Inside the Coming Power Surge: Beacon AI Centers’ Bet on Alberta
North American electricity demand is growing fast, driven by the increasing presence of data centers, as well as other load growth. In Alberta alone, the AESO has reported about 12 GW of requests for load grid connections from data centers.
This week, our guest is Josh Schertzer, Chief Executive Officer at Beacon AI Centers, which recently announced plans to develop up to 4.5 GW of AI data centers in Alberta, representing an investment of up to C$10 billion.
Here are some of the questions that Jackie and Peter asked Josh: Should Albertans be concerned about this substantial load growth and the potential for associated higher electricity prices, given that 4.5 GW would account for nearly a 50% increase in current average provincial demand? Can data centers introduce flexibility by curtailing demand during periods of power shortages? Will Beacon AI Centers rely on grid electricity, or will they build their own generation sources? Do the clients of data centers, such as major IT companies, view natural gas as an acceptable energy source, considering its greenhouse gas emissions? How much cooling water is required, and could freshwater availability become a growth constraint? Lastly, what is the employment impact of the projects, including construction, operations, and spin-off jobs associated with establishing a data center hub?
Content referenced in this podcast:
- Globe and Mail, “Carney to brief premiers on plan to fast-track major nation building projects” (May 29, 2025)
- The Logic, “Calgary firm eyes $10B AI data centre building blitz” (May 15, 2025)
- Duke Nicholas Institute for Energy, Environment & Sustainability, “Rethinking Load Growth: Assessing the potential for integration of large flexible loads in US power systems” (February 2025)
Please review our disclaimer at: https://www.arcenergyinstitute.com/disclaimer/
Check us out on social media:
X (Twitter): @arcenergyinst
LinkedIn: @ARC Energy Research Institute
Subscribe to ARC Energy Ideas Podcast
Apple Podcasts
Amazon Music
Spotify
Episode 287 transcript
Disclosure:
The information and opinions presented in this ARC Energy Ideas podcast are provided for informational purposes only and are subject to the disclaimer link in the show notes.
Announcer:
This is the ARC Energy Ideas podcast, with Peter Tertzakian and Jackie Forrest, exploring trends that influence the energy business.
Jackie Forrest:
Welcome to the Arc Energy Ideas podcast. I’m Jackie Forrest.
Peter Tertzakian:
And I’m Peter Tertzakian. Welcome back. Well, Jackie, you did it. You got that coveted camping spot.
Jackie Forrest:
The BC campsite I’ve been trying to get. And now Peter, I don’t want to jinx it, but it looks like I’m going to actually see Mount Assiniboine, which is very exciting.
Peter Tertzakian:
And I’ve done that.
Jackie Forrest:
Something I’ve wanted to do for a long time.
Peter Tertzakian:
Good, good. Excellent.
Jackie Forrest:
Of course you’ve done that.
Peter Tertzakian:
Not for a long time, but I have done that. Okay, great. What’s in the news?
Jackie Forrest:
Today, we’re recording a little bit early. It’s May 29th and The Globe and Mail has just reported that Mark Carney will be visiting Western Canada. There’s going to be a Premier’s meeting in Saskatchewan on Monday, but we hear there may be some other meetings coming.
Peter Tertzakian:
Yeah, I think there’s some meetings here even over the weekend.
Jackie Forrest:
And we had Gordon Campbell on last week being a bit skeptical about if this government was going to act, but the reporting in The Globe and Mail, and I will put an article to it and by the time this is live, maybe they’ll be a lot more to say on this, but they’re proposing changes to legislation to fast track projects.
There’ll be a list of nation-building projects, and that these will have a much faster approval process and they will adjust regulations to modify that and to relax some of the requirements, and if they are projects of national interest. So more to come on that, but we want to see fast action. They’re talking about potentially passing this by Canada Day.
Peter Tertzakian:
Well, next time we record, we’ll know a lot more. We’ll know the outcome of that first Ministers’ meeting in Saskatchewan, potentially even some hints on what the list is of infrastructure projects that are prioritized. And I do think there’s going to be some legislation that accompanies that. So stay tuned.
Jackie Forrest:
Yes, stay tuned.
Peter Tertzakian:
All right. Well, we talk a lot about the supply side of energy. We talk a lot about politics lately, unavoidably. We talk about policy that emanates out of the politicians and government. I’d like to come back to one of our favorite subjects, which is on the consumption side of energy, and that’s data centers and still the tremendous momentum behind data centers. The tremendous excitement.
I can tell you that I voraciously use AI in my day-to-day productivity. I know you do too, Jackie. So we want to find out more. We want to find out more also because there’s some really fantastical numbers and excitement here in the province of Alberta with respect to the build out of data centers.
So who better to invite than a CEO of a data center company? So today we’re delighted to have a special guest, Josh Schertzer, who is the CEO at Beacon AI Centers based here in Calgary. Welcome, Josh.
Josh Schertzer:
Thanks for having me.
