The AI Power Puzzle: Grid Flexibility vs. Off Grid Microgrids

The power demands of AI data centers are skyrocketing—but how will we meet them?

In this episode, James sits down with Tyler Norris (Duke University) and Kyle Baranko (Paces) to dig into two influential white papers offering different solutions. One explores off-grid, solar-plus-storage microgrids as a fast, clean option. The other highlights how we can better utilize existing grid capacity by embracing flexible interconnections and load curtailment.

Together, they explore:

• The case for behind-the-meter energy systems
• How hyperscaler reliability demands shape power strategies
• What ERCOT is getting right—and where other markets are lagging
• The regulatory, financial, and technical barriers to scaling solutions fast

If you're working on energy, data infrastructure, or just curious about how AI’s growth will reshape the grid—this one’s for you.

Paces helps developers find and evaluate the sites most suitable for renewable development. Interested in a call with James, CEO @ Paces?

Transcript

00:01.96

James McWalter

Hello, today we're speaking with Tyler Norris, J.B. Duke Fellow and PhD student at Duke University, and our very own Kyle Baranko, Head of Product at Paces. Welcome to the podcast, Tyler and Kyle.

00:13.34

Tyler Norris

Great to be here

00:14.85

Kyle

Thanks, James. Great to be here.

00:16.78

James McWalter

Amazing. Well, the purpose of this podcast, you know I think everyone's obviously looking at what's happening this in this world. There's a lot being written. There's maybe things being written that are not always the most backed up by research and and deep thought.

00:30.29

James McWalter

um But you know we've been fortunate enough at Paces to partner on an interesting white paper with folks at Stripe Climate and Scale Microgrids. and This has gotten a lot of traction. This came out last December. Kyle, myself were the co-writers on that paper.

00:45.37

James McWalter

And then I think even more interestingly and more recent, Tyler was one of the co-authors on this really, really interesting paper from Duke, again, about how to meet data center demand, um but with a bit of a different perspective.

00:56.63

James McWalter

And so we put this podcast together. We actually want to kind of like get into deep. What are the kind of pros and cons of the various approaches we're talking about? The truth is really going to be somewhere in the middle or maybe slightly on one side versus the other.

01:07.04

James McWalter

um And so, yeah, we have these two experts who have very, very particular perspectives. um And so love to hear, ah i guess, a little bit more about you, Tyler, um just yeah a bit of your introduction and your background in the energy system.

01:19.43

Tyler Norris

Yeah, sure thing. I'll keep it quick. But yeah, I've been in the space for 15 years, give or take. i Started actually at the Department of Energy working on tech commercialization programs, um mostly in energy efficiency and energy, but also helped stand up this new office called the Office of Tech Transitions.

01:36.82

Tyler Norris

Then sort of caught the electricity bug, went to S&P, was working on their North American Electricity Outlook for a couple of years, and then kind of got impatient. you know The utility-scale solar industry was taken off, so i went to Cypress Creek, Renewables.

01:48.70

Tyler Norris

Ended up there about five years doing large scale solar and solar development, some distribution scale, and ended up running our Southeast division. But, and you know, spent about, I'd say like 40% of my time on regulatory fare sort of in front of these state utility commissions. and working with consultants and doing all these electricity simulation models to try to open up the markets.

02:10.36

Tyler Norris

And that was the part of the job I love the most. So decided to go for a PhD in electric power systems. So i'm here at Duke University in Durham, North Carolina.

02:19.38

James McWalter

Amazing, what what what a fascinating background. um And you, Kyle?

02:24.43

Kyle

Yeah, so I've been in the industry um for about six years now. ah My background was originally in energy marketing after undergrad, so worked for a marketing agency that got exposure to lots of different technologies under the cleantech umbrella.

02:40.38

Kyle

And over time, my practice area really took me more into demand response, distributed energy resources, and a lot of machine learning type startups working on on power system analytics. so Also got the the power data bug and kind of pivoted into data science from there. So spent time at a startup doing lots of distribution planning, consulting work. So working with utilities and regulators to basically simulate adoption patterns for electric vehicles and battery storage systems and in homes and and um commercial buildings and think through non-mars alternatives, how to mitigate those those upgrades and whatnot, and had a stint as well, working for a demand response aggregator.

03:21.33

Kyle

forecasting ri reliability of demand response assets. And then I felt really pulled to get back into into siting, get close to the geospatial data. And then I met James and Charles about three years ago. And um yeah, been been with Paces ever since in a hybrid data science and and product capacity.

03:38.95

James McWalter

Amazing. um Well, so let's get into it. You know, I'm actually going start with the, I'm call the behind the meter white paper and the Duke white paper for purposes of clarity. um And so starting with the behind the meter ah white paper, Kyle, can you speak to, you know, this white paper that came out in in December?

03:57.21

James McWalter

What were the kind of high level findings of that white paper and the underlying reason why we produced that paper in the first place?

04:03.43

Kyle

Sure. So i'll I'll start with the reasons first, because I think it helps set the scene a bit. So you were approached by a good good friend of Pace's, um Duncan from Scale Microgrids, who was speaking with Stripe.

04:17.84

Kyle

over a variety of topics, but specifically honed in on the topic de jour, which is citing the hyperscaler, basically it's citing hyperscaler data centers and in in wake of in in in light of the AI boom. And so essentially like the big like conversation there was how do we prevent um this from being a lot of off-grid natural gas projects. And so essentially thinking through, well, a lot of what these hyperscalers care about is really, really the time to power and they are going to deploy these assets regardless. And so, you know, one of the things in response to long interconnection queues and essentially connecting those assets to the grid, a lot of difficulties around taking a long time to connect those assets to the transmission network.

