What does “sustainable” even mean when AI demand – and data center projects – are accelerating? In this episode, we unpack what sustainability means in the AI era, where power and water factor in and how data center sustainability is increasingly about more than just how electricity is being generated.
Welcome to The Five Nine miniseries digging into what's changing in the infrastructure behind modern businesses.
Catch the video at top, listen to the audio edition and read our transcript below, or watch this and future episodes on YouTube.
To learn more about the topics in this episode, check out:
- Where are renewables in the data center power conversation?
- Is Google’s big hydropower deal a sign of things to come?
- Trump’s big AI energy pitch is just microgrids by another name
- Data center growth has sparked a natural gas 'gold rush'
- Are batteries the next big thing for data center power?
- Meta goes all-in on nuclear power with 6.6 GW plans
- In focus: Data center nuclear tech creeps closer to reality
- The Five Nine: What happens when data centers come to town?
- The Five Nine: Can nuclear fix the big data center power problem?
This podcast is written and hosted by Diana Goovaerts. It is edited by Diana Goovaerts and Matt Rickman. Liz Coyne is our executive producer. Special thanks to guests Wannie Park, Yasir Arafat, Bala Naidu, Marissa Hummon and Dan Ye.
Diana Goovaerts, Fierce Network: A few years ago, sustainability was all hyperscalers wanted to talk about. Renewable power this, carbon offsets that. Then AI hit and the sustainability conversation took a back seat. But this begs the question: what does “sustainable” even mean when AI demand is accelerating?
The answer? These days it’s less about 2040 targets and more about what communities – and build timelines - will accommodate right now.
Wannie Park, PADO AI: Yes, sustainability is part of the data center conversation. What I would say, though, is it's probably not number one on the priority list because I think things like time to power and activating economic development within the data center and AI market is probably number one.
Yasir Arafat, Aalo Atomics: The constraint has been evolving over time, right? In the past, the constraint was, "Hey, you know, are there enough land or chips?" I think that has shifted over to, “are there enough electrons and liters of water?”
Bala Naidu, Bloom Energy: In the past, there were a lot of questions about greenhouse gas emissions, but nowadays it's more about impact to the local community, like noise, water, you know, air pollution. Those are the things that we are dealing with.
Marissa Hummon, Utilidata: I think sometimes operators frame up the sustainability question as a contrast to them getting to market faster or to being the most efficient data center, and we really try to reframe that conversation with them to focus on what are the underlying technologies, what is the visibility, what are the control systems you need to be both sustainable and to run a very successful data center.
Diana Goovaerts: Welcome to The Five Nine miniseries Data Center 2.0, the show where we break down what's changing in the infrastructure that powers modern business. I'm Diana Goovaerts, and today we are unpacking what sustainability means in the AI era, where power and water factor in, and how data center sustainability is increasingly about more than just how electricity is generated.
It won't come as a surprise to anyone paying attention, but power is the number one constraint for data centers today. It is top priority, and the faster data center operators can secure power, the better. That means that while sustainability is still on the slide deck, it is increasingly competing with time to power.
Bala Naidu: Definitely power is the constraint. And, you know, when customers come to us, they've done all the research on the land and the water availability and where the gas is available, and then they're really talking about deploying power in the next six to 18 months. They don't have time for waiting for power to come three to four years from now. So, they really want solutions right away.
Diana Goovaerts: So, what kinds of power options are on the table? Well, as Fierce has covered before, there is a rush to natural gas as a short-term solution, but data center operators are also exploring nuclear as a cleaner, longer-term option.
Dan Ye, CollegeNode: So, the conventional answer by far is nuclear power. By far. It's the cleanest power. It's a great baseload player. It doesn't have intermittency issues, and the new reactors do not even have meltdown concern. So unfortunately, there is a political problem with nuclear power. And I think if that bottleneck can be solved, I think nuclear power is the way to go.
Yasir Arafat: The current playbook isn't working and it's not gonna work in the long term if you continue this way. Something structural has to change, and that is bring your own dedicated power generation that is also clean, right? Nuclear is the firm, clean, base load piece of that overall puzzle.
And so if you think about it, right, you can't train a frontier AI model on intermittent power twenty-four/seven, but you can’t also like decarbonize on gas.
But if you look at nuclear, this is a very important question that everybody asks: "Well, how, how far are we out from getting a real nuclear solution?" And that question arises because if you look at traditional nuclear, it has not been very fast, right? We essentially need a completely different model.
If you have a nuclear reactor that is the highest power you can put on the back of a truck, then try to build hundreds and thousands of these so you can come down the cost curve and increase your speed. So we are building our first nuclear reactor. We've actually finished building our first nuclear reactor at Idaho National Lab, which will power on very soon. So hopefully by the time this episode is released, that reactor has achieved criticality.
We're building then a power reactor by the end of this year, which will make electrons at full scale and power a co-located and co-built small data center. If we can prove that out, then starting next year, uh, and, and, and 2028 and onwards, we're going to ramp up our manufacturing capability.