Jackie Forrest:
Okay, Josh. Well, your company wasn’t maybe not known to everybody a few weeks ago, but there was an article, and we’ll put a link to it, that introduced your company and talked about potentially a $10 billion investment here in Alberta, four and a half gigawatts potentially of electricity demand. So that got a lot of people’s attention and we’re really glad you could join us and tell us a bit more about it. So maybe we could start off with just telling us a bit about yourself and the company.
Josh Schertzer:
I’ll go quickly just into my background first for 60 seconds or so just to give some of that and then I’ll dive into Beacon and what we’re up to. So I am just newly appointed CEO at Beacon. I started in the middle of March, so I’m fairly fresh.
I’m coming off of what I like to joke about as just about a 17 and a half year sprint working at a private equity firm called Blackstone where I had a number of roles. I was the chief technology officer of enterprise technology. It was a team of five or six that led early stage investing off the balance sheet on behalf of the firm.
I advise both our portfolio companies and our business units on technology-related strategy and investments. And most recently, I was up in New York for the first 13 years of my career at Blackstone and in 2020, I relocated to Miami to both launch and co-lead that office.
That was a really exciting journey for me. It was pretty much the only place I’ve ever worked other than New York Stock Exchange for a brief period of time. And the story behind me coming to Beacon is Beacon is invested and owned by an investment firm called Nadia Partners that’s chaired by someone named Aidan Kehoe, who’s a dear, dear friend of mine.
I sat on the board of Aidan’s company years ago that he had successfully sold and we always knew after he sold that, that we would end up working together somehow. And credit to Aidan, he’s been trying to figure that out, try to pull me over for years and I could never bring myself to leave in the exciting role I had at Blackstone.
But when Aidan approached me again last summer about what they were working on at Beacon AI Centers, and it really resonated with me just mainly because of a few things. One is my technical background as someone who’s built and operated data centers across my career, someone that has built technology including artificial intelligence technology for quite some time.
And really just on the Blackstone side of the house, it’s very public that Blackstone is a very large, if not the largest data center investor in the world, and so I had this practitioner experience coupled with the exposure I got on the investment side of the house. And so when this opportunity presented itself via Aidan, it excited me quite a bit.
Peter Tertzakian:
Well, your pedigree is amazing and my experience with American PE and American entrepreneurs and CEOs is go big or go home.
Josh Schertzer:
For sure.
Peter Tertzakian:
And so here it’s go big, and come here and $10 billion. So tell us where you are going to place the data centers here in Alberta.
Josh Schertzer:
Yeah, sure. So a few things. So thank you for the support and stuff as it relates to the launch and to the release just a few weeks ago. If I’m being honest, we expected some inbound. It was quite more than we expected in a good way, especially up here in Alberta. If people didn’t know who Beacon was before, they certainly do now, and we’re very happy about that.
We are a data center development company. The problem that we’re trying to solve for is essentially the constraint that the large technology companies, the hyperscalers face as it relates to the crunch, as we call it, for compute.
As they continue to develop and build out artificial intelligence technologies and features and those things get released and as they train models and as these models continue to enhance, there’s just a big strain on compute technology, compute power to continue to power that.
And compute is really just CPUs and GPUs, mainly GPUs, used to crunch that information, but compute really equals power. If you don’t have significant power, you can’t provide that compute. And so it’s really a power constraint at the end of the day.
And it’s no secret that in North America and more broadly globally, power is obviously very scarce and grid power in particular is very, very scarce. And so the credit’s to the team. The company’s founded by a person named Joe Shovlin. And credit to him, we’ve been here in Alberta since 2023 quietly using data and other sources to really try to find where the right places are to locate these centers.
Alberta is a really exciting place when it comes to building data centers because a number of things. It’s obviously got really great access to natural gas. It does have some excess grid capacity that can be used in the short term. It’s got a great workforce, it’s very business friendly, et cetera. So all those things really really matter quite a bit.
And so after, again, some really careful research and decision making over the last two years, we made some investments in a number of sites. The six that you read about in regards to that 10 billion that you just alluded to are the ones that we are committed to. But we do have a few other sites up here as well that are a bit more opportunistic in nature.
And so we’re still trying to explore how to leverage those three sites in the greater Calgary area and three sites in the greater Edmonton area. Of those six sites, five of the six of them are capable of what we call what’s 400 megawatts of power each. And then the sixth one is actually capable of up to, call it between 1.4 and 1.8 gigawatts of power.
And so what does that mean per site? So for the five of the six sites that are capable of 400 megawatts, it’s essentially four buildings, so 100 megawatts per building. And the sixth one can obviously go to four buildings, but that’s a 960-acre parcel, so it can go quite larger to that, assuming that we can provide the power to it.
Peter Tertzakian:
Well, I always feel like we need to put these numbers into perspective. So to me, one gigawatt is the equivalent of a mid-size nuclear power plant.
Josh Schertzer:
That’s correct.
Peter Tertzakian:
And so 400 megawatts is like a very large gas-fired power plant that serves a large fraction of a city.
