Stargate, AI and Energy Demand – EcoTech Insights

Hi everyone. I am pleased today to be joined by David Wilson, who is the CEO of Energy exemplar to discuss energy challenges posed by the 500 billion Stargate Project. It's a massive venture by OpenAI, Oracle, and SoftBank to reshape the data center industry.

David, it's great to have you on. Thank you, Bonnie. It's pleasure to be here.

Well, tell me about this. I'm sure you've been reading all about this venture, and one of the implications, of course is energy use and what does this mean? I mean, it's such a large project.

That's right. So data centers over, um, recent years have got materially larger. And so that's not just in terms of their capabilities to process information and, and store data, but it also considerably larger in the amount of energy they consume.

So if you, if you wind the clock back a, a few years, a data center may have used 30, 40, 50 megawatts of power. And today they're sort of the, the 1, 2, 3, 4, 5 gigawatts of power. So we we're a hundred times bigger than, than just a few years ago.

And the, the latest announcements, um, just take that to the next level, It's really almost hard to imagine the scale of it. Uh, I'd like to learn a little bit more about your background. And I, I detect a slight accent here.

I'm assuming you're not foreign America, but let's hear all about it. I mean, I wanna hear, uh, about exemplar and your role. You, you have a very extensive background in energy, and I think our audience be interested in that.

Sure. So you, you're right. Um, I grew up in Australia, um, and been fortunate to be involved in energy in some form or another.

My whole, um, professional career. I started out designing, um, power plants. I went and worked upstream oil and gas with Schlumberger for a number of years and got to live around the, the world, which was fantastic.

Spent a decade with General Electric, um, running the transmission distribution business in Asia Pacific and a bunch of other, um, fun roles. Spent a few years at, uh, McKinsey and Company doing transformations, again, primarily energy sector via utilities and other things. And that's where I was when, um, I got the opportunity to join Energy Exemplar in 2017.

And, um, so Energy Exemplar has been around a, a while now. It's, um, was founded in 1999 by, um, Dr. Glen Draining.

And it's really, um, a platform that enables companies, uh, organizations to, um, look into the future and understand what's gonna happen, um, under different scenarios in the, in the sector. And so, um, we've really grown that materially. So when I joined, we were 50 people and, um, we're about to pass through, um, 600 in the next couple of days.

Today we're serving 550 organizations in 80 countries around the world. And that's everything from, um, governments and system operators looking at how do they, um, tackle the energy developments, uh, across the next few decades, right down to literally what to turn on or off in the next few minutes. And everything, uh, in between.

That is, uh, a big endeavor. And, uh, speaking of which, uh, we've been talking a lot about data centers just in the news even before Stargate. There amount of energy that's needed and what's anticipated due to ai.

So energy use for data centers, it's already under scrutiny with this big project. Uh, where do you see energy sources, uh, being needed, the, the stress on infrastructure? What is your take on this?

Yeah, So it's, it's fascinating and particularly timely. Um, as, as we, um, have this conversation today with the Deep sea announcement, AI boom, if you will, um, has really accelerated over the last couple of, uh, years. And so, particularly in North America, if you look at it, energy growth has been, um, consistent but quite slow.

And the, the, the reason I say that has been quite balanced between energy efficiency measures because we're getting, um, how devices and industry uses energy coupled with, with, with growth, which is, um, typically around electrification as we move industrial processes to being electrically powered as we move more of our in-home devices to being electric as we move transportation to electric. And those things have almost balanced out with the percentage of tool growth. What we're seeing with the AI boom in the last few years is it's really created a new source of load and it's been really quite dramatic.

And so some of the forecasts say that by 2030 8% of the energy use in the, in the grid will be going to data center consumption from really, um, you know, a percentage point or two a few years ago. So it's really a, a dramatic increase. The, the questions at the, um, at the heart of this is just how much of it is real, right?

And the, the, um, because, and I think, um, been around for a few, uh, tech cycles, but this one feels pretty, um, pretty real. There's material investments going in, there's radical productivity, uh, to be gained from using the ai, um, uh, tools. But then we had the announcement this week of, uh, uh, a new competitor who's managed to, um, release their model that's dramatically more efficient and uses less energy.

So it's a, it's quite an interesting time in the sector and we've seen the stock market and everything whip sword around as everyone tries to make sense of, of where will this, where will this lead? Can you talk about the technology you're using to model different scenarios? Yeah, so the way we look at these things is what we would call a fundamental model.

And so we really look at the, at the whole grid, and we build it up asset by asset. And so really the physics behind every asset, right through to the economics. And just to give a quick example, ar around that, if you think of a, a power station, it will have physics, very much like a car in the, in terms of how, when you turn the key, how quickly does it start, how fast can it go from zero to 50, 50 to a hundred, how much fuel does it use at those different speeds?

And um, how long till the maintenance interval and those things are the physics associated with it. If you know how much the fuel costs and you know what a maintenance event costs, now you've got the sort of the economic model of that car, if you will, and it's, or, or a power plant. And we thing.

