It’s no secret that hyperscalers’ behind-the-meter gas bonanza has generated enormous CO2 emissions. But the AI boom is responsible for another, less-publicized emissions issue: upstream and onsite methane leaks that could derail decarbonization efforts.
In this episode, Shayle Kann speaks with Olya Irzak, founder and CEO of Frost Methane Labs. Olya unpacks the full lifecycle of methane emissions in gas-powered data centers, and considers why the actual warming impact of behind-the-meter gas may be double current estimates. Olya says practical, relatively low-cost strategies could cut those emissions in half.
Shayle and Olya also discuss:
- The big-picture data: Why tens of gigawatts of new behind-the-meter gas capacity could generate 380 megatons of CO2 and methane emissions over the next 18 months.
- How methane can leak across the value chain, from wellheads and gathering pipelines to facility leaks and engine slippage
- How methane’s extreme short-term warming potency means off-grid gas emissions could dwarf coal’s impacts over the next 20 years
- Why running gas generators at partial loads increases leaks
- Why low-leak gas procurement, leak detection and repair, and catalyst additions add less than 1% to total data center build costs without causing operational delays
- How big tech’s current emissions reporting relies on averages that overlook actual leaks
Resources
- Catalyst: The unexplored frontier of methane removal
- Catalyst: Why are we still flaring gas?
- Catalyst: Serving data center load with carbon capture
- Catalyst: Can the fossil gas supply handle the AI boom?
- Open Circuit: The off-grid data center fantasy
- Latitude Media: The problem with the US gas bonanza for data centers
- Latitude Media: Senate Democrats probe gas-powered AI data centers
Credits: Hosted by Shayle Kann. Produced and edited by Max Savage Levenson. Original music and engineering by Sean Marquand. Stephen Lacey is our executive editor.
Catalyst is brought to you by GridBeyond. GridBeyond is the world’s leading technology platform for managing distributed and flexible energy resources. Learn more by visiting GridBeyond.com.
Catalyst is brought to you by Antenna Group, the strategic communications and marketing partner behind the biggest names in energy, climate, and infrastructure. For three decades, Antenna has helped breakout companies amplify their stories, build reputation, differentiate from the competition, define new categories and accelerate growth. Learn more at antennagroup.com.
Catalyst is brought to you by EnergyHub. Peak season puts every grid to the test — and the utilities that pass are the ones that built flexible capacity before they needed it. EnergyHub works with more than 230 utilities to coordinate over 2.6 million DERs and more than 3.6 gigawatts of dispatchable flexibility through a single platform designed to perform when it counts most. See what that looks like at EnergyHub.com.
Transcript
Shayle Kann: I’m Shayle Kann. I lead early-stage investments at Energy Impact Partners. Welcome to Catalyst.
So let’s take one thing as a given. Behind-the-meter gas generation for data centers is happening. Interconnection queues run for years, the hyperscalers and frontier labs aren’t waiting, and onsite gas is often the fastest route to firm power at scale, at least today. Undeniably, tens of gigawatts are coming in the next couple of years alone, and who knows how much in the long term.
So it isn’t really a question of whether it’s going to get built. It is a question of how. The emissions story that most people think about for natural gas is about CO2, what happens when you combust the fuel. But an equivalent variable is the methane that actually never gets burned. There are some leaks at the wellhead, some leaks at the generation site, some slips unburned out of the exhaust of the generator depending on the kind of generator. And if you add those up, at least by the math of my guest today, then an onsite gas plant looks actually kind of like a coal plant from a holistic emissions perspective.
In aggregate, we’re talking about adding annual emissions just from these behind-the-meter assets that could look like all of global aviation, for example. But make different choices on sourcing and monitoring, some equipment, some efficiency, and that same plant lands closer to average grid emissions. That’s the same fuel and the same megawatts with roughly a 2x swing in emissions.
So what does it take, and is it realistic? Olya Irzak is the founder and CEO of Frost Methane. She’s mapped out where the methane comes from in an onsite gas data center and which levers actually move it. She’s coming up next.
Shayle Kann: Olya, welcome.
Olya Irzak: Thanks, Shayle.