05:01.91

Kyle

um And then in light of that, thinking through, okay, well, the next, another option might be, you know, a lot of what you hear about in the news with Amazon and the Three Mile Island partnership. So, you know, we can start mothballed facilities and look for, you know, instances of latent capacity around the transmission network and and get those up and running again.

05:20.18

Kyle

um And then I think a lot of the discussion as well was thinking through, okay, um but what about other types of resources like clean from power and geothermal and new nuclear facilities like SMRs?

05:32.41

Kyle

um But the reality is a lot of those technologies are are a little bit of a ways away um from coming online. um We're looking for, you know, at like solutions that can really be deployed within the next five years.

05:43.75

Kyle

um And then finally, i would say going through going through We're essentially, sorry, I got tied up here, James. um

05:56.20

Kyle

And so we were essentially also, one of the, another option there was, James, can we cut this out actually? Sorry about this.

06:04.05

James McWalter

Yeah, that's fine. yeah um Just reset and then keep going.

06:05.18

Kyle

Okay, cool.

06:06.81

James McWalter

The editor will clean

06:08.49

Kyle

go Um, so yeah, I'll start back with, um, getting into the, the, why we wrote the paper before proposing the solutions and the conclusions. And, um, I think really what this started with was thinking through, there is a huge demand for power by the hyperscalers and how are they going to attain that power and listening on, okay, stack ranking, what are the most likely scenarios for them to attempt to get that power?

06:34.44

Kyle

and talking through, well, how can they actually get that power um quickly and cleanly and working from those principles? So, I mean, the first option to connect, um you know, to bring a bunch of data center capacity online is to expand the grid. So essentially connect to the transmission network. But what we're seeing there is a long interconnection queue, a long wait time.

06:56.16

Kyle

Those upgrades are are very time consuming and expensive. um So thinking through, okay, how do we get something faster then expanding the transmission network. And then another option we kind of like to consider when we were looking around was restarting mothball facilities like Three Mile Island. So basically saying, okay, where is latent capacity around the transmission network that we can restart, come online.

07:16.61

Kyle

um And one of the solutions would essentially be to you know find find those mothballed facilities and empower them up again. The problem with that is there very, very few decommissioned nuclear plants sitting around. So looking for something with like something that could actually serve the total addressable market um of what you know the data center capacity that's being forecasted.

07:37.75

Kyle

um Another solution would be to build off-grid, co-located clean-from-energy resources like geothermal and nuclear facilities. So yeah what we're thinking there is a lot of new clean-from-resources are in the pipeline. You have you know Fervo Energy and a lot of SMR startups, but really what we're looking for is is something a bit more like you know a faster path the power, um and those are unlikely to be built in the next five years, and and there's a lot of technical risk due to the novelty there.

08:08.08

Kyle

um And then additionally, what you could also do is say, okay, instead of finding a mothballed nuclear facilities, what if you found existing utility scale solar and wind projects? um Again, the total addressable market of those systems is fairly low in that a lot of those projects already have like financial contracts and are operating in the market. So that's another kind of limitation in terms of the the size of the opportunity there.

08:31.22

Kyle

And then um another solution is power these data centers with rented portable gas diesel generators. So essentially saying, okay, we're going to get these online very, very quickly with, you know, backup diesel generators.

08:44.66

Kyle

The problem is that the slack of supply and diesel generator markets is actually very, very tight. So this again, kind of limits the TAM. um And then finally build off-grid co-located natural gas projects, which is really like the outcome we were where you're hoping to avoid.

09:00.12

Kyle

um So many groups we spoke to when we were kind of you know thinking through this idea was that this would be the most viable near-term option um given, in light of all the other concerns,

09:10.94

Kyle

um being able to just build a gas plant next to a data center um as a way to get you know firm capacity fast. So when we surveyed that landscape, we were really you know asked the question, why is no one thinking about building solar plus storage plus gas backup microgrids fully off grid right so it's basically taking a system that scale that builds regularly and expanding that up um into a larger system and so what is what we did is we basically um collaborated with with scale and scale ran tons of different you know power simulations to essentially say

09:45.54

Kyle

um over the course of you know a specific site in West Texas, what is the lowest cost of energy um to supply a fully off-grid data center with the various combinations of solar storage and gas backup.

10:00.26

Kyle

And then from there, they were able to essentially conclude um that the the cost of that power from a megawatt hour basis is like at near cost power, near cost parity with a off fully off grid natural gas system um and also cheaper than other clean alternatives from what we consulted to.

10:20.60

Kyle

um Additionally, they're typically faster than new grid connections based on like the interconnection queue timelines. And they're also faster than off grid co-located gas turbines given, you know, the the weight, the lead times to essentially procure those gas, you know, gas turbines for for full natural gas power plants.

10:38.23

Kyle

um And then finally, the um off-grid, we also found from this was like kind of the the PACE's component, um we ran a full simulation with the geospatial data to say um the total addressable market of where you could build these systems, right, is actually very large. So when you're essentially putting down hundreds of megawatts of solar to power, you know, 100 megawatt data center, it's going to take a lot of land. And what we found um was that there was about 1200 gigawatts of data center potential in the the US Southwest alone.

11:06.66

Kyle

So, you know, combining all aspects of the The cost of providing this power, the time to power critically, essentially taking into account procuring equipment, building permitting times, as well as like the total addressable market of where these locations exist.

11:21.00

Kyle

um And it ended up being a very compelling um argument.

11:24.80

James McWalter

And so, yeah, so just to, I guess, to summarize on that, it's you have grid connected, you have folks, but that's not taking five to 10 years, depending on where you can get that supply of load.