So, we're not talking about 2030 beyond. We're talking about Aalo speeding through all of our development cycles to get solutions to the customer late '27, early '28, and then ramp up by '29. And we have a very aggressive target internally, which is by 2030 we want to be the first nuclear company to deploy at least a gigawatt every single year from there onwards.
Diana Goovaerts: And then there are renewables like hydroelectricity, wind, solar, though some of these are harder to fit on a data center campus and others are tricky to use due to intermittent power generation issues. But the need for power is so great that there could be opportunities for renewables on the horizon.
Dan Ye: So, hydropower is ideal because it get rid of one of the key bottleneck of solar, right? It doesn't have intermittency. But it's a niche solution because number one, you can't build hydropower wherever you want, right? It has to has the right geographic, geological formation.
Number two, the permitting process is significantly more difficult because you have to flood a large area of land. I used to work at the Federal Energy Regulatory Commission on hydropower permitting, and -- it is a bureaucratic nightmare. So, I am a big fan of hydroelectricity, but given the political reality of this country, I think hydroelectricity is still not... you know, it's a niche play.
Energy storage was my specialty, and I've been watching the industry actually grow from 2006 all the way to now, and I am very happy with where it's going. It's going a little slower than expected, but there are now such myriad of options. We're seeing batteries taking a larger part in the heavy lifting. We're looking at forced air. China just created the first forced air energy storage, and it seems to be running pretty well.
We're looking at, you know, sand as a form of heat storage. So overall, I think If we just make the incentive right we can even accelerate the growth of energy storage.
Marissa Hummon: So interconnection queues for data centers that are going to be built in the next few years are long and very uncertain, and I think this is an opportunity to consider or, or for data centers to move towards sustainable, renewable onsite generation in a way that the circumstances really haven't been there in the past, where that choice is potentially the fastest time to market, the fastest time to power up on the data center than waiting on an interconnection queue that has a high degree of uncertainty.
Having the visibility and the controls gives you a lot more space to be flexible in what you choose to power that data center with.
So, um, our technology operates really, really fast. It can also inform those slower technologies, the demand management technologies that are kind of shaping the whole data center. That also means, though, that you can choose to have intermittent variable renewables as your power source if your data center has some flexibility to meet that variability.
Diana Goovaerts: And then there are fuel cell options like what Bloom Energy is deploying with Oracle.
Bala Naidu: We convert natural gas or any other fuel directly into electricity without combustion. Because we do that without any combustion, we don't produce any of the harmful pollutants that other technologies produce.
Diana Goovaerts: So suffice to say, the sustainability decision tree for data centers has many, many branches.
Yasir Arafat: When it comes to power solutions, there are three major options that we have regardless of what you pick in terms of deployment. One is, you know, "Hey, delay your project until you get clean firm power."
Number two is, "Well, build dirty now and then clean up later." Or three, you have all the clean firm power, you need to just bring it along. So, if you see like today, hyperscalers are not choosing number one, right? They're not waiting for power to come. Most are doing some version of number two, right, with gas turbines as a bridge.
Gas turbines fill the near-term rational choice because of speed, but it does not make sense unless there's a committed plan to replace them.
Diana Goovaerts: Power may get the most headlines, but guests noted that water also shows up fast on the list of issues for data centers operating at scale.
Yasir Arafat: More and more what we're seeing is water is becoming a kind of a siting veto, right? Nevada has already banned evaporative cooling for new builds, and there's a lot of other states like Colorado River, Texas, and other areas that are feeling the real pressure of water shortages. So, water used to be an afterthought in site selection. I think today it's often the first conversation along with power before we even talk about, “oh, do we have enough fiber in the land?”
Bala Naidu: A lot of people use water for cooling. So that is one aspect of the water usage in a data center.
The other usage is in the power, when you get the power from the grid, the grid intrinsically uses a lot of water. You know, a lot of the legacy grid is filled with combined cycle gas turbine plants or coal plants nuclear plants. They all use a lot of water.
And so when data centers are looking at sustainability, the overall water usage, there is a metric called water usage effectiveness. It only typically looks at the water usage inside the data center. It doesn't consider the water usage coming from the grid.
Diana Goovaerts: That last point really matters because there are a few key metrics in sustainability. There's water usage effectiveness or WUE, which Bala mentioned. There's PUE, which is power usage effectiveness, and there is carbon accounting, which can get a bit creative.
And all of this means that a data center can look efficient on paper and less so in reality.
Yasir Arafat: The challenge there is PUE does not measure where your power is coming from, whether it's clean or not, right? A coal-powered data center can have a great PUE, you know, but your emissions are really high.
So that doesn't really work towards our sustainability metric. The other one that is important these days when people are looking for, "Okay, how do I cool my data center?" is water usage effectiveness. And that is becoming more and more important question, especially in drier regions. But a lot of people, that's easier to game those numbers.
The industry average is about, what, point six liters per kilowatt hour. But it doesn't really distinguish where the water is coming from. Is it potable water, reclaimed water, or am I putting strain on the region that I'm in? And the biggest flaw in WUE, in my mind, is that it only measures direct water usage in the data center. It does not account for how much water did I use in the power generation facility that's powering the data center. To me, that may be five to ten times more than, you know, what the data center uses.