Josh Schertzer:
We were having a meeting earlier today and somebody actually was also trying to put that into context, and don’t quote me exactly on the numbers here, but that 4.5 gigawatts is basically almost three Calgaries if you talk about the power required to power the entire city.
Jackie Forrest:
Well, and the province is, on average, consuming about 10 gigawatts, although our peak is closer over 12, just to help our audience understand what a big load that is. Now, that load has come with some concerns as well.
Josh Schertzer:
Sure has.
Jackie Forrest:
… because it’s not just your project. In fact, the AESO, which is our electricity system operator has 12 gigawatts of data center load in their queue. You can see your six projects, but there’s others there as well. And if all this demand were constructed, we’re talking about doubling our consumption.
And that doesn’t even include, by the way, these other projects that have been announced that aren’t connecting to the grid, which would add another 10 maybe to that. So how should Albertans look at this? You say we have some excess capacity now that could be easily consumed.
And I know Beacon’s aware that we’re redesigning our electrical system right now because you are actually contributors to some of the feedback to the government on that, and right now it’s not really the best environment to get people investing in generation. So how should we look at that? Or should we be concerned by it?
Josh Schertzer:
I would say a few things. One thing I should clarify, it can be consumed, but I should say we feel it could be safely consumed. Now granted, there’s obviously concerns around some of that, but first off, everything that AESO and the broader regulatory priorities are doing here is the right thing to do.
They’ve got obviously a lot of demand that came within a very short period of time, coupled with, as you mentioned, trying to do things to expand and reinforce the grid more broadly for the future. And so we were just with one of the ministers today who was alluded to something very similar.
And it’s a legitimate thing to focus on and so I think that the steps that the AESO are taking here in terms of ensuring that the reliability of the grid is there, ensuring that the cost for electricity for the broader Albertans doesn’t go up as a result of what’s happening here with this large demand that’s coming in is imperative. And so our hats are off to them for the hard work that they’re doing to do this in such a short period of time.
Our view is this. The great thing about Alberta is that it is an unregulated market. So to some extent, even though there is work and there’s policy and things that need to really be either updated or reinforced, et cetera, the fact of the matter is that the opportunity to expand this grid safely and quickly is pretty remarkable when you compare it to other regions, particularly in the United States where it’s a lot harder to do this.
It does come with things that we need to remain focused on. But that being said, it is the right environment for it, if that makes sense.
Peter Tertzakian:
Yeah.
Josh Schertzer:
Our approach is a hybrid one. So we are running a business here, but one of our biggest values at Beacon is essentially really to make sure that we are really working with the communities in which we’re operating. The broader community in Alberta essentially, everyone really is excited for Alberta to become the next hub for artificial intelligence.
In order to do that, to some extent, you need data centers to do that. The approach we’re taking at Beacon is in order for it to be an AI hub, that means technology companies have to come here, they have to land here, they have to work here. People would obviously get the opportunities to work at these companies, et cetera, but you ultimately have to land the technology companies here for this to effectively become an artificial intelligence hub.
So because we are solving the compute constraint that the hyperscalers, the technology companies are dealing with, we’re representing what they’re looking for.
Peter Tertzakian:
Let’s unpack this a little bit more because we throw this term data center around as if it’s some monolithic building filled with racks of computers with blue flashing LED lights. So is Beacon constructing and running the building like a commercial real estate enterprise? Are you actually running the GPUs, CPUs and the server racks or all of the above, or what are you doing?
Josh Schertzer:
So it is commercial real estate, but I would say it’s fairly specialized and building a commercial office property is nothing like building a data center. Data centers are way more complicated, way more expensive to build. So there’s three stages to building data centers; powered land, which is basically just a large land parcel that’s got close proximity to power, water, fiber and transmission lines and things like that.
Then the next phase of that is what’s called powered shell, which is essentially everything in the powered land phase, but you’re also building the walls, you’re building the cooling, you’re building all those other pieces. Then there’s called built-to-suit or turnkey where you’re actually customizing it fully for a particular customer, hyperscaler.
And then there’s the running of those data centers. Currently, our strategy is to take it to the powered shell level. And the reason that we do that is a couple of things. One is there are obviously other data center operators out there that have done this at scale, and so for us it doesn’t really make too much sense to invest in something that someone else is great at and we prefer to partner with them in that.
The second piece of that is a lot of these large-scale technology companies, these hyperscalers prefer to either run it themselves or use one of their operators. Does that make sense?
Jackie Forrest:
So the Microsofts or the Amazons, they’d prefer to buy the servers and configure them the way they like.
Josh Schertzer:
Yeah. They actually all have their own specs. And so for us, we’re trying to create the most optionality we possibly can for them and not make it for a particular one. Now, if one of them came to us and said, “Look, we want three of these sites,” and we know exactly what you want to do, we would absolutely be open to that too.
Peter Tertzakian:
What is Beacon selling is selling and to whom? Are you selling to Microsoft and to Google?
Josh Schertzer:
Yeah. So we’re definitely going to those large-scale technology companies; Microsofts, Googles, Amazons, Metas, Oracles of the world, basically coming to market to them with a powered shell and then basically saying, “Here’s a powered shell. How much capacity do you guys need and how far do you want to take it? Do you want us to build it all away or do you want to take it to powered shell and go from there?”