11:36.21

James McWalter

You have ah folks going, trying to go very, very large scale behind the meter, primarily gas, um power systems, but that's running into procurement time horizons that are now also reaching into five-year time horizon.

11:47.33

James McWalter

um To meet the demand from a timeframe perspective with net new generation, um you need to to use some combination of solar, gas, and wind, sorry, solar, BESS, and wind, sorry, solar, BESS, and gas, um because those are basically the only things we can build fast. And ah and for the gas kind of constraints from procurement perspective, layering in solar and BESS, start to pencil out from about a timing and a cost perspective.

12:11.19

James McWalter

um Tyler, you had, I think, quite a different ah approach, right? you you know I think you guys modeled um in a very, very interesting way um the existing system and like and a deeper ah you know really, really deep understanding in terms of like where Slack could potentially appear, as well as, you know i think, you did a very, very sophisticated modeling on um you know things like peak load versus average load and those kind of elements.

12:34.83

James McWalter

um Yeah, I'd love to hear, I guess, you know your kind of the overview of of your approach in the Duke paper.

12:43.22

James McWalter

sorry Sorry, Tyler, i you were muted. Sorry,

12:48.82

James McWalter

Tyler, you're still muted. you just hit that mute button? don't know why it's muted.

12:55.61

James McWalter

Oh, weird.

12:58.73

Kyle

That's weird.

12:58.72

James McWalter

my God.

12:59.09

Kyle

Yep.

13:01.26

James McWalter

don't know why that's muted. Oh, weird. Can you just refresh your page, please, Tyler? You mind? Thanks.

13:25.02

Tyler Norris

Can you guys hear me now? Okay,

13:26.24

James McWalter

We can. Great. Thanks for that.

13:27.39

Kyle

yeah

13:29.38

Tyler Norris

interesting. All right, I'll try to keep it brief. I'm gonna start over. Yeah, sure thing. I'll try to keep this relatively brief so we can keep the conversation rolling. but Yeah, so we were motivated by these you know eye-popping figures that you know AI specialized data centers could drive 44%, perhaps up to 44% of US growth of the next five to seven years.

13:54.88

Tyler Norris

That's obviously colliding with resource and supply chain constraints, right? Multi-year timelines for new transformers, now circuit breakers, and now it looks like gas turbines. So it looks impossible to meet some of these forecasts in key markets, at least in terms of the you know the projected peak expansion to accommodate us you know this data center load growth uh but the good news right is that the existing system is just heavily underutilized and so we found looking at know the 22 largest balancing authorities covering 95 of the country's load that the average ah load factor meaning average load over the realized peaks is 53 which means in any given hour

14:33.49

Tyler Norris

On average, half of the system is unused. um Another way to say that is that 90% of hours, 30% of the power system, actually more than 30% of the power system is unused. see actually more than thirty percent of the power system is is unused So the problem here is really on the most extreme peaks, right? When we have heat waves or cold snaps like polar vortexes.

14:55.22

Tyler Norris

And so if a new load can avoid putting more burden on the worst peaks, right? can defer or even mitigate the need for new generation capacity and transmission delivery capacity. And this was kind of an a natural topic for us because I've been As you know, really interested in energy only interconnection service inspired by the example that ERCOT has set, especially in the past five years, achieving rates of interconnection for new generators that far exceed any other market but and and much faster speed interconnect right than the rest of the country.

15:29.73

Tyler Norris

And so turns out, right, the energy only interconnection, if you're not adding more the peak and you're not seeking capacity status or what we call, you know, fully firm status, either as a generator or a load, you can you can interconnect a lot faster.

15:44.99

Tyler Norris

And ERCOT demonstrates that. And, you know, they've recently, you know, developed a service tier that can allow new loads to get online. And they stay, you know, in and approximately two years.

15:55.86

Tyler Norris

So so the findings were that. um If the new load, in this case, so let's just call it data centers, if the new load was willing to curtail 0.5% of its maximum potential uptime in megawatt hours, right, on an annual basis, that across those 22 balancing authorities, so nationally, you could add up to 98 gigawatts of of new load to the existing power system,

16:27.10

Tyler Norris

Or if you drop that to 0.25% of their max uptime, it aggregates the 76 gigawatts. And you know actually, if you accounted for, we discounted the reserve margins. So if you accounted for the the reserve margin, the numbers would be higher.

16:43.08

Tyler Norris

Obviously, you know once you account for transmission constraints and the intertemporal constraints on the generation and load, that's gonna that's going to be a constraint that sort of pulls those numbers in. But but even if it's half of those figures, it's It's still quite substantial. And the other thing I'll just say is, ah you know we thought there'd be a lot of hours where you would require like curtailment of the entire new load. But it turns out that in the vast majority of cases where curtailment is required, it would be partial curtailment.

17:13.34

Tyler Norris

And so actually 90% of the curtailment hours, at least half of the new load is retained. And Just to give a number on that, like at the 0.25% curtailment rate for the new load, it it ends up being 80 to 85 hours of annual curtailment. But again, in 90% of those hours, at least half of the half a new load is retained.

17:36.79

Tyler Norris

um So happy to kind of go through some of that in more detail, but that's kind of the headline.

17:42.73

James McWalter

Yeah, it's that's super interesting. I think the thing then um I'd love to get into as as a contrast ah across the two sets of findings. and One of things, we work directly with folks building these projects, um data center developers and so on.

17:57.75

James McWalter

And there is, I think, a um an expectation, maybe a a demand that we're trying to navigate where you hear this thing of five nines reliability.