Shayle Kann: Okay, so I think our listeners probably have some general sense of this, but not nearly as much knowledge as you do about it. So in brief, why should we care about methane?
Olya Irzak: So everybody knows that it’s 28 times more potent than CO2 over the 100-year terms, and 84 times more potent over the 20-year terms. What people don’t know as much about is that today, it accounts for 30% of the warming that we currently experience. Also, it has pretty major air quality issues. So, 850,000 premature deaths every year and 860 million crop loss. And this is because methane is a precursor to ground-level smog.
Shayle Kann: Okay. So suffice it to say, we care about methane. And it just so happens that we, in the United States mostly, but in other countries as well to some extent, are on the precipice of building kind of an enormous number of data centers that are powered by onsite natural gas generation of one kind or another. And you’ve done a bunch of interesting analysis on methane emissions that might come from that, and then how to mitigate them. Let’s start with walking through where those emissions would emanate from. There’s multiple stages. So kind of walk me through end-to-end, where do you get methane emissions from in the full value chain of a behind-the-meter gas-powered data center?
Olya Irzak: Yeah. So, data centers are building this natural gas capacity because it takes too long to interconnect to the grid. And so first, you need to actually extract that natural gas. So let’s say your data center is hypothetically speaking located in Texas, you’re getting your natural gas from the Permian Basin. And so the two sources of leak there are upstream, right, so at the wellhead, and then midstream, the gathering pipes that take it from the wellhead to the processing unit within the basin. So between those two things, 4% of production is lost. So that is a very significant amount of methane that really could be powering more data centers instead of going into the atmosphere.
The long-distance transmission pipelines are mostly fine, right? So that has a negligible amount of leak, so that’s not something that people should be worrying about. Then you have leaks at the actual facility level, right, and then finally there’s what’s called methane slippage. So for anybody that’s using reciprocating engines as their generators, as opposed to like turbines or combined cycle or fuel cells, there you have a very significant amount of methane that slips from the exhaust unburned. So those are the four things that people should be paying particular attention to.
Shayle Kann: Okay, and you said 4% on the extraction leak side, that’s presumably average across the Permian. There’s a pretty wide variability, right?
Olya Irzak: Exactly. So you have, on the upstream side, the best players are probably leaking like 0.2% of their production, and of course the worst are, you know, at least double the average. And then on the gathering pipes, that’s a little bit more tricky because you don’t always control which gathering pipes you’re in, so you have to put together both the upstream and the less leaky gathering pipes within that. And there’s huge basin-to-basin variation as well, right? Like, the Permian Basin, San Joaquin in New Mexico, like the Uinta Basin, those are particularly high leakage. But if you’re, say, getting it from Pennsylvania from the Marcellus, that’s a much, much lower leakage rate for both the wellheads and the gathering pipes. So, basin-to-basin variation and player-to-player variation.
Shayle Kann: And we’ll come back to that in the how do we mitigate it piece, but just to put numbers to each step of the value chain. So, okay, average of 4% across the Permian on the extraction side, orient me numerically on like the facility leaks and the potential methane slip at the generator.
Olya Irzak: Yes. So let’s start with the methane slip in the generator because that dataset is better. For four-stroke lean-burn engine, the EPA average when it’s running at full capacity is 3% of the methane coming in slips out the back unburned. If you’re running it at partial load, that can be significantly, significantly higher, right? I’ve seen measurements up to 12% on that.
On the facility leaks, that is probably the weakest from a data set standpoint because we can’t say that these are the same as utility-scale facility leaks, right? When you are building a gigawatt power plant for utility, you’ve got two turbines and like a second stage, right, all in the same building right next to each other. Some of these data centers that are getting built right now are doing 35 reciprocating engines sprowled across a pretty large area in order to generate that same gigawatt. So that means that the amount of piping and welds and connections that you have goes up quite a bit, and therefore the leaks also go up quite a bit. So there it’s, you know, we can make an estimate, and what we’re seeing from satellite looks not so great for some players.
Shayle Kann: Right. I mean, can you say more about that actually, like how much fidelity is there in all of this data and why?