Jackie Forrest:
Okay. Now you’ve come to Alberta. That’s great that you’ve scanned all of North America and see this as a great opportunity. But one question I have, one of the criticisms I’ve heard is that although our energy price at this moment is quite low, it’s been volatile, and then we have fairly high transmission costs relative to some other jurisdictions. So how important is electricity price? Obviously you settled on Alberta even with that, so just explain that to us.
Josh Schertzer:
So two things. Well, electricity price is clearly important to us. It’s also important to all of Albertans or any community that you would host this type of stuff. So today, and this is public data coupled with our own math and analytics, but the current price per kilowatt for electricity here in Alberta is quite similar to many of the popular regions in the United States that host data centers. In some areas, it’s cheaper. It’s actually slightly cheaper than Texas at the moment, according to our math.
Jackie Forrest:
Even with the transmission costs and everything in and the distribution?
Josh Schertzer:
At the moment.
Jackie Forrest:
Right. Because we have really low energy cost portion right now.
Josh Schertzer:
Yes. At the moment. But there’s a few things I would say there. One is in order to maintain costs and for operators like us that are going to draw this much power not to create more expensive electricity both for ourselves and for the broader Albertans, there’s got to be a way to do that.
So our approach, it’s really a hybrid approach. So it’d probably be helpful for me to explain how we’re thinking about the power and how we plan to come online and phase it in because I think it’ll clear things up a little bit.
So as I mentioned before, we’re representing customers and their needs and anybody that’s speaking, any data center developer or operator that’s speaking to hyperscalers today know that all the hyperscalers, the names I just mentioned and a long list following that, they are all shopping for what they refer to as called in-service dates, when they can take ownership of the data center and put their stuff in it.
They are shopping for the end of 2027 and beginning of 2028. And so we’re trying to help solve for that in the short term one, so that we can land customers, yes, because we’re a business, but two, to get them into Alberta.
For example, if we can’t land them in here in Alberta where they’re actively excited about, what ends up happening is again, they have to solve this compute challenge for themselves or their evolution of their artificial intelligence solutions will stall because they’re going to hit a brick wall when it comes to compute.
So hence they will essentially go shop in other markets, they will land there, they’ll commit there. And then it’s not to say that Alberta will completely miss the boat, but certainly now that these folks are all committed financially, they probably won’t be able to circle back until 2031, 2032.
So we’re really striving for, how do we land at least some of them here now and then ramp up over time? And so when you think about that as it relates to power and our power strategy, our approach is going to be draw from the grid. And by the way, most of these technology companies, the names we just mentioned, they require at least a strong percentage of grid power.
For the folks that are saying they’re going to only self-generate, especially if they’re a new company, and this is not just my opinion, this is industry knowledge, it’ll be very hard to land them because they have these requirements. And that’s really because they care about reliability, and to some extent, sustainability falls into that as well.
So we are committed to hybrid both grid power and self-generation. That being said, self-generation, if we start today, it takes quite some time to come online. That’s not something that happens in 18 months or two years. That could take four years. I think the earliest we’d probably be able to come online with self-generation is 2029.
And so as a result of that, what we’ve been trying to collaborate with AESO on quite a bit… And to be fair, it’s been a great partnership and they’ve really been listening, as has Minister Glubish, Premier Danielle Smith. Everyone is really trying to listen and understand the problems and, is there an incremental approach that can be taken to get Alberta started?
The incremental approach is to some extent grid power. Now, it doesn’t mean we have to draw all the grid power the second you say we get it, but it’s just a matter of, the only way to land someone here in 2027 is to give them some grid power. So let’s say, for example, we got 800 megawatts to a gig allocated of grid power out of the excess capacity that’s there now.
First off, we wouldn’t even start doing construction or development until 2026. So in reality, 2026, we’re not really drawing much except for construction stuff. And then really what it comes down to is when the hyperscaler lands and puts servers and racks in your data center is when that draw really starts to happen.
So even if we were given a gig today, it’s not really going to get pulled until the end of 2027. Fast-forward, say that happens, we get that, we land a few long-term leases on these hyperscalers via grid power, we know there’s not 4.5 gigawatts of excess grid capacity in the queue, and so self-generation is going to be required. Even if the grid is expanding, it’s still a significant amount.
And then again, to your point before on reliability and making sure the costs remain effective and doable, that self-generation is going to be required. So we definitely intend to do self-generation as well, and our thoughts are that we’re going to do self-generation, but that’s grid connected self-generation. And so the thought behind that is that we can also be contributors to the grid to some extent.
Peter Tertzakian:
So you will fund the development of self-generation-
Josh Schertzer:
Oh, yeah, absolutely.
Peter Tertzakian:
… like a natural gas-fired power plant? So we had a guest, I don’t know when it was, Jackie, several months ago, about the backlog of gas turbines and things like that. So you have to order today if you probably want one now by 2030.