18:08.31

James McWalter

ryan um Data center developers, hyperscalers have lived in a world in the earlier, you know, previous to 15, 18 months ago, where they would need financial reliability. That's how you build these data centers.

18:21.29

James McWalter

And they're willing to wait a little bit longer to get the interconnection requirements needed to get to that point. I think in both, you know, in the case of of what you outlined, Tyler, you know, basically you're we're directly kind of proposing them to sacrifice on those five nines.

18:35.55

James McWalter

um And then on the behind the meter approach, what we've heard is that this is this makes sense, but we don't really like it as a hyperscaler. We want to be grid connected for the very same redundancy reasons.

18:46.40

James McWalter

And so even if it begins islanded, where's the pathway for this you know large scale off behind the meter project to eventually be grid connected? And in both cases, I think there's this very interesting kind of operational risk element that the data center is saying, hey,

18:59.87

James McWalter

i'm you know I'm assessing that I need to keep this project up, you know I need to keep this development, this this data center up at Five Nines level. A 0.25% on an aggregate basis, that sounds great, but for my project, can we just make it Five Nines redundant and ignore some of these concerns?

19:16.79

James McWalter

So I guess i starting with you, Tyler, you know are what is what have you seen as like the initial reaction to ah adjusting for that the kind of risk and the output that you mentioned?

19:26.90

Tyler Norris

Yeah, thanks thanks for setting that up. There's a lot to say on this. I mean, the first is that right like curil or curtailment from grid draw, right like reducing a load's draw from the grid doesn't necessarily mean ah reducing those uptime the uptime for the servers, right? Because this this notion of curtailability or flexibility could mean like going to on-site power. So switching to on-site power and there are documented examples of this happening. Of course, you already have the you know the backup ah for the case of emergencies.

20:03.65

Tyler Norris

ah So just just want to want to clarify that you know this arrangement doesn't necessarily require any reduction in uptime. um The other thing to say is that there are other forms of flexibility here that wouldn't necessarily.

20:21.21

Tyler Norris

and so you know one option is to the extent that they can reallocate some of the computational workloads, especially the batchable deferrable training loads ah to another data center data center temporarily that's not located in the region that's experiencing that that contingency. right That's an option Google has been obviously very public in advertising.

20:42.42

Tyler Norris

Their capabilities there, it's not very well quantified, so we don't know to what extent they can do it or how often or sort of at what level. We don't know what extent the yeah other hyper-skillers have invested in that capability, but it is an option.

20:56.03

Tyler Norris

um and And then yeah I think it's also worth just reminding ourselves that the 5.9 supplies to the servers, right the chips, but of course, there's a bunch of other infrastructure around the chips, including the cooling infrastructure.

21:11.15

Tyler Norris

And so, right, the data centers are not running at 100% utilization, right, of they their nameplate capacity. And actually, in some cases, much lower than that. I've actually been shocked by some of the numbers. LBNL and their recent data center energy usage report, they actually put the number out of 50% utilization rate.

21:30.36

Tyler Norris

And EIA and E3, their latest numbers are like in the range of 85%. So this is accounting for right all all the other infrastructure.

21:40.95

Tyler Norris

And so just to give like a concrete example that here in the Southeast right are like periods of highest loss of load expectation or winter mornings. during polar vortexes. And you know when it's 15 to 20 degrees outside, like you're probably not going to be needing to run but cooling infrastructure at 100% utilization rate. right So like you can actually reduce draw from the grid temporarily um in in a way that creates create some of that headroom.

22:08.39

Tyler Norris

So a lot more to say on that, but I'll pause there.

22:15.02

James McWalter

Yep. I guess then, you know, from your perspective, Kyle, any any kind of reactions is the kind of thing that Tyler's saying, um as well as, I guess, you know, kind of navigating the expectations of the developer on the data center side. Because I think, like, thinking about what you're saying there, Tyler, that all makes sense to me.

22:32.89

James McWalter

and One of the kind of maybe unfortunate things though is even at like at the Hyperscalars where they have a pretty tight team, the folks who like understand that operational gap right where you might have only 85% utilization um are often not the folks who are like you know have grown up in a world of 5.9 redundancy from a power system.

22:51.78

James McWalter

So maybe some of this is like basically it's like internal kind of workflow improvements that need to be made. um But I guess, you have Kyle, what ah your reaction is one you've seen.

23:00.57

Kyle

Yeah, for sure. And I think, um yeah, what I like to consider here is just like the operational risk from the lens of the data center developer. And I think, you know, going to the fully off grid case, from what we've heard from a lot of developers, and especially those have been around for the and for a long time in the industry,

23:17.79

Kyle

is that they are so risk averse in terms of really thinking through that five nines that even in their grid interconnections, they're looking for two end redundancy. So even for that data center in a grid connection, they're thinking about, okay, if I have one line come two lines coming in and one of them comes down, come falls out due to a contingency, can that other entire line power my entire operation so that I can continue to operate my my data center at sufficient capacity to pay off the very expensive chips that i I purchased. And so when you kind of talk to a data center a developer and you say, okay, we're gonna bring the system fully off grid, even though you have a solar storage and and gas backup system, I think the kind of reframing of risk to a fully off grid system

23:59.04

Kyle

is is pretty remarkably high and that you're kind of speaking an entirely um different language despite you know moving across. um yeah know we're kind of You're kind of like speaking about a ah totally different dimension of risk, right? You're saying, okay, even if one line goes out from like my grid connection, thinking about the diverse resource mix of the actual grid, um now I'm kind of have my my resource mix from one like on-site system.