Olya Irzak: Well, the data for methane slips, like reciprocating engines and turbines and fuel cells and all of that, that is like really quite well understood. There’s newer models, typically the manufacturer will have methane slip as just part of their data sheet at full load. But when you’re running at partial load, that’s where you typically don’t get a characterization from the manufacturer, and that’s where it goes up quite a bit, right? You also see it in the efficiency drop. So that is pretty solid.
The basin leaks, those have been done depending on the specific basin, but there’s been a lot of flyovers, flyover campaigns with methane sensors that, you know, as well as satellite observations, that I think are all pretty good, sometimes they’re a couple of years out of date, but lots of pretty accurate data there.
And facilities, because of availability of shortages of equipment, they’re being built in pretty bespoke ways, so you see a lot of variation there.
Shayle Kann: I’m curious on the extraction leaks side, is the trendline positive or negative? In other words, are we getting better and better over time at minimizing leakage at the wellhead, or is it the opposite direction?
Olya Irzak: Great question. I think the best study I’ve seen on this was from RMI, there was a curious but very logical study that showed that there’s some variation based on basically gas prices, right? So natural gas prices go up, the extractors are willing to spend more money to have people doing more frequent inspections, for example. Prices go down, all of a sudden your expensive person doing inspection isn’t worth the amount of gas that you, the extra gas that you can sell if you prevent a leak. And so there’s a little bit of that correlation that RMI found. But in general, like I think New Mexico had a big improvement recently, but other than that, I think gas prices are the big ones given that potentially EPA regulations did not end up coming in.
Shayle Kann: Interesting. Okay, so there are all these sources of methane leaks in the value chain. Give us a sense, just like at the high level, using averages obviously, how much total emissions might this add given how much new behind-the-meter natural gas capacity is being built or planned?
Olya Irzak: Yeah. So we’ve done the calculation on assuming that 50 gigawatts of new capacity is coming in. The latest reports are showing more like 75. And just for context, California has 40 gigawatts of natural gas generation capacity installed. So data centers are no longer a rounding error on our energy usage, like that world is gone.
So assuming that 50 gigawatts, if we’re assuming the 7.5% of methane ends up in the atmosphere, which is what happens if you have some optimistic assumptions on facility leaks and, you know, the upstream, midstream, and engine slip, in that case, we would add basically 380 megatons of CO2e to US emissions in the next 18 months for those 50 gigawatts to come online. So that’s really quite significant, and a lot of that, or, you know, 40 to 50% of that, is preventable if we tackle the methane issue, right?
Shayle Kann: And this is, of course, not including the CO2 emissions from combustion from the generators themselves?
Olya Irzak: The 380 includes the CO2 emissions.
Shayle Kann: Ah, okay. So how much of it is the CO2? Of that 380, how much is the CO2 from combustion and how much is from the methane upstream?
Olya Irzak: Depending on your assumption, between 50 and 60% is CO2 and between 40 and 50% is methane.
Shayle Kann: So there’s another way to think about it, which is that like if people have been thinking about, okay, how much CO2 are we going to generate because we’re building all this new natural gas capacity, they’ve got roughly half the equation, actually. The actual warming impact is basically double what they’re imagining because there’s an equivalent warming effect amount from all the methane upstream of that.
Olya Irzak: Exactly. It’s double if you use the 100-year equivalency. If you use the 20-year equivalency, it’s much, much worse than that. It ends up being like 724 megatons of CO2e, with at that point the majority being methane. So there’s kind of a big question of like, are we looking at a 100-year time horizon for warming or a 20? If at 20, the methane completely dominates. If 100, then yes, then it’s double. So it might be even worse depending on when you think we will start reducing our emissions.
Shayle Kann: Sticking, I guess, with the 380 megaton number, can you contextualize that a little bit? You said it’s a lot, but like, how much?