Josh Schertzer:
Hence why I was saying everybody’s going to commit to self-generation, but in reality, what that means due to exactly your point around data center supply chain, it’s just physically as much as our desire would be to do it tomorrow, that’s just not feasible.
Peter Tertzakian:
No. I’m hearing that some of the data center proposals are also co-locating close to natural gas fields because then you can-
Josh Schertzer:
You can draw from them.
Peter Tertzakian:
… they could basically have the fully integrated supply chain from the gas molecules all the way to the power generator all the way to the data center and o-
Josh Schertzer:
To the extent that’s possible.
Peter Tertzakian:
… and only have the grid as a backup. Is that…
Josh Schertzer:
Yeah. To the extent that’s possible, that’s definitely one of the areas we’re exploring. The other one is you do virtual PPAs with some of the power companies as well, or you do it yourself. And so to be honest with you, all of those options are on the table. I think you said 12 or 14 gigawatts that’s in the queue. Some of that are actually these power companies.
So in reality, and respect to all of our peers and everyone that’s working here in Alberta, it’s pretty common knowledge that not everything in that queue is legitimate data center building right now. It’s a lot of them at the power companies trying to clean some of that power as well.
But it’s okay because the expectation’s going to be, and if you speak to some of the folks within the government, within AESO, we know that post allocation coming up in a couple of weeks that some of that is going to not clear itself out, but some deals will probably be done, but once people know what they’re getting and that number’s probably going to come down.
Jackie Forrest:
Right. Maybe some duplication.
Josh Schertzer:
Yeah, there’s certainly some duplication in there.
Jackie Forrest:
Now, Alberta, we have renewables, but we’re primarily a natural gas grid and we have the perception that a lot of these IT companies only want clean energy because they’ve made commitments to reduce their emissions. What has the response been when you talk about using a grid that’s using a lot of natural gas and has emissions?
Josh Schertzer:
It’s a couple of things. So I think people have come to the, call it understanding that in the short term, natural gas is the only way to get them what they need. And so we’ll be doing some renewables too, but the fact of the matter is renewables cannot power a fraction of the amount of capacity that we’re looking to bring online. And so it’s going to have to be, again, a hybrid approach.
If it’s not renewables that can do it all, coal is obviously not an option anymore for a good reason, and then you have natural gas and you have nuclear. Technology companies are also investing directly, the Microsofts of the world are investing directly in the small modular reactors nuclear power. But what comes with that is very, very long lead time, still.
Faster than 20 years ago, but still very long lead times and the cost to do so, and obviously the regulatory hurdles that they’re going to have to jump through to get that built, which again, gets you right back to natural gas is really the answer. And so to answer your question more directly, the way that they’ve been getting comfortable with using natural gas more.
First off, I’d say if you look recently, Microsoft or one of the large technology companies, I don’t want to quote the wrong one, but one of them recently, within the last three weeks actually, it was in the news because they actually pulled back on some other sustainability targets for the next two or three years simply because they are prioritizing continuing to move forward with artificial intelligence over that, and to some extent it’s not possible for them to hit those targets.
So that that’s number one. Number two, and obviously quite important is natural gas coupled with carbon capture. And so if you’re going to say you’re going to use natural gas to do that, do you have a carbon capture plan as well and, what is that? And can you show me how that’s going to work and what your emissions are really going to be?
And so it is a lot of work because you got to really… Based on the power that they’re going to draw because it’s per customer too. People think about it in the context of per land or per piece of land or per plot, but if you have four buildings on a particular land parcel and it’s different clients, it’s actually different targets. So you might have one piece of land and four different sustainability targets. And so we have to work through that complexity
Jackie Forrest:
One grid. Well hey, in the long term Alberta, I think we can do CCS. We already have proven that.
Josh Schertzer:
Absolutely.
Jackie Forrest:
We have the reservoirs, we have the people. It’s more about the economics, but that can also be the part of the future if that’s something the clients demand.
Josh Schertzer:
It is. And to some extent, you also have to think the way that these leases are pretty, not say flexible, but they’re structured differently. And so when you work with the customers and some of these costs are not just bared onto the data center developer, they’re also passed right through to the customers in certain areas based on the targets they want to hit.
Jackie Forrest:
I got to think if you’re willing to back a nuclear plant and pay that much for power, then you might be willing to pay for natural gas with CCS too.
Josh Schertzer:
It’s that coupled with, I think the technology companies, the customers, to some extent, and there’s a limit here, but to some extent, have also acknowledged that they can’t put all of that cost and burden on the data center developers, so there won’t be anybody building data centers anymore.
Peter Tertzakian:
Can you talk about reliability? And I guess they call it five nines, 99.999% uptime because certainly nobody wants to be interrupted during the making of a CAT video. So can you tell us about that?
Josh Schertzer:
Yeah, sure. I think it depends on the type. We call them AI factories. It’s an industry term that’s been used over the last few months. Essentially, it’s data centers that are purpose-built for artificial intelligence, but within that, you still have different categories.