Then lastly, carbon intensity. To me, that's the most important one in principle, but that's the one that's gamed the most in practice that I've seen. Companies routinely use market-based Scope 2 accounting. You know, they get renewable energy credits from a grid they don't even reside in. So, the true metric in my mind ultimately boils down to: What is the carbon intensity per useful work?
Diana Goovaerts: But data center sustainability isn't just about where a facility's energy is coming from. It's increasingly also about efficiency and innovation.
Wannie Park: There is so much, for lack of a better word, leakage, right? Leakage of power, leakage of compute, and things like that.
In a lot of these data centers today – and these may be more on the training side – the utilization of the chips are probably 12% to 14%, right? Now, you can't really get to 100% 'cause you have to manage some headroom and efficiencies and, you know, all that stuff, right? You don't want the entire thing to come crashing down.
But the fact that it's only 12% – what it is, is it's actual leakage over phases of like actually generating the compute. And if you can actually juice that a bit, right, and get that from 12% to let's say 20% without impacting uptime and all the different business requirements, the ability to flip that on the other side of it and saying, "Hey, that means less water," right? That means less power. That means potentially more sustainable energy, renewable energy, things like that.
And you can actually create a model that is actually more sustainable, more efficient. And for the data center operators, it's actually higher revenue, right?
Marissa Hummon: So on average, a power system that is not managed in real time is utilized between 30 and 50%. And the reason it is so low is that the power consumption spikes, which on the grid might come from air conditioning use or EV charging. In the data center, that comes from, you know, coincident workloads all spinning up at the same time.
But those spikes drive the capacity ceiling for the data center. If those spikes can be managed, they could either be tamped down a little bit, they could be spread out in time a little bit, they could be spread out geographically between data centers.
Now you can raise that average floor from 30 to 50% up to 70, 80, 90%, and now you're utilizing that already built capacity two or three times more than what you can do without that real-time observation and control system. That's what we have been demonstrating,
Diana Goovaerts: This is the under-discussed side of sustainability. Workload placement, scheduling, and better utilization can turn the same megawatts into more output, reducing the pressure to build the next gigawatt campus immediately.
Wannie Park: As companies and data center operators and the actual users of these data centers become a little bit more sophisticated, there is probably ways to actually shift jobs – which is also power – both spatially and temporally, right?
So if you think about demand response or any sort of grid flexibility mechanism, it's the same thing with workloads. It's not mature enough where you can just say, "I'm gonna take any job and move it tomorrow to a different data center." It's not there yet, but I think that's the direction that we're headed.
And what that also allows then us to do is basically say, "Hey, it's gonna be really congested. It's gonna be very dirty if you, if you run this job right now. Maybe we push it out to the night and we send it over to somewhere where it's powered by hydro," or something like that.
If there was a way to provide more access to all the data, whether it's workload to power or cooling, the amount of innovation that you'll see in the space would just skyrocket, right? Because now what you can do is really figure out ways to become more sustainable, become more efficient, you know, better ROI on power, all that stuff.
Marissa Hummon: So before we should build another data center plant or location, we should make sure that the current data centers are actually utilizing what they have available to them.
And then second, we want all of those electrons to go into the work that that data center is supposed to do. We don't want it to get lost as heat or wasted or stranded. And in order to do that well, you have to know where your power is flowing and what are your options in real time for controlling that or moving that into an optimal state.
Diana Goovaerts: When thinking about where we go from here, it's easy to think sustainability may be lost to an ever-rising tide of compute demand. But it seems the conversation around sustainability is actually poised to shift yet again in the near future.
Wannie Park: A lot of what we're talking about for these big build-outs are for training jobs, right? As you evolve into inference, which is really where it's headed, and I mean, a lot of the market forecasts are basically saying we're gonna go 80/20 by 2030, right? 80% inference. And so what that really means is that the requirement and the necessity of those one megawatt racks no longer exist, right?
The way to look at it is that it's a lighter, less dense infrastructure for data centers. What's gonna happen is that you're gonna be able to repurpose or build modular data centers, which is much smaller footprint, faster to build or easier to activate. And then in those instances, the ability to actually bring in storage and power, whether it's renewable or not, is much more reasonable.
Marissa Hummon: I see a big disconnect between those who sit on the direct current side of the data center, the actual server side, and those who sit on the AC power flow side, you know, the actual power delivery infrastructure. And it takes both of those pieces working together to create a sustainable and efficient system.
So if I were to say where the biggest problem is with the sustainability conversation, it is that we're moving really, really quickly, but we are still moving in silos, and we really need to look at that system as a whole in order to be able to get the most out of it.
Diana Goovaerts: The takeaway? Watch for three things over the next year. First, how fast behind-the-meter and onsite projects scale. Second, whether operators adopt more honest metrics like carbon per useful work. And third, whether utilization and orchestration improvements reduce the rush to build the next giant campus.
That's all the time we have for now. Make sure you like and subscribe wherever you listen or watch so you get the latest releases from our channel. And stay up to date with news in this space at fierce-network.com. We will see you next time.