Olya Irzak: Fair point. So the preventable part, right, so let’s call it 50%, that’s 15% of US grid emissions is what we can prevent by building these data centers with tackling the methane in mind, right? So that’s like the biggest climate, in the next 18 months, right? It’s not like we’re going to stop building the data centers afterwards. So this is like by far the biggest climate lever that I ever had a chance to pull in my career. I don’t know if you had 15% of electric grid lying on the ground in your career, but I’m like—
Shayle Kann: I’d like to think so, but despite a long career in this space, I don’t think I have. Okay, so you said preventable. Let me talk about what it takes to prevent it, and I guess, you know, it makes sense to probably similarly walk through the value chain, because there’s different things you have to do at each step in the value chain, starting with the upstream wellhead stuff.
Olya Irzak: Exactly. So in the upstream wellheads, like, there are two things you can do. A, you can buy it from a basin that has lower leaks, right, that obviously has a bunch of knock-on effects about where you locate your data center or buying pipeline capacity to pipe the methane from where it’s produced to your data center, that is possible, but, you know, obviously extra pipeline capacity buying that is involved. So that’s one thing you could do.
You could buy from the best player within a given basin, right? So there is a set of certification, let’s say MiQ from RMI is one of them, where the best players, according to their certification, leak like 0.2% of their production, and the additional that those players charge for this like certified low-leak natural gas is pretty low. Now, we can argue whether that’s the best standard and how we can improve that standard and so on, but directionally this is correct, right? Essentially, the fact that there are profitable companies that can leak the methane and not leak the methane, that the inspections roughly end up costing the same or something close to the additional revenue from the gas, means that it’s very easy for the data center companies to tip everybody into the low-leak regime, right? Because we know it’s possible, there’s existence proofs of the good players that do this, and the folks that leak less, these aren’t like small companies, right? Like BP is one of the certified ones, for example, in the Permian Basin, so it’s large supply that is available at the low leak rates.
And the data center people together get to essentially decide who drills the next hole in the Permian Basin, right? And they can make that decision, right, because this is such significant amounts of natural gas that especially if they got together and did a coalition for responsible gas buying or something like that, you’d be able to, I think, shift the entire industry, not even just their direct suppliers, I mean starting with their direct suppliers, to that regime.
Midstream, I think very similarly, right? That there are basins that have much lower leakage on that, so I think just the attention to it and the demand and the slight, maybe slightly higher pricing at least at the beginning would probably shift that pretty easily. We know how to keep pipes intact, we do this over long-distance gas transmission as well, right, it just has different regulations on it.
On the onsite, I think like, I understand why this is difficult, right? You have the hyperscalers that have a subcontractor that has a subcontractor that has a subcontractor by the time somebody welds anything, right? And so this is pretty difficult, and it’s not that all of these people are thinking heavily about methane leaks. And so this is something that needs to go down both in the contracts, and where there needs to be some certification.
And leaks are going to happen. The 35 reciprocating engines to be the equivalent of three across like sprowled over like a mile, right, you do what you can with the quality of construction, but leaks are just going to happen in that kind of a regime, and so having leak detection and repair with really fast response is going to be key, both for the climate and for the bottom line, because this is methane, at that point, you’ve already paid for the methane. The methane that’s leaking at the wellhead, those you haven’t paid for yet. And of course for safety, right? You have a bad well, like that thing is ignitable, you have one smoking colleague and that’s not going to be good, this is pressurized gas. So that’s like, I think, something that people should be pushing a lot harder on as far as requirements for specific kinds of detection and response time.
And then the last thing is the methane slip, right? Pretty large number there. First, if you have availability of equivalent efficiency that isn’t a reciprocating engine, then you just get less slip, like this is just not an issue for fuel cells or turbines or combined cycle.
Shayle Kann: Is it an issue for like aeroderivative turbines, or no?
Olya Irzak: I don’t think so, but I could double check on those, but I don’t think any turbines have this issue. I think they all have much more complete combustion.
And then for reciprocating engines, you know, you have this issue. And so first of all, just don’t run them at partial loads. So I know a lot of folks out of like the best of intentions are trying to match it to solar production, so shut off half of your reciprocating engines instead of running all of them at partial load is something that people can legitimately do, right? So the running at partial loads is surprisingly bad for methane slip.