So you have AI model training, so think of ChatGPT and ChatGPT constantly training and becoming more intelligent and providing more intelligent answers. And then you have things like inference and things where it’s doing true reasoning and reference. So there’s two things there.
One is the compute requirements are quite different. Once you start to get into inference, the compute requirements are like nothing the world has ever seen. Training was like that over the last two years, but that was just the beginning. And the simple example of that is one ChatGPT query is basically 10 times more in terms of the draw on the compute side and the power side as a result than a Google search is.
Peter Tertzakian:
But we don’t pay more than a Google search.
Josh Schertzer:
Well, the technology companies do though.
Peter Tertzakian:
But I find, and we’ve had this discussion on podcasts before, is that okay, then you go to the making of say, an AI query to write a paragraph and then it’s a short video and there’s a long video. The power consumption of these things is just astronomical.
Josh Schertzer:
Astronomical.
Peter Tertzakian:
Yet you only still pay a flat, relatively meager monthly fee.
Jackie Forrest:
A lot of them are totally free for individuals anyway.
Josh Schertzer:
For them. I’m sorry, I thought you meant for the…
Peter Tertzakian:
Yeah, correct. My thesis is at some point we’re going to have to pay per use, aren’t we?
Josh Schertzer:
So a couple of things. So one is technology features in general, not so much an AI, AI for sure, but in general when they come out they tend to be free and then they figure out how to monetize after the fact. You see that happening with OpenAI changing their corporate structure to be able to monetize that more. So for sure they give them out for free because that’s the best way you get them hooked. Once you get them hooked and then down the road, essentially they’ll put some version of-
Peter Tertzakian:
The day is coming, in my opinion, where people are going to have to pay for the computing power you use and make a decision about whether you’re going to turn the stove on or whether you’re going to make an AI video.
Josh Schertzer:
I think to some extent, it’s the same thing as how many subscriptions do you feel like paying for when it comes to video streaming or something else? I think people are going to have to be really smart about that. But let me get back to your question on the five nines.
Peter Tertzakian:
Yeah, because the five nines, because each incremental nine past the decimal point costs exponentially more, I would imagine. And so how much do you really need?
Josh Schertzer:
The first thing on five nines, and to your point on customers may like, “You know what? Who cares if ChatGPT goes down? I’ll just wait.” It all depends on the technology and the use case. So what do I mean by that? Training. So training models. So for example, I keep using ChatGPT because that’s the most known.
ChatGPT is constantly learning. So when you are giving it whatever prompt, it’s giving you an answer, you’re then iterating on that answer. You’re saying, “Oh, do this, but do that too, do that too.” As you’re doing that, ChatGPT is continuously learning, and that’s on a per person, per user, per prompt basis.
That training, you can argue, does not require five nines because it’ll just produce an answer with what it has at the moment. And it doesn’t have to keep learning to give you that answer. The answer might be slightly stale, but essentially it could still give you an answer. So training can be paused because training itself is not user-facing. Does that make sense?
Now, when you get into things like inference, let me give you a prime of where it does matter. Think about autonomous vehicles. So if you think about the proximity of those data centers, but also the speed in which they respond to something, that’s not only running computers in the car, it’s talking to artificial intelligence over internet and other things.
And so in certain situations, time does matter. Imagine a data center went down and all of a sudden, autonomous vehicles got funky or something. And I’m just giving you one example. You can think about in the context of airplanes, a number of things that autonomously that if they went down, it would create quite the scare.
Peter Tertzakian:
But isn’t there network redundancy in all of this? If one goes down, the other one’s still up, so it just quickly instantaneously-
Josh Schertzer:
Yeah, but five nines doesn’t just only apply to power, it applies to uptime as a whole. So you’re talking about network connectivity factors into that, power factors into that, all those things factor that. So if the network went down, the power went up, you’re down. If the power goes down and the network stays up, you’re down. So there’s always that balance.
And so again, my point being is, one last thing is outside of artificial intelligence, there’s other types of data center usures, for example, cloud computing, which was obviously a big thing prior to artificial intelligence. That’s essentially every company, every government hosting their stuff on Microsoft and Amazon.
Imagine your banking site went down or something like that. Well, some people might argue that they can wait for it. Others would say the banking systems must stay up at all times, and so that requires five nines. And so five nines was a thing way before artificial intelligence is my point. It’s always been a thing, it’s just becoming more of a thing now.
The difference being now is because of the different phases and use cases for artificial intelligence, companies can be smart about, where do they really need five nines, to your point, and where do they not? And not only can they be smart in that regard, they can then say, “We can do training in a very far remote region and it can turn off and we can lose it for a day and it wouldn’t be the end of the world.”
Jackie Forrest:
Following on that, and there was this recent study by Duke University, there’s been a lot of talk about it. I will put a link to it in the show notes, Rethinking Load Growth: Assessing the Potential for Integration of Large Flexible Loads. Basically the takeaway from the paper is that if these data centers could be a bit more flexible, that there’s a bunch of spare capacity in the system. The issue is most of the time the day, we’re not using nearly the capacity of the system.