24:24.91

Kyle

I think what um is interesting though about the curtailment piece and and operationally how to implement this, um when you think about the connection to that that data center and and what i'm I'm curious to get your take on Tyler from a physical interconnection point and grid flexibility is that The argument really goes that in order to bring a lot of this grid flexibility, you know, new data center demand online more quickly, um you need to be able, you're you're able to curtail around the peak, right?

24:55.79

Kyle

And so I'm curious though, from an interconnection perspective, how you see if that data center demand, um if it's coming still coming fully from the the grid, can you actually like, is there a possibility from a flexible interconnection perspective to size the system according to

25:06.78

Tyler Norris

Thank

25:13.27

Kyle

um like if you're If you're basically going through an interconnection process and you're saying, okay, I have 100 megawatts of data center load, um the actual utility can give me 80 megawatts of firm capacity, like um most of the...

25:30.24

Kyle

can give me 100% of 100 megawatts of capacity most of the year, but only like 80 megawatts, a couple bit of a couple of like a little bit at the time, you still have to size that to the max of the system, right? So I'm curious, like most traditional demand response programs, like they're thinking through implementation wise, you're reducing energy cost. But in this case, like how do you see about that, like the physical grid connection? Can that actually be like reduced? Or would utilities be open to this when when going through an interconnection process?

25:59.44

Tyler Norris

Yeah, so you're you're spot on, right? It's all these, it's steady state study on on peak, right? That's the traditional approach. So you take your existing peak forecast, you load on, right, whatever the nameplate is of the request, and then you subject it to various contingencies and local stress scenarios.

26:18.72

Tyler Norris

And it's much more likely to trigger the need for substantial system upgrades, which are the primary cause of all the interconnection delays, right? um But RICOT takes a fundamentally different approach right for for generators, and they do not allocate any cost of network upgrades beyond the point of interconnection to the generators, and they can just connect freely to the grid.

26:42.45

Tyler Norris

And they're responsible for managing their own basis risk. And that's exactly what we're talking about here. And that's the direction RICOT is going on the load side too. and you know So for the controllable load resource service tier, they've kind of revamped it over the past couple of years to explicitly kind of quantify the straight off of like faster speed to interconnect in exchange for flexibility on peak.

27:05.93

Tyler Norris

And they're now encouraging as many loads as possible to become controllable load resources because it actually ends up offering sort of various other benefits of the system. So i think we like we have to embrace this mind. like just We know this, but energy just like fundamentally different from capacity.

27:25.30

Tyler Norris

And we can connect energy-only resources much, much faster and in much larger volumes. And i think that that is the direction that this is going to need to go if we're if we are in fact going to accommodate as much of this load growth.

27:42.06

Tyler Norris

Now, the thing is, we don't have a lot of jurisdictions that offer that sort of service tier for loads. We have basically like Burkow, we've got PG&E FlexConnect program at distro scale, same in Southern California Edison, and like a couple other concepts popping up.

27:58.91

Tyler Norris

But it does it hasn't really been promulgated. so I'm here, I'm like cheering on, hoping that more data centers will announce definitive intentions to go off grid in order to send the signal to like the last the rest of the transmission providers. Like if we don't get serious about creating these kind of service tiers, like we might see very substantial amounts of grid defection.

28:24.50

Tyler Norris

um so I'm hoping that what you guys have articulated and that value proposition can sort of put some pressure on here to to get some more streamlined grid access.

28:36.37

Kyle

Yeah, absolutely. And I think um to your that kind of gets down a rabbit hole. I was interesting to and interested to get your perspective and as well as like, ERCOT has always been the shining example in terms of connecting large amounts of generation, especially clean generation, um you know,

28:52.57

Kyle

I've, you know, followed the, you beating the drum of ERIS over NRIS in terms of the interconnection processes and the unique facet I think of um like what ERCOT really isn't able to provide is the the lack of capacity in markets and the idea of deliverability and how that's studied.

29:10.07

Kyle

um But also just like thinking through something I experienced through distribution planning in terms of like the, the regulated rate of return that utilities are able to provide. So one of the issues like we would run into um when I was doing some distribution planning work is like not utilities are inherently averse to non-miles or alternatives and and mitigating grid upgrades, um you know, through more flexible means to actually increase that, that load factor. And so like from what, you know, if you're taking about the utilities idea of saying, I want to increase capex and boost the grid, be able to rate base that and, you know,

29:44.57

Kyle

get my return on capital investment. Do you see that like fundamentally at odds with um the ability to increase that load factor? And where are you kind of most optimistic in terms of regular regulatory channels in order to mitigate, you know, the the institutional inertia?

30:05.69

Tyler Norris

Yeah, no, I mean, it's a real challenge. I mean, I think When we're talking about these types of numbers for load forecasts where, i mean, literally it would be physically impossible even with the most aggressive version of permitting reform like anyone could come up with.

30:20.84

Tyler Norris

I mean, it's like AEP. They process five gigawatts of data centers and they had 30 gigawatts more requests, like physical impossibility, right? And of course, not all those are real requests anyway, but um even the ones that are real, like probably physically impossible to to meet them on a commercially meaningful timeframe.

30:36.80

Tyler Norris

So like they're going to get plenty of CapEx opportunity. I don't think that that's really the concern, but it's still legitimate. and funny It's funny you mentioned it because I was actually talking with the representative of a major but major um US utility a couple weeks ago, and ah we were kind of spitballing and I said, you know maybe we should explore sort of you know performance incentive mechanisms type structures that like would give you give you guys some incentive to increase the you know the utilization rate of your existing combined cycle units um so that it's not just a fuel pass-through, so that you don't you don't feel like there's like a 100% trade-off between like new you know new capacity versus increasing utilization.