And then, you know, just like you have a catalytic converter for your car, you can oxidize methane at the exhaust, right? There’s catalysts, there’s regenerative thermal oxidizers, although those would need additional energy input at the kind of percentages that you get, and we can dive into the options for either one of those things, but those are essentially your options for that.
Shayle Kann: Honestly, you know, looking holistically at all of that stuff that you can do, to me, it doesn’t seem like it’s that hard. It definitely doesn’t seem like it’s that expensive relative to anything else that these data centers are spending money on, right? Like it should be doable.
Olya Irzak: Exactly. I don’t think it would add more than a percent to the cost of the data center. And I don’t think, more importantly, the question is, does it increase your time to market, right, like time to FLOPS? And the answer is it doesn’t, right? And all of these things, like even if you can’t do it at that moment, do you leave the hook in to put the catalyst into the exhaust to make it easier as soon as you can afford it or as soon as the technology becomes available or the supply chain constraints are done? So those kind of like hooks in for later, people are just too constrained and are having too much of a race to be thinking about anything kind of like that. And so that’s why a third party like us is really useful to be talking to all of these folks to say, look, here’s the things that if you do it now, it’s going to be a lot cheaper later.
Shayle Kann: Yeah, I mean, I know you’ve been out talking to the hyperscalers about this and preaching the gospel of methane reduction at data centers. I mean, I guess the first question is like, have they been accounting for this historically? Have they been, when they, you know, they do like pretty public emissions accounting, are they accounting for all this methane slippage?
Olya Irzak: Not mostly not correctly. There are certain things like, like people use a lot of country or even a lot of averages that may be really, really underaccounting it in a lot of places. And definitely what we’re seeing from satellite just does not match. So I think there’s been a lot of underaccounting of that by basically using averages. And like to their credit though, it is unlikely that the person that’s doing the accounting has expertise about methane slip. Like most data, like even data center builders, not just the hyperscalers, all the way down, right, the number of people that know what methane slip is is tiny, right? And so I don’t fault them for, you know, not having the technical expertise in the different kinds of ways that you can generate electricity from methane, it’s just like hyper-specific. And so, but yeah, I do think that there will be a lot of accounting that would need to get adjusted, especially once more of this comes online.
Shayle Kann: You know, if you could wave a magic wand, obviously you would have everybody go do all of the things, but to get there, like what’s the interim step in your mind?
Olya Irzak: Yeah, I mean, I think that some of these things are really cheap and easy and like available now, right, and some of this stuff is just like some of the things which have like a money, a safety, and a climate win are like a no-brainer. But really because there’s no methane experts all the way down, it’s really useful to just have a central party set the standards, right, kind of like LEED sets the building ones, and we set the leak standards, right? So I think that like is really useful to basically say, well, what should you be asking of your suppliers, right? And then once you show that you care and that there’s an influence to like who you choose as your supplier and how much you pay them is related to their quantification and reduction plans for their methane, even if they can’t immediately do this because of the kind of race we’re in, I think that gets us a lot of the way in.
Shayle Kann: I mean, you said, you know, even if they can’t do this immediately because of the race that we’re in. Like, there’s an extent to which I don’t want to give them that pass because, as you said, I think, look, as a practical matter, you speed-to-power is the most important thing right now and there’s so much money behind it and so much momentum behind it, you’re like tilting at windmills to propose anything that slows down speed-to-power, but if as you’re saying, none of these things would actually slow a project down, they might add some fraction of a percentage of cost to that project, let’s acknowledge that, but they wouldn’t slow it down, if that is indeed true, then like what’s the excuse?
Olya Irzak: No, I think that’s, I think that’s fair. I think the question is maybe at like some larger scale, like at some point, are we going to run out of like certified low-leak natural gas producers, or at some point do we hit shortages on the catalysts? And the catalyst for the methane slip, like that’s not a solved problem, there are solutions, but they’re, all of these solutions will get a lot better with increased use, right, so there’s definitely scale issues. But I think I’m, in general, I’m with you, but I don’t know what influence we necessarily have on this.
