fellow. Back then, hyper-scale data centers didn't exist. Instead, Google rented space in a warehouse and put some computers in cages. An energy inefficiency was a real problem. We hadn't either the time-know-the-expertise to do data centers ourselves, and we're also too small. And so these data centers are basically just rooms full of servers and cages. So why are cages that separate the different customers? When did it become clear to you that there was an energy challenge to be solved in data centers? That happened actually very early when we were still pretty small. I think it must have been 2000 or 2001, and I think we were planning to put several thousand servers there. And then we were only half done, and then one day the provider calls us and says, "Hey, you've got to really be careful because our alarms are going off and our your breaker is about to trip because you're using too much power." And so we went over to our numbers and they just didn't work out. Like it was not possible to use that much power with the service we had. And so the question became, "Where is that extra power?" And so through a series of tests, we discovered that really the power supplies that were in the servers were just very bad. They were wasting a lot of energy. They were not efficient. So of the energy that was coming in, a little bit more than half actually went to the actual servers, and the rest just became heat in the power supply. And when we looked a little bit further, we kind of realized that this was happening because the power supply vendor was saving a dollar or two of costs per power supply. So everyone knew how to make efficient power supplies, but they were a little bit more expensive, and they weren't paying the power bill. So in the end nobody was asking for efficiency. That was when we discovered that, "Wow, the world really isn't paying attention." And there's a lot of money that we are wasting because our equipment is not efficient. And that's when really that whole journey started. Between 2010 and 2018, compute workloads increased by 550%. But data center energy demand only increased by 6%, according to research published in the journal science. But back in the early 2000s, that trajectory was not guaranteed. As Ores and the data center team prepared to build the first Google HyperSkale facilities, energy use was a top priority. And at first, it was just about buying better components. You could actually buy a better power supply. But then it was clear that there were other reasons, for example, industry standards that really prevented the power supply from being very efficient. I think 2004 is when we really had our first server that was optimized for energy, and then 2005 was when we had our first data center that was almost twice as efficient as the industry average, even though it was the first one we did. And so we had all kinds of bugs, probably. We weren't the best in the world at figuring out data centers at the time, but it was much, much more efficient. So in the 90s, there were a lot of people saying, "Oh my gosh, the internet is going to gobble up half of US electricity." And in reality, while we've seen the internet traffic explode, data center energy use has grown incrementally. So did that surprise you or did you kind of know that as you were building it? No, so that actually didn't surprise us. So if you have something that takes a certain amount of energy, you do do something, and then you think you're going to need a thousand times more, then obviously you have forecasted as well. There's a thousand times more energy. But technology is not constant. In fact, highly not constant. And so you get improvements every year in the energy efficiency per unit of work. And so maybe you do have a thousand times more demand, a furious down the road, but you got 700 times more efficient at doing it. And so in the end, there's only factor 1.4, increase versus a factor of a thousand increase. So that's kind of the first one. And then the second one is actually economics. If it really was true that you're using 10% of the world's energy or something like that, you would spend enormous sums to build and operate these data centers. And nobody can afford it. There isn't enough revenue to actually pay for all this, which means the economic pressure of a company to actually find these improvements. And also the ability just to pay for that is limited. So sort of the forecast cannot be true because nobody can pay for it. It would be so large that nobody can pay for it. So it's really a management theme to say, we can do better. We can do a lot better. We actually have to find ways to double the speed, to have the cost so that we actually survive. So that really is why these forecasts don't play out. First of all, the world is not static. Improvements happen, like really meaningful improvements happen all the time. And second, the economic pressure. In the end, you have a business. You can't afford your costs to go up exponentially. So you're really motivated to find ways to solve that. So apply that to the AIH then. AI workloads are much more computationally intensive than conventional computing. How do you minimize the spike in energy use from AI? AI is really a huge computational problem. You need a super computer to make a new model like Gemini. And then that super computer runs for weeks or months to just build this one model. And then you also need a lot of hardware to survey efficiently to actually answer questions using them all. So absolutely. But we're really in the very early days of AIH. And we're learning very rapidly how to do both of these things training and serving much more efficiently. So I recently did a study that showed that from 2017, when sort of the modern way of ML was invented to 2022, there was an efficiency improvement of a factor of 88. Right? So in just five years, things got 88 times faster, slash cheaper, slash less. energy off that kind of AI. And so even though the demand obviously went up a lot, a VACR-88 is huge. And so that's going to play out for the next several years as well, because we're really still at the beginning of understanding how these things work and finding ways to make them more efficient. And so pretty much every new time we're building a new version of Gemini, that version is substantially more efficient in its training or in its serving than the previous one. And then the last effect is that the data sensors aren't there just for themselves. They run stuff. And so many of the applications that run in those data centers actually help reduce energy. If you think you're perhaps, you know, email instead of physical mail, video conference instead of a physical meeting, like these use much, much, much less energy than the physical equivalent. And AI is just the next step of being able to optimize things very efficiently and very flexibly. And that really offsets an enormous amount of chemical or energy or material usage that happens elsewhere. Because that same approach can be used to make a building slider to make a car slider or to find better ways of doing batteries that use less material to have more energy storage. And therefore in that battery manufacturing save you to the amount of energy. And so if you have AI or if you have this sophisticated algorithms, you make other processes more efficient. And therefore reducing energy by using compute to run the rest of the world more efficiently. So AI can help save energy by designing better batteries, finding more efficient ways to cool buildings, reducing materials waste and manufacturing, or as we heard earlier, helping data center shift demand during critical times. But what can you do for the broader electricity system? To tackle that, it's helpful to understand how the grid is set up. The electric grid is the largest machine that humanity is ever invented, the most expensive machine that humanity is ever invented, the most complex machine that humanity is ever invented. And it was built as a kind of one way, highway. Astro Teller is the captain of Google's moonshot factory, called X, an organization designed to invent and launch cutting-edge solutions to tough problems. It's known as an innovation factory. He helps build teams that are inventing radical solutions to big problems, like modernizing the electricity system. So you had these very centralized fossil fuel burning plants that then just broadcast electrons to all parts of the let's say a city. And statistics would tell you, minute by minute, kind of roughly how many electrons were going to get used. That was basically how the electric grid was set up. It has expanded through a series of acquisitions. And so if you are the operator of a large grid in the United States, in any country around the world, you as the system operator literally do not know where every wire is, where every transformer is, where every inverter is on your grid. That makes it very difficult for you even to take care of the grid as it currently is. And it's absolutely out of the question that you can start to use 21st century tools and to take advantage of the fact that computers can now do things that humans can't do in terms of making trade off and creating additional resiliency. So the architecture of the electricity system hasn't changed much since it was first built. But in the last two decades, there have been a lot of external pressures forcing change. And Paige Crayhan was at the leading edge of one of them. I started as a solar sales person and would go to like literally farmers markets in Fresno, try to help people figure out what it meant to put solar panels on their roof. I used to carry dog treats in my pocket because a lot of the solar customers in the early days had dogs. Paige leads tapestry, a team based out of X, the moonshot factory. Tapestry is aiming to virtualize the world's power grids and use data science and AI to shift that outdated model that Astro described. But before that massive undertaking, Paige spent a lot of her time at farmers markets, store steps, and retail outlets pitching people on the benefits of rooftop solar, trying to figure out how to help explain the way people buy energy and explain the opportunity to do things differently. And over the last decade and a half, a lot of people have done things differently. Residential systems have exploded across the US. In 2024, nearly 5 million households in the US have solar on their roofs, up from a few hundred thousand in 2010. The number of installations will likely double again by 2030. So we're talking about very different levels of scale from when you first started in solar. How is it now starting to impact grids? At this scale, instead of a couple early adopters having rooftop solar panels in their neighborhood, in some places around the world, you have more than 50% of a neighborhood. The end of the cul-de-sac has all rooftop solar. And it's exciting because you're covering your generation with low carbon energy sources. But if everyone at the end of the cul-de-sac goes solar at 2 p.m. in July, all that energy is feeding back into the grid at one node and causing fits and starts for the folks that are responsible for keeping the lights on and me keeping it reliable. And it's not just solar. Similar challenges can happen on local grids with technologies like batteries and electric vehicles when they aren't charging or discharging at the right times. Of course, we need a lot more of these technologies to clean up the grid. But they're presenting new complexities for utilities and grid operators out of unique point in history, right when power demand and climate threats are surging. You've got wildfires. You've got a three-day deep freeze in Texas where it's not supposed to freeze for three days. You've got new storms and new cadence of weather patterns. And so it's the weather plus the aging infrastructure, which are incredibly challenging to manage. Then let's make it even more exciting. How about we ask you to double the size of the network while you're managing the aging infrastructure? Because we're going to electrify transportation. We're going to do things like add more data centers. We're going to electrify heating. We're going to start building and manufacturing more things. So we just need more electricity. Do that affordably. And if you could go ahead and do that, but you can't rely on predictable fossil fuel, you need to rely on weather-based energy. So you're also going to need to forecast the wind and water and solar. And if you could do that right now because climate change is upon us, but also affordably. But also, this is just like-- I mean, it is an absolute quagmire conundrum. It's just such a challenge. And it's a beautiful challenge because when you have all these things coming together, I think you have a real opportunity to do things differently. But from an infrastructure management perspective and a grid management perspective, the stakes have never been higher. All I want to do whenever I talk to folks that work in this space is roll my sleeves up and have a big coffee and see if we can help. And maybe give them a hug too. Huge hug. Huge a big hug, a coffee, and we got you. And what are the risks to not properly tackling these challenges? If we don't solve these quickly, I think in the US, at least we looked at what happened with Hurricane Sandy and some of the major storms we've had, people die. I mean, if we don't get our reliability and resilience issues sorted out, there is a human life and a human health and safety component that is absolutely critical. And probably sits at the front of every governor and regulators mind as they think about taking care of their communities. But at the same time, we're really thinking about the criticality of adjusting the way we deliver energy, because if we keep going with the status quo, and all we've done is keep the lights on, well, we've certainly introduced a real problem for our kids as the climate issues continue to persist. So stakes are high if we don't get it right. So to recap, an aging grid, an explosion of renewables, rising demand for electricity, worsening extreme weather, and a grid that is in desperate need of an upgrade. You might think of the grid a bit as, if I'm going to date myself on the guy of our episode, or a bit bubble gum and duct tape, where you sort of say, OK, we've got an issue, let's fix it. Alice Sheena Jackson is the VP of Strategic Development at AES, a global power provider. It's a very large company that owns multiple utilities and operates a wide range of power plants around the world. It also develops solar wind and battery projects, often serving data centers. And she leads a team of people trying to create the clean grid of the future. To have it move from the 100-year-old grid that worked really well in an analog world, but is not going to work as well as we are moving to more digital assets connecting to it. So we don't always know the exact state of the grid. That mission includes some really exciting work, like using AI to dispatch large batteries at the precise times they're needed. It also includes a lot of mundane stuff, like knowing the age and location of pieces of equipment. We're solving problems all the time. From a maintenance perspective, just saying, OK, here's an issue that came up how might we solve it. Getting an idea of what's actually on the grid, even. Even when it's digitally assisted, it's still long and lengthy and a bit perhaps boring. I mean, if you really love Transformers, you might love to go out and gaze at them. But for the vast majority of us, we'd say, "Okay, wouldn't it be just really great if like, you know, as someone changed something out in the field, they just updated that into a system?" And they input what they had installed and took a photo of it. And then you knew, you know, from that photo, maybe it's serial number. And you could then compare that serial number, you know, in the type of asset that it is against a large database of failure rates and performance standards. And all of these things that exist in siloed different areas of information, but are not in a singular centralized place. So if there was somehow, magically, a near perfect digital twin for your grid, you the system operator, then you could start number one. Just managing things, super basic things like, of all of the trees I could go prune today, which of those trees are most likely to take down one of my power lines. Or you have a 10 year wait line for getting renewables onto the grid. You are legally obligated as a system operator to have a bunch of people sit around for about a month running what if scenarios before you plug a new solar field or wind farm onto your grid. While you spent the month thinking about that, four more just got on the back of the line. So enter tapestry. Tapestry's mission is to make everything on the grid visible. Using data science and AI to plan, predict, and monitor every asset across the vast electricity network. We really need something like a Google Drive for energy. Like we need a place where the information about the energy system is organized, easy to access, correct, and you can collaborate on it. So you talk about virtualizing electric grids. What does that mean? So we think there's a combination of a few things. One, we take information wherever it's available. So we sort of take spatial information. We use street view and satellite imagery and lots of tools that we have at our fingertips and in the industry to complete and correct some of those models. And then we also understand that we need to think about power flow. So can we run simulations about what's happening at that grid? We need to understand the emissions profile and trying to synthesize all that together so that the folks who have to make decisions about what to deploy, what to build. So what you can do differently to think about a virtualized grid. And I am sort of a nerd. So I think about SimCity and like drag and dropping things, but having the power flow is actually work. So what you're saying to that planner is, all right, I'm going to speed up what took you a long time to do because let's just make your job easier. But then I'm going to introduce tools that you didn't know you had, right? So can I show you the, let's say you have a congestion issue two years from now at this wind farm in the middle of the