31:21.60

James McWalter

I think building on some of that, I had a, you know, ah not everyone kind of an interesting conversation with a utility, someone who runs utility in WEC, and they were like, i've gotsh I've got a gigawatt available right now, and can give you, because we just have five very peaky days a year, just because of the nature of of of our local climate.

31:43.04

James McWalter

um We are talking through data centers, and but we are struggling to actually attract on a kind of regional and utility basis the hyperscaler because they're not willing to take on pretty much any risk whatsoever um because they don't have to, right? ah you know You have this kind of situation that's ah occurring right now.

32:01.01

James McWalter

I mean, maybe this is more in the kind of financial set of risks where folks are taking a lot of land bets. um And then, you know, it's kind of a bit of a Wild West, you know, if you and all of AirCash is crazy. You know, the the ring out around some of the other utilities you mentioned, Tyler and Kyle, is very, very crazy in terms of like our customer base right now.

32:20.80

James McWalter

But then you also have ah this kind of like at the very end, if the project is fully de-risked, right, to the kind of requirements that we're discussing, they you know this is investment-grade capital, um and the Hyperscales are just leasing at this point anyways, right? They're just going, hey, we're we're going to lease, we're going to supply it you know our our chip contracts, et cetera.

32:38.90

James McWalter

And so you have this middle stage of capital risk where a lot of the actual block and tackling to set up a project in the kind of, whether it's a large cut behind a meter project or it's something that's around a more flexible load, where a lot of that work has to happen.

32:51.40

James McWalter

But in essence, what we're seeing is that the folks at the end are just like, we don't actually have to make too many trade-offs. Let them, you know, let a lot of folks kind of in the middle stage fight it out from a risk return perspective. and then whatever the best projects that get to us, great, we'll sign off on them.

33:06.36

James McWalter

um That's what we're starting to see more and more because I went to, i went to a hyperscale. was like, Hey, I, there's this utility. They'll give you a gigawatt today. Like there's nobody competing for it. um It's a vertically integrated utility. They can build the generation to like supplement it.

33:20.61

James McWalter

um The CEO, I can put you in I got on the call with the CEO of the utility and they were like, yeah, like it's not in our top 15, you know, areas and you know, it's five days a year. going to love that.

33:33.82

James McWalter

And I was like, okay. And then this is again, very senior person at the Hyperscalar. And so I think like one of again, one of the things is um where risk is sitting across the set of decisions, I think is driving in like overall a set of bad decisions, right?

33:48.30

James McWalter

And i agree with you when you mentioned mentioned earlier, Tyler, we, you know, again, we're supporting Bindometer because like it's a failure of, you know, the, uh, all the other kind of counterparties actually, you know, be able to, you know, both accept and and absorb the load as well as like be bit more flexible in terms of the, you the folks who are generating the load.

34:05.64

James McWalter

Um, And so we want to see some big behind the projects because we think this will also attempt to push some reform. And yeah, and we would love to see ARCOT type model for the rest of the country. um I guess like what are some other risks that you have noticed, you know, ah definitely want to get into where we see these types of projects actually getting implemented and and where, you know, where if you guys have heard of particular projects that want to kind speak to and and what if we think on the 2030 time horizon um which of these approaches will potentially have you know the most amount of potential upside um but yeah any other risks tyler or kyle you want to cover before we get into the kind of outlook of the next five years

34:44.19

Tyler Norris

Kyle, anything else there?

34:46.44

Kyle

Yeah, I think I'll just kind of start with like the natural reaction, I think when we published our paper. So why isn't this this happening? And I think the most common consensus in terms of pushback was there's a lot of off-taker risk. So it's a chicken or the egg problem. Who's going to build a solar farm and have the data center come first, or who's going to spend the capital to de-risk on the power side?

35:09.92

Kyle

um Or where are you going to, like, you're going to build a data center and then bring the power next, right? So it's like kind of the the two components of successively de-risking each side, um I think has been one of the things that's a bit challenging where like as you continue to expend more and more capital and get close, then you arrive at bigger and bigger stage gates.

35:31.48

Kyle

And essentially like a lot of the market as, you know, James, you were referring to it, the hyperscalers, they're just waiting for the finished projects that are, you know, almost a hundred percent de-risk to come to them. Right. So like you see a lot of movement and announcements, um, in terms of like high publicity off-grid stuff, in terms of Stargate, the energy abundance development corporation, I believe that was like this week's news cycle. So you see a lot of like headlines in terms of these projects. And I think surprisingly, we also saw a lot of, you know,

35:59.20

Kyle

off the beaten path types of developers reach out to us or you know to the the group that wrote um you know the off-grid white paper saying like, hey, I have a pitch deck for this and I'm actively getting site control to build this exact thing you're talking about. So there's plenty of movement there, but I think in terms of like the the financial aspects, it's like how far do you have to get in terms of de-risking and procurement and how much risk are you willing need to take to get that set up in order for a colo or a hyperscaler to come and actually you know mutually invest to bring that project to fruition.

36:27.71

Kyle

um So I think like just, you know on the financial side, that's like the the biggest risk from the data from the off grid perspective. um And in practice is, you know, we've mentioned like, you know, that the most common approach seems to be like this, this hybrid um approach where you are targeting these systems and it's always seen as a temporary solution and you're building close to an actual, like you are building with a line of sight to a grid connection at some point to de-risk that offtake um eventually. So you aren't on the hook at the end of that.