Shayle Kann: I don’t know if you’re willing to name names, but like is anybody out in front on this, any of the hyperscalers? I mean, historically if you look at like who’s been out in front on, you know, early renewable PPAs or 24/7 clean power, like usually it was Google followed by Microsoft followed by Meta and AWS behind them. Is it kind of a similar dynamic here, or is nobody out in front at this point?
Olya Irzak: I think I would be breaking some NDAs if I were to say that, but there are definitely, there is a big difference in the players.
Shayle Kann: Yeah.
Olya Irzak: But, yeah, sorry.
Shayle Kann: That’s okay.
Olya Irzak: But there’s, it’s definitely, it’s been very interesting to see the internal organizations of the various companies, right, like who has how many sustainability people, whether having more sustainability people means that you can move faster on this or not, like all of these things have been, I think I understand a little bit more what management consultants do now after this experience. The internal structures and its influence on who can move how fast and on what particular topics has been fascinating.
Shayle Kann: I guess to close out, I mean, you know, you said 380 megatons plus or minus, of which half is avoidable through this set of measures, that leaves half, which is still 190 megatons, still a lot. What do you do about the other half?
Olya Irzak: What a great question. I mean, really there’s only two things that you can do, right? Like, you can capture the CO2 onsite. I did the math on that, that would take about 15% of their energy, and we know that energy that isn’t going to the FLOPS is the one thing currently people aren’t willing to give up, so I think that’s going to be hard. The other thing you could do is pay to take out or prevent an equivalent amount of emissions somewhere else, right, so this is offsetting. And this is something that could be potentially done in the interim of can you offset what you can’t reduce.
You know, of course, there is can you combine it with more renewables, I think a lot of people are trying that. You know, can you combine it with the grid under certain circumstances, you know, that obviously still runs into the interconnection queue issues, which is why they built the gas in the first place. But I think using the gas less and less and less, and then offsetting, and then maybe eventually capturing when somehow the energy becomes available and you’re still using those, that might be another issue.
The other thing that I would say is planning a little bit for the future in case these things continue to be around. So for example, none of them are planning to add a second stage later. And as far as getting 50% more energy from exactly the same amount of gas, like that reduces your emissions dramatically while also helping a lot on the energy availability and the money. Those kinds of things, like people are now building structures which do not allow you to add a second stage later when those second stages become available, so I think that’s like one of the biggest mistakes we’re seeing people make.
And I think the one other thing like I’ll say is I think if there was a massive shift for the data center people to care about this problem, I think the benefits would go beyond the data centers themselves. Right, if you can shift the entire oil and gas industry and eventually as a result of that shift make more cheaper methane leak detection and response systems available and more deployed at scale, right, that could be good for any usage of natural gas, not just the data centers, and I think similarly the catalyst for the slip, on the one hand, you know, we want to do the, like data centers are the ones that have the margins to potentially pay that extra percent. But over time you can imagine that that’s, could be applicable to any reciprocating engines, just like we need to have a catalytic converter on our car, this could now become a standard, a standard thing that you just buy and bolt on as opposed to have to like design a little bit more bespoke for each case. So I think that it could move, like it could develop technologies that would move other industries into a better place because now they’re available without having to spend so much of your, so much extra. Not to say that it’s a lot in data center terms, but it is a lot for a lot of other industries that wouldn’t be able to foot some of the scale-up costs.
Shayle Kann: Well, Olya, this was really informative, and I also very much appreciate that you are doing this work. Like, I think I had an intuitive sense, but I had not seen prior to seeing with the work that you’ve been doing anything quantitative on just how much methane we are soon to emit in all likelihood, barring some change because of all this behind-the-meter gas. So thank you for that.
Olya Irzak: Thanks, Shayle. This was super fun.
Shayle Kann: Olya Irzak is the founder and CEO of Frost Methane. This show is a production of Latitude Media. You can head over to latitudemedia.com for links to today’s topics. This episode is produced by Max Savage Levenson. Mixing and theme song by Sean Marquand. Anne Bailey edits the video version of the show. Stephen Lacey is our executive editor. All of our episodes are on YouTube, subscribe to Latitude Media for episodes of this show and Open Circuit. You can also find the audio version of the show anywhere you get your podcasts.
I’m Shayle Kann, and this is Catalyst.