Josh Schertzer:
Correct.
Jackie Forrest:
And so they said if just 0.25% of the time in a year, that these loads could be curtailed, then there would be an extra almost 80 gigawatts of new load in the existing system. So do you think that data center owners are going to become more flexible because you say the compute is so important and hard to get?
Josh Schertzer:
I’m happy you asked that because I think there’s probably a perception here that’s actually really good to address. This doesn’t solely fall on the data centers. The data centers are hosting it, but remember what’s in that data center, what’s drawing that power, data center owners like us, we don’t control that or even touch it or even see it for that matter.
So it would be the Amazons, the Microsoft, the Googles, the Oracles of the world. It’s their servers drawing the power, and to some extent, there’s two ways to solve for this. One is on the data center, and I’ll explain that. But the other is even Nvidia, they’ve announced that they’re working to get a lot more intelligent with their GPUs and their chips around, when is it going to draw all the power versus when it’s actually not being used why does it have to draw all that power?
So the chip technology in itself is going to get more intelligent, more sophisticated and say, “If I’m not being asked to do anything at this moment, don’t draw the power right now. Essentially, for lack of a better term, tune yourself down.” When that happens, the data center draws less power, so part of this will fall on them too.
Now, the other part, now, back to the part on the data center providers, this is actually something that’s come up quite a bit. Here within Alberta, the AESO has been very vocal about this for legitimate reasons. What we have proposed and what we’ve actually committed to with AESO is that in such situations, particularly in the summertime when everyone’s blasting their air conditioner or whenever peak demand of electricity is being used, there’s things that can be done.
We can work with AESO and any other organization that would be monitoring and operating this to essentially say, data centers will always have, in addition to five nines… The other thing that contributes to five nines is not self-generation to produce the power but backup generation. So if you were to lose, for example, grid power, the data centers can’t go down back to the five nines situation.
And so what happens there is we have backup generators. And so what we’ve been working with AESO on is saying, “Look, in the summertime if we need to pull 400 megawatts off of the grid because of peak usage, if you have a sophisticated way of effectively telling us, we can start the backup generators up and run it for 24 hours until the peak goes down, and we’d be happy to do that.”
The second thing is once we get to doing our own self-generation, we’ll actually be helping contribute back to the grid because if we’re not using it at that moment, then the grid can have it. We hope that that helps at least not shelve, but at least address some of the concern around peak load usage because it’s a legitimate thing.
Not only Beacon, any data center operator, that last position they can be in is the reason that they caused a brownout or some type of over usage on the grid. You think about it from our perspective, in addition to being good neighbors, if you will, we’re absolutely incented to ensure that this stays up because think about the burden and the perception that would fall on us if we created some type of unreliability as a result of what we’re doing.
Jackie Forrest:
Okay. So another concern you sometimes hear around data centers is their use of fresh water. And in fact, there’s been moratoriums in some locations, not here obviously in Alberta, but in other places because of water use. So how much cooling water is needed? And we have a constraint on water in southern Alberta. Do you perceive that to be an issue?
Josh Schertzer:
We’re in the design phase ourselves now, so it’s a bit hard to say how much water is going to be used. What I will tell you, though, water in general, the six sites that we committed to in Alberta, the solution for water for almost each one of those sites will be different based on what the local communities and municipalities allow for.
And also, you have to factor in things like environmental studies and other things in the surrounding areas before you can really nail down a water source solution. And not every land parcel is the same. It’s not close to the same things, the same proximity to other things like natural habitats, et cetera. So all of that has to really be factored into it.
So there’s a combination of that. Obviously reuse of wastewater is going to be a priority to the extent possible. While it’s not fully there yet, liquid cooling technology for sure helps with that outside of the water. What I would say in Alberta…
And again, we’re still doing math here with respect to our design, but one of the great things about Alberta, not if you’re from Miami like me obviously because it’s a bit cold for me, but one of the great things about Alberta is that it is a dry and pretty cool climate for the majority of the year, I would say.
And so while air can’t be used entirely, it’s definitely not that cold, especially with GPUs producing this amount of heat. It certainly can help and it can reduce the amount of water required because you can do a hybrid approach here using air and water.
Peter Tertzakian:
Surely you can do a closed loop water system?
Josh Schertzer:
Yeah, and you can do that too, but my point-
Jackie Forrest:
It just costs more, right?
Josh Schertzer:
It’s quite expensive. But my point is to say that is Alberta has a little bit more of that option. There’s other areas, like Texas, you can’t. You can’t use the air like that. You have to cool that air no matter what. We don’t have to cool the air. Even if we have to cool the air, we don’t have to cool it as much as if we would’ve to cool it in Texas in the summer.
Jackie Forrest:
So you’ll need it at some times of the year. But for a big part of the year, you don’t have the cooling requirements of a Texan, et cetera?
Josh Schertzer:
We are building AI, GPU stock data centers. They will produce heat. My point is that we have a lot more options here when it comes to solving for that cooling issue than we would do in other regions in North America.
