This is Tom Rowland's Reese and you're listening to Switched on the BNF podcast.

Today we're discussing BNF's new Energy Outlook, our annual flagship report that represents our long term energy and climate scenarios for the transition to a low carbon economy.

AI data centers have dominated conversations in the clean energy sector of late, with forecasts varying wildly over just how much extra load there set to place on already strained power grids.

As with every hot button topic, comes the risk of hype versus reality.

While data center energy demand is unquestionably growing at speed, it currently represents just one point four percent of global power demand and only around two percent of that figure is actually consumed by AI facilities.

The demand itself is also uneven, with some regions like the US needing significantly large amounts of power than others.

And when it comes to meeting the needs of these energy intensive facilities, can the development of clean power sources even be done at the scale and pace required.

To learn more about the forecast for data center energy demand, I'm joined by b and EF's head of Energy Systems Modeling, Ian Berrman to discuss findings from the twenty twenty five edition of our new Energy Outlook.

BNF plans can find the full report at BNF go on the Bloomberg terminal or on BNF dot com, and if you're not yet a client, you can also download the executive summary at BNEF dot com.

All right, let's get to talking about the impact of data centers on this year's NEO with Ian.

Ian.

Welcome to the podcast.

Thanks Tom.

So Ian has.

Been on this podcast before.

We had a freewheeling discussion around the power system as a whole.

He had this crazy metaphor involving multiple people on like a bike, I mean, as in millions of people all on the same bike like this is not a very relatable metaphor.

But okay, I got more messages on LinkedIn saying what a great episode after that particular one than I've ever had before, So I'm sure this's gonna be a fascinating discussion.

I'm going to talk about NEO.

So Ian tell us about NEO, which is obviously something as well that we hype a lot at BNEF, but so spell out what any means for those of us that are not familiar with it.

NEO is the New Energy Outlook and it's benef's flagship publication.

It's our house view on the future of the energy transition.

We've been publishing it for many, many years now, since before I started, so seven or eight years ago, and in some ways it's the culmination of everything we do because we are looking at the entire energy transition.

We're bringing in work from all our colleagues and all the various different sectors that touch on that into one comprehensive and coordinated house view on the.

Future of energy and the transition.

Awesome, and so I kind of have this view of it as this it's like an orchestra and we're pulling together everything that BNFS done in this coordinated way.

And so where do you sit in this?

AREI the composer, the conductor?

Are you the star violinist?

How do you read?

Why is it that I'm talking to you on this podcast and not someone else?

That's a good question.

I'm I'm trying to figure out where to go with this analogy.

Now, I'm definitely not the conductor that would be my boss.

Do I make the instruments?

Possibly?

Possibly I write the music.

Maybe that's that's the way so I sometimes will play along in an instrument when when and where I'm needed.

But the Energy Systems modeling team, which I head up, our role is to work on the models which go into producing this.

So I guess, I guess we make the instruments that other people play well.

I feel like I mean, Firstly, obviously the people who make the instruments are fundamental to the orchestra, even if they don't get the limelight.

So maybe I'm glad we're having you on this podcast so you can get your flowers.

Finally, I feel like you might be being a little modest because the music that we create with NEO, I think very few people understand it in its completeness more than you do, which is which is why we have you on today.

So Neo, it's this outlook to twenty fifty, a comprehensive look at everything.

And obviously, if we're going to.

Do that every year and just update our view on a twenty five year twenty five year plus forecast, it's not you know, you question why we're doing it, because not that much can Chette have changed or how we see twenty fifty between one year and the next.

So we always do new things to kind of keep it fresh, keep it interesting, have some other insight to say.

So this year, what would you say the focus was?

So this year, which is probably not going to surprise anyone that's listening, the hot topic has been data centers and AI, so we went away.

We developed our own in house short and long term forecasts for electricity demand from data centers and AI.

That's a big feature in the report.

We refreshed our entire base case, which we call the economic Transition scenario, which is not exactly a business as usual scenario.