And then we connect that to another. The next, um, part of the chain, which will typically be a transmission line, has some physics as well, hot day carrying capacity, cold day carrying capacity. And then we'll connect this out to below that you or demand, which will be industrial or residential and, and the models of those.

And we build these, we take that supply chain, if you will, and you can start to then answer questions like, what's the, if the demand this, what will the cost be or how much, um, low growth can there be before we run out of power from those existing assets? And when you're looking at these, um, at a whole grid, this become very sophisticated. So for an example, when we're looking at the Eastern interconnect in the us, which is the largest, um, energy network in the world, which spans right from the, from the Canadian border right down through to to Florida, we'll be modeling 250,000 objects in that.

And then we'll run scenarios, what happens under different load growth scenarios, what we're talking about here with data center, what happens with different technology scenarios, what will be the cost of batteries or different forms of generation into the future, what happens under different weather scenarios. And so we'll run, uh, thousands of scenarios and we'll typically take with millions of hours of compute to look at all those scenarios. And that gives us, uh, a range of insights into the future.

Your background has given you the agreed ability to kind of look at trends and things, things that have happened before. But when we're looking at Stargate, again, it's such a large project that the size and scale unprecedented. What do you think some of the challenges that they may face in developing this?

Yes, and so when you look at the grid of today, 'cause um, stargates not the only one, it's by by far the largest. But even just the, um, other, um, data centers that are coming in have consumed most of what we call the available sites, right? So visualize that, um, those sort of data centers, uh, span over a thousand acres and they use, um, uh, a modern data center like that will use five 10 gigawatts of power.

It's hard to put that into scale. My background is an Australian, the Australian grid is 75 gigawatts. We've got single data centers that are looking at using more than 10% of that, um, uh, that, that kind of scale.

So then they have to, um, solve the problem of where can you get really good data access? So fiber interconnect that's close to the, the demand good talent. So how do you staff this with tech expertise?

And then how do you cite it with close to, to energy? And given that there's very few sites that are just, um, sitting there, well, probably none now with, um, that sort of supply, how do you, um, either augment the grid, so build transmission lines, connect it to new power developments or co-locate. So they're also looking at how do you build more solar and storage or even gas fire generation as part of that campus so that they can generate more of their, their own energy or get the right balance between being able to use what capacity is available from the grid, plus generate the, um, the, the remainder.

So these are quite complex challenges 'cause you're trying to find all these factors to find the ideal side and their huge capital investments, um, uh, that go with it. Where do you see the input or the growth of renewable energy kind of going hand in hand with the growth of data centers? Yeah, and so around the world, renewables is dominating new builds because the economics are so compelling and by far the cheapest, um, foremost energy there is even without subsidies.

And so if you look in 2023, we're still waiting for 20 core numbers to come out, but there was 500 gigawatts of solar deployed across the world, which is by far the, the vast majority of, of new build, possibly wind and geograph things. The other, um, piece that's sort of been a complication in the past for renewable is that variable supply that, you know, generates when the sun shines and when the wind blows. Other big transformation in the last few years is just how fast, um, the batteries have become cheaper.

It's a little unexpected. It's exceeded, um, everyone's forecast. And so the way that people are looking at, um, uh, handling, this is sort of the term of round the clock renewables, how do you couple solar and wind with sufficient storage, given the demand for power is so high and particularly low carbon power, people are willing to pay a premium to bring those assets back to life, if you will, and, um, and use them.

But that's for a short term. Um, oh, there's only so many of those assets if you or people have to get out and build new things if they want, um, more supply. What do you think, Stargate, the implications of it are for the future for data centers and for energy?

Is it a signal of, of this rapid growth that we're seeing, uh, going forward? What, where do you see all of this? Yes.

Um, well, um, you know, so it's like literally is a fascinating day to today as we, we look at the, the competing models come out and the, the active debate about how efficient can you, um, AI models be? And so really the deep sinks come out and said it, it uses only 5% of the compute resources that, um, some of the other models like open AI and, and that have used today. If that's true, that's actually, that translates directly to energy consumption.

So you'd only need 5% of the energy that was anticipated. We'll wait and see where that shapes out. But regardless, there's a huge, um, future for ai.

I think it's one of the, if you look at the transformative things in human history, industrial revolution, the internet, and, um, AI is certainly in that, um, in that, uh, that bucket. And so the demand will continue to grow. It's just a little hard to put a, um, an exact number on, um, how much energy will come into the future.

But even if you took out, you know, um, uh, the, the bulk of what's forecast, it's still a massive number. So we're gonna see a huge amount of new bill, um, generation, an huge amount of transmission, um, uh, upgrades to move the energy, uh, uh, around. So it's a, it's a fascinating time.

It really is an, uh, Dave Wilson, CEO of e Energy exemplar. It's really a pleasure to have you on and kind of walk us through this exciting time, as you mentioned with AI and the news data centers, and of course, the Stargate Project. Thank you so much for your insights today.

Thank You, Bernie. It's been a pleasure.