Peter Tertzakian:
Let’s talk about employment because the criticism is that, okay, yeah, you build these giant warehouses and the racks of flashing lights and who knows, in a few years, it’s AI robots walking around doing all the tasks? And so actually there’s a surge of employment for the construction of these things and putting them together. And then after that, it’s just this concrete shell, a artificial brain that’s operating on its own. What do you say to that?
Josh Schertzer:
So I say a few things. Jobs, we think about it in different ways and phases, meaning there’s construction jobs I don’t think will be automated by AI for quite some time. So as it relates to the build out of data centers for the foreseeable future, construction jobs will be there. The second piece of that… And so just for context for us, we are estimating, give or take 1500 construction jobs per each site that we look to develop for those four buildings per site.
Jackie Forrest:
Is that over a couple of years?
Josh Schertzer:
That’s over probably a two-year period, two to three year period. And then subsequent to that, when the buildings are in service or where they go live, we’re estimating between two and 300 in terms of permanent jobs.
And my view, we’ve had third parties actually come in and vet this for us to make sure that we are thinking about this the right way, a lot of that obviously pertains to the size of the buildings and the things that you’re doing in it, but ultimately two to 300 jobs. If you look, Microsoft’s got a 350 megawatt campus somewhere in the US and they’ve got over 300 permanent jobs.
Peter Tertzakian:
So maybe a better question based on experiences that you’ve seen in these Microsoft campuses of already established data centers is what sort of peripheral technology jobs spin out in the immediate geographic orbit of a data center? Do you know what I mean?
Josh Schertzer:
Outside of the business?
Peter Tertzakian:
Like tech startups and things like that. Is that happening or are those still-
Josh Schertzer:
So we’re exploring that too. With respect to tech jobs, I think two things there. There’ll be data center related engineering jobs and the two to 300 number I mentioned. Separate from that, will data centers themselves create more tech jobs? Not on the data centers themselves is what I would say. You’re going to get the typical jobs that you expect, security, HVAC engineers, things like that.
However, when you build data centers like we’re building, hyperscale AI data centers, what happens, and this is proven all over the United States, all over Europe, all over Asia, is when you have data centers here, what happens is companies that also are building AI stuff, think of companies that are doing robotics or self-driving car technologies, things like that, where they need to be in close proximity to the data centers, they follow the data centers.
So the jobs that come outside are not because they’re working for the data center, it’s because they need to be close to the data center to build their technology. And as a result, they tend to come. So we view Alberta as not just a data center AI hub. We view data centers as the, I’m going to say draw, but excuse the pun, but the draw to make it an AI hub.
Because once this is here, yes, you’ll get some technology companies here and they’ll be hiring with those technology companies, et cetera. But more broadly, anybody that’s going to be building new artificial intelligence technology that needs to be close to a data center, and that’s a significant amount, they would move here or start from here as well.
Jackie Forrest:
Okay. And I have another one for you to add to your list, Josh. I’ve been thinking about this.
Josh Schertzer:
Sure.
Jackie Forrest:
If we see all 12 gigawatts, which I don’t think will all happen, but if we did, that would equate to about a BCF per day of natural gas demand if it was all supplied by gas. So think about it, it’s like basically half of LNG Canada phase one. So we’ve been working very hard to get LNG exports out of this country.
Well, the data centers can provide a new demand for our natural gas and we don’t have to build pipelines to get that. And then that provides jobs to the province in terms of drilling wells, royalties, taxes.
Josh Schertzer:
You’re right. Thank you.
Jackie Forrest:
So that’s another benefit.
Josh Schertzer:
That’s an excellent point.
Peter Tertzakian:
Well, we’re running out of time, but I want to ask you a final question. When’s the first shovel going on the ground?
Josh Schertzer:
No later than the first quarter of 2026, but we are striving strongly to get there by the end of 2025.
Peter Tertzakian:
Wonderful.
Josh Schertzer:
But you should definitely expect the, see our first data centers come online towards the end of 2027. January 2028 would be the latest.
Peter Tertzakian:
Well, Josh Schertzer, chief executive officer of Beacon AI Centers, thanks so much for coming in. I want to obviously welcome you to Alberta.
Josh Schertzer:
Thank you. Thanks for having me.
Peter Tertzakian:
And thanks for all you’re doing. It sort of blows my mind in terms of the details that we went into and all the information. It’s been great. All I can say is I think it’s a good time to be an electrician-
Josh Schertzer:
That’s for sure.
Peter Tertzakian:
… out of all this. And I’d say, yeah, thank you for coming in.
Josh Schertzer:
Yeah, thanks for having me. Thanks for the support. And thanks to the broader Albertan community for the love and support they’ve showed us. We couldn’t be more grateful. And now that we’re pivoting into this next phase, it’s our turn to deliver and execute and do what we say we came here to do. So we intend to do that.
Jackie Forrest:
Good. And thank you, Josh, and thanks to our audience. If you enjoyed this podcast, please rate us on the app that you listen to and tell someone else about us.
Announcer:
For more ideas and insights, visit arcenergyinstitute.com.