36:55.88

Kyle

um And then I do see just last one, like the grid flexibility side, like operationally, um you know, Google has mentioned a lot about load shifting. i do think there might be an element of like risk for new build data centers in terms of, okay, um you know, are you going to like, I've heard a lot of like interesting strategies around,

37:15.11

Kyle

um thinking through buying older chips and having kind of like a load following type of resource from the data center perspective. So you have like low CapEx chips that are being used less frequently. um or As you mentioned, Tyler, like you're kind of experimenting with ways to cut um cooling loads. But I still think for like the broader data center community, there's still a decent amount of operational risk and thinking through, okay, what are the new technologies and ways of operating a data center that would allow us to, you know,

37:39.75

Kyle

introduce flexibility at at the scale that we're seeing. So I think there's kind of like, you know, financial risk on on both sides there.

37:46.97

Tyler Norris

Yeah, so maybe I'll just make a couple of points. You can just X this out right here, just in this transition. um

37:58.55

Tyler Norris

So ah couple of things. I mean, I think there's this very interesting and sort of strange dynamic where the actual utilization rate of many of these data centers is going to be well below 100%. And and yet it's almost like they're actually not getting credit for the flexibility that they will already be actually experiencing.

38:24.58

Tyler Norris

and There's something breaking down there. and I think you sort of mentioned this, James, a little bit, like maybe they're just not connecting dots or it's just those those people, they'll think about ways of like designing right like interconnection service constructs that account for the sub 100 percent utilization rate they're already going to be experiencing.

38:47.81

Tyler Norris

And that's what's funny. Again, it's why like everyone's like, five nines. And it's like, yes, it's five nines po for the chips, not for the whole package. So I just sort of like once you guys actually connect the dots and just figure out how to like take the existing sub 100% utilization rate, you're already going to experience and translate that into a product gives you an advantage, right especially in speed to power.

39:12.23

Tyler Norris

like You guys are going to have a significant comparative advantage over your competitors. You're going to get online faster and maybe you'll win the AI a i race. but hate So, yeah, it's it's very interesting.

39:25.07

Tyler Norris

And in James, the way you articulated like, why is there this sort of breakdown?

39:29.53

James McWalter

Yeah, and i I can't say who it is, but I think there's at least one large player and they're starting to explore extensions to existing to existing projects with, because like you're basically, if you're taking an existing project, you're exactly following what you're saying, Tyler, right? Like, you know, the actual operational uptime of of the of the literal project. And now we're just going to add on an extra 50, 100, 200 megawatts.

39:48.80

James McWalter

you know just add a couple buildings to an existing campus. And what we're seeing with that particular entity is a high degree of openness and flexibility, which we haven't seen from nearly anybody else.

40:00.40

James McWalter

like I think there's there's kind of one player that that might have been mentioned a couple times earlier, who is very, very open to a lot of this. And then that's kind of it. We're hearing, and actually, and part of the reason also that we've heard is that that you know one ah one or two of have a particular, they're not they don't want to build gas Right.

40:19.12

James McWalter

And this is actually also a thing that that we're seeing where you have folks who are trying to do, Hey, we're going to do some gas offsite, like that behind meter. We'll do a gigawatt then we'll, and so now it's a dirty project, but then we're going to reconnect it to the grid.

40:31.41

James McWalter

And now my co-located, data center next to that gas project is now clean project, right? Because now we're just interconnected to. um to the grid and hey, that gas happens to be, you know, now it's like electrons amongst all the other electrons.

40:43.80

James McWalter

So we went from grid connected kind of clean power, sorry, we went from a fully behind the meter gas to grid connected clean power, ah with you know, when literally nothing happened, right? It's like electrons is literally flowing right past the exact same substation and transformer bank.

40:56.67

James McWalter

um So yeah, so i think I think, you know, one of the lessons, and we just came out of a big data center conference, and I think oh the people needing, and I think this is why it's been so powerful, like the conversations that you're having, and and, you know, I think you speaking to elected representatives recently, Tyler.

41:12.39

James McWalter

I think it's like, there isn't really someone who's taken the kind of metal of like, this is the orchestration that's needed to actually meet this demand. What's happening is everyone's not making fairly understandably, like just their immediate incentive decisions.

41:25.51

James McWalter

And that is aggregating into a bit of a mess. And so I think one of the things that we're trying to do at Paces, know you're trying to to your work is like, just be like, Hey, let's, you know, all this stuff is up for demand. All this stuff is new.

41:36.73

James McWalter

What can we do different? um And I guess with that in mind, like what, what do you want to see? And I guess, what are your kind of expectations over the next couple of years? um I guess, starting with you, Tyler, again.

41:47.74

Tyler Norris

Well, I was just going to say this, i I'll never forget and Brian Janis in conversation with Steven Lacey, maybe six months ago. And Steven asked him, Brian, if there was one thing you could do to help us integrate all this new load in a sustainable way, what would it be?

42:04.23

Tyler Norris

And Brian said, point blank, I would force the data centers to be more flexible, which is you know quite a statement, right? Like coming from him and his background,

42:15.55

Tyler Norris

And I think it's it does sort of speak to this theme we're getting at here where you know everyone's acting on the margin, what they perceive to be their own sort of interest. And you end up with this sort of tragedy, the commons type situation. and I think this is the a perfect role right where sort of regulators can come in and nudge the system right in a way that can lead to to better outcomes that can actually be maybe be ah more optimal for most of the actual agents on the margin and certainly for society at large.

42:47.27

Tyler Norris

And You know, look, like that's, um we're starting to see signs of that, right? So we were very surprised Duke Energy came out last week and stated publicly they are now going to start requiring all hyperscale loads above 100 megawatts to participate in demand response.