night when the wind's blowing. What if we put some batteries here? Then you wouldn't have to reconduct or build a new transformer or build a new transmission network. So we're allowing more tools to come into the view of those planners. So what we're really trying to do in the virtualized energy system is put all of the tools that the folks who are responsible for the grid put all their tools at their fingertips so they can run more scenarios more quickly so that we can have high reliability, low cost, not overbuilt, low carbon network. So when we deploy this, the very first thing we did was say, listen, you're going to have to raise your hand and say you're, you're okay going on an innovation journey because it hasn't been done before. One of the companies willing to go on that journey, AES, AES runs multiple utilities and also develops large-scale renewable energy and battery projects. So it understands all the challenges outlined earlier, intense extreme weather and aging grid and the need to build lots of zero carbon energy quickly. And meeting those challenges requires a single digital window into the entire network. What are the stakes of this work in your mind? We've got certainly the sort of existential things around climate change and weather. Can we keep the grid running under more stressors? And having this visibility will help us maybe even predict where our grid is at greatest risk and early respond to events that can occur. And also do we want our large consumers like data centers or industrial factories to be able to bring economic opportunity into our communities? Because if we are addressing climate change and we want those big sort of consumption centers of energy for the society that we want both economically and experience wise, then we need to get a lot of renewable energy to those locations. And that renewable energy is dynamic. The way we consume energy is dynamic and in fact the lines we send energy across are dynamic. And so being able to see that dynamic production, dynamic consumption and dynamic transmission allows us to be less constrained to make better use of the assets that we have on the grid, then we can be more confident about the reliability of that transmission system because we're just more confident in what it's going to look like, what it's going to be demanded from it. So for example, it is complicated to put a solar field onto the grid. It is also complicated to put load onto the grid like a data center. They're both hard. But if you knew about both of them to some extent, they can cancel each other out. But if both of those decisions are siloed and they don't talk to each other, there's no mechanism to help the system operator have the load that they're putting on the grid and the power they're putting on the grid, meet in the middle, then they may literally end up saying no to both. Page, over the course of your career, you've personally witnessed all of these forces putting pressure on the electric grid, renewables, climate threats, and even new load like data centers. But all of these changes have collectively brought a new opportunity to modernize the system. Have you had tried to build tapestry a decade ago? Would it have been possible? Absolutely not. Tapestry could not have existed 10 years ago. Everything that we derive information from, whether it's an inverter or a PV panel or our coffee maker or thermostat, that amount of compute and information and sensory insights, that didn't exist a decade ago. And the ability to process that and run that in the cloud and serve it up, this is becoming more and more ubiquitous. And that level of information can help us. We don't need to go out and re-sensor and redo things. We can actually take the inputs that already exist and are being processed in today's infrastructure by the existing data processing mechanisms. So we have more information coming from more sensors being processed all around the world. We have smarter data centers, right? So we're using global load shifting and insights around how you optimize the data centers to be able to run processes that we would have never thought possible. And so I think from a technical perspective, we've crossed a trancem of what is possible. But now we also have this sort of really important inflection point in the industry where folks want to do things differently because they see the capability. So I think there's this dual moment of technical opportunity that didn't exist 10 years ago, cost of compute, sensors everywhere, data centers. And then you have, I think, maybe for the first time, I think from deep, deep challenge comes deep, deep opportunity. So you have some of the folks who might have been most hesitant to do things differently, kind of raising their hand and saying, we do need to do things differently. Let's partner, let's think about how we put this data forward. Because I think at the end of the day, what I hear when I talk to folks all around the world is just like a really deep, personal and professional commitment to helping, to keeping the lights on affordably with low carbon. That is what people want to do. And I think, you know, now we have the technology to really help some of that willpower with new tools. That's it for the second episode of the season. Next up, how data centers are strengthening a critical area of society, food and agriculture. In this episode, you heard from Tim McAravy, Savannah Goodman, Oors Holtslay, Astro Teller, Alex Sheena Jackson, and Paige Crayham, where the internet lives is produced by latitude, media and collaboration with Google. You can subscribe to the show anywhere you access your podcasts. Please give us a rating if you're enjoying our journey together. And if you want to learn more about how Google's data centers are benefiting communities around the world, click the link in the show notes. To see what keeps the grid operator in Indiana up at night and learn how AES is pioneering a new digital grid in partnership with Tapestry, watch our companion mini documentary, Keeping the Lights On. It's linked in the show notes. I'm Stephanie Wong. Thank you for listening.