And it's probably worth unpacking that very briefly because I think economic Transition scenario really reflects what makes I think NEO unique and BNF's approach to producing this type of global analysis unique, And what makes the Economic Transition scenario unique is that we're taking an economics led approach, and we're doing this informed where we can by bottom up modeling, so very detailed and granular modeling.

And when we say we're putting economics in front, that means we're trying to strip out policy, particularly if that that's aspirational, any sort of targets which are non binding, et cetera.

And we're really trying to hone in on the technology and economics story, and that this leaves us with a story where economics are in the driving seat and technology is a story.

Right, it's almost I mean, this is a fair way to describe it.

What we're painting there is a picture of a world that follows what we would consider to be an economically rational ideal without the interference of inconvenient human beings who are either pushing for more low carbon technologies to reflect the challenges or the crisis should say of climate change or other ideologies, or one could even say global dysfunctions that can get in a way of maybe that pure economic vision.

Yeah, exactly.

And policy can push to carbonization faster or slower than what the economics would tell you is least cost.

And as soon as you try and produce like a four car if you will, and you're making judgment calls on what policies are announced, when, how long they last for, etc.

You're really putting your analysis on shaky ground.

So by focusing on that economic store, it's always useful to know what the least cost system is, even if you don't know what politics are going to be.

And I don't think many of us know what politics are going to be at the moment.

We do forecast some sort of human I say, brilliant if I might, in terms of we look, we do take into account cost declines of technology and the sort of technological progress we expect to have happen.

But we yeah, we don't forecast the politics.

We don't forecast either people being able to miraculously get on the same page.

And we don't forecast in which particular way people are dividing among themselves and tearing each other apart, which is obviously impossible to forecast.

So it's kind of like a north star in a certain sense.

North starts a good way to put it, and may I also don't want to give people the impression that it's sort of too cold and rational and is not based on the real world necessarily.

And I think that there's two main reasons why that's not the cases.

First, I mean, from our sector teams, we do have a very good handle on what we think will happen in the next at least at least five years for most sectors.

We've got huge project databases that BNF is very very famous for renewable assets and other types of asset classes, so we know what's coming and that feeds into the model, and it's sort of there's a transition away from what is known and fixed in the short term for the next few years, and then what the model proposes to do after that, and that's sort of a gradual transition and handover in the results.

And then also, not.

All sectors that we model are as rational, as for example, the power system, and so I mean consumer behaviors in important part of how we model our ev uptakes.

So that's just one example of where it's not like completely without taking humans into consideration.

But the focus is the economics.

Ultimately, that's the story we're trying to tell.

And my impression that that's complex enough as it is.

I think that you said something interesting before the podcast.

That is the number one kind of piece of feedback you get from clients about this is that they don't like a particular kind of charts.

Tell me what kind of chart don't they like?

Yeah, So I've been involved in here for a while now, and I mean, sometimes, well maybe this is feels like a couple of years ago now, but sometimes there's not big disturbances in markets, and things tend to tick along, and I feel like people sort of get used to charts and they can go up, they can go down, but they tend to be sort of like quite constant.

They show a direction, they show a direction, which direction plays itself out in a way that you're like, okay, cool, so that's the trend.

And yeah.

I mean one of the charts that I've got the most questions about, or the general type of charts, is where that doesn't happen, where you'll have something that goes up and then it goes down and people immediately latch onto that and they're like, what's happening here?

And I mean, I think that's part of what makes us a unique cause we try to do that bottom up modeling, and I think when you do that, you can see these sort of weird changes of directions occasionally in your results.

I think if you're doing a trend based analysis more top down, you wouldn't see that at all.

You would miss that.

I'm always amazed by client's ability to hone in and identify those charts that change tackicy.

It has come at me with questions.

If you're been a f client, presumably by the fact that you even find our content useful.

You do not fear change, but you might still fear change of change.

And that's what those charts are, is like the change.