42:59.92

Tyler Norris

A certain number of hours a year details are unclear, right? We're all like waiting to see the details. We'll see how, you know, how much it sticks. But, and we're hearing from other state regulators that this is something there they're very interested in exploring.

43:14.96

Tyler Norris

So i think it's, you know, it's it's just like, a lot of areas of contract and everyone is used to firm firm firm that's the gold standard and especially those that are like cutting the deals right they'd have a perception like that's the only thing that goes that's the only thing the the financing counterparties will go for And so I think we we almost we kind of have to like nudge towards some contractual innovation to get some quasi-firm structures actually banked.

43:39.14

Tyler Norris

And once it's done and we can see it's working, then and then hopefully it can sort of proliferate. So that's what we're excited to work on and see if we can actually you know get some of the service tiers in place and and operational.

43:53.35

Tyler Norris

And I should say, from a research standpoint, you know we want that it was very much a first order analysis. they about that um You said to really sophisticated the BA. the Actually, the the elegance actually was sort of in the simplicity of the approach, which is the only way we could do it at 22 BAs, right?

44:10.06

Tyler Norris

But where we want to go really second and third order here and do some pretty actual sophisticated modeling in like one or two large markets to run a bunch of other scenarios and sensitivities.

44:24.74

James McWalter

That makes total sense. And hey, if if you need help of picking out the best data center sites and some of those geographies, we would be very happy to to be helpful there. um I guess, like, Kyle, you know, as as you kind of think from here, you know, but again we're seeing a lot of different projects at varying stages.

44:41.97

James McWalter

You know, i think you kind of alluded to this. We had like this influx of inbound folks who, and I remember this funny moment like like a month ago where somebody was like, have you heard of this behind the meter strategy?

44:53.42

James McWalter

ah he like Like whispering it to me. I was like, yeah, we're very familiar with it. Do you guys have the ability to actually um to run this because you need a pretty big balance sheet for the procurement of something else? um I guess, what are you seeing, like Kyle? And how do you kind of think about how this is going to evolve over the next couple of years? You know you you sat in with a meeting, closed-air meeting with some folks on the investment banking side who are like kind of like modeling you know some of these outcomes. And yeah they're mostly looking at chip, right? Like the supply chain and the chip supply from NVIDIA and other providers.

45:24.33

James McWalter

um And they just don't know the power space, but they're like, hey, there's this many watts being deployed. Presumably, that's going to be built out or met. um How do you see it going?

45:33.77

Kyle

Yeah, i think um what I really liked about the Duke paper and in Tyler's work is it really illustrates like like what a really big conclusion, which is the grid is underutilized, right? And I think that that really resonates through thinking through like you know, through the experience doing distribution planning work and really thinking through like, okay, like if this is really as lucrative business, lucrative of a business as we think,

45:58.61

Kyle

which, you know, given the AI race, I, you know, I strongly believe that we're, we're going to have, it's going to, you know truly transcend the economy and very, very, and be very, very important. If that is the case, why aren't we getting more out of our existing grid, right? And I think that the conversations we've, we've heard a lot of entrepreneurs in the space, both on like the financial side, from a business model perspective, as well as technology providers,

46:20.15

Kyle

about thinking through you know i referred to it earlier but leveraging like older compute technology that doesn't have as high of a capex and so you're able to to run it um with lower utilization rates and then even getting creative um as you mentioned with you know There's new startups that are essentially rethinking data center operations and how to peak shave and really you know keep five nines redundancy for the chips and load follow and do some other sophisticated stuff with the the rest of the load that would make up the total data center peak. So I think that given the demand here, there's always gonna be strategies to basically

46:56.99

Kyle

you know generate as much value from these these new AI technologies as possible. um And really what your paper did is like allocated, okay, this is the this is the this is the universe, this is how much opportunity is on the table, let's find ways to to go and make your inroads there.

47:11.11

Kyle

um I'd say at the same token, um what What I'm seeing from like the reaction of the off-grid papers, what the big conclusion there is not the greatest underutilized, but solar solar and battery storage is cheap.

47:23.58

Kyle

And we should also take advantage of that when we're thinking about this. And so I think kind of like the most likely um scenario is... like kind of a, you know, a combination of of both of these approaches where you you do see specific instances where there's like off grid systems were built in mind with the grid tie.

47:39.87

Kyle

um And then you also see a lot of, you know, entrepreneurs thinking through different ways to squeeze the most, um you know, inference and training out of assets that are connected to the the grid and and really leveraging those existing connections and and building around the network.

47:54.82

James McWalter

Absolutely. um We're just about at time. um It's been great. you know I'm really kind of excited to see how this will pan out. you know We're seeing these projects directly. yeah I'm really excited about this a kind of deeper study, Tyler, that you alluded to.

48:06.23

James McWalter

um i guess before we finish off, is there anything I should ask you, Tyler, that I did not?

48:12.51

Tyler Norris

No, but I just want to point out that one of the entrepreneurs that you allude to i had actually written a white paper comparing and contrasting our two papers, which I think is is forthcoming. So it's Andrew at at Critical Loop, and i think he intends to release that the next couple weeks. So it'll be interesting to see his conclusions.

48:35.10

James McWalter

Amazing. and we'll we'll add that into the show notes. Perfect. um Well, Tyler, thank you so much. Kyle, thank you so much. it was a great conversation. And yeah, look, it's go to be exciting time. You all got into this industry. You know, I'm the more recent so entered just in the last three years, but it's been an absolute blast and we're solving big problems and very, very exciting.

48:55.86

Tyler Norris

Thanks so much.

48:57.72

Kyle

Thanks, James. Thanks, Tyler.