We've got some change, and then the change changes, and that's a little bit weird.

Yeah, I mean, it's probably usual to talk about a little bit.

I mean, the main culport of this has been when you oom out, our results are basically saying, if you look at US gas demand, we see a decline until twenty thirty and then growth again, and sort of people immediately what's behind that, And it's not a super complex idea to unpack.

Essentially, what we have today are many parts of the US where we're still building renewables, and renewables are still in the money, and that expansion of renewables continues, and so there's some displacement of natural gas demand in the power system towards twenty thirty.

But then around twenty thirty at least when we've modeled this to the last year or two, I mean, the whole system is moving so equally.

Reinpoints the wrong sort of terminology here, but we sort of reach a penetration level of renewables where the system struggles to move past that and so post twenty thirty, what happens.

We still build renewables, but what you see is the whole system is getting bigger, and renewables role in that system grows more or less in proportion with that system wide growth, which means the other parts of the system also grow in proportion there and you see a return to gas demand growth after twenty thirty.

I would go so far as to say that this is part of the real value add I mean the fact that you have these and clients hone in on them and ask questions.

And maybe it's not because they fear the change that's happened to change.

It might just be that they've actually identified this is where the value really is.

Is Some of these things kind of reminds me back in my early days as an analyst.

My team, as an April Full we wrote a pretend research note and sent it to the editors.

I mean, this is how nerdy we were.

And in it we had a forecast, we had the methodology, and we had developed a methodology called the IACAGR, which stands for Indiscriminate application of a compound annual growth rate.

So obviously just making fun of like how you can make something look with the sort of like the laziest possible methodology, and I think this just really emphasizes that you definitely haven't done the indiscriminate application of a compound annual growth rate.

There's a lot of complexity behind what you've done, and so even with some of the policy interventions going one way or other stripped out, there's still a lot of the igzagging that happens within this as some of these different dynamics play out.

So speaking of different dynamics that up playing out, let's get on to one of the big additions of this report, which was bringing in an analysis of the impact that data centers, principally driven by power demand for AI, how they could impact the energy transition.

So for full context, we did a podcast with our colleagues Natalie Lamandebrata and Helen co and they were talking about specifically maybe the nearer term view and particularly to the US, although I think a lot of those themes apply more broadly, and so that was part of obviously your analysis, but you also looked in the longer term as well.

You extended this out to twenty fifty and obviously the team as well took this global.

So, yeah, what were some of the things that you found doing this analysis?

So, I mean, first off, I'm definitely standing on the shoulders of giants when it comes to the work that Natalie and Helen have done here.

This is a couple of couple of interesting dynamics here is one, because of latency concerns, that forward looking view of where where is data center demand coming from?

It's a lot more globally distributed than you might otherwise imagine.

I mean that's not to downplay the role that the US plays, Like the US is still huge, say China's massive.

Approaching fifty percent of data centers today are in the US.

I think, well, yeah, no, that's that's more or less on the money.

And yeah, and so we see like a large geographic spread, at least towards the longer term.

And there's sort of a dynamic here where I mean, the US is definitely sort of leading the pack here, but other economies who might be a bit later that party will eventually catch up in their demand for general data and AI specific use cases will grow.

Over time too.

There's also I mean there's a seg here as well, because that story of renewed gas demand growth in the US post twenty thirty is partly responsible from data centers as well.

There's definitely a long term growth there data centers and interestingly electric vehicle demand as well.

So that was another sort of high level result that popped out which was quite interesting, is that evs in most places by twenty thirty is still much more significant demand than data centers may potentially not much more significant.

Other places are a bit closer, and I mean US is one of the few exceptions where we think that data center demand will outpace demand from EV's in the short term, but eventually I think EV's will catch up by about twenty thirty.

So where else would we see that trend.

That's a good question.

I think we're probably if you looked at markets where EV growth is a bit behind the curve, where you still have strong demand from data centers.

So I think I haven't seen these charts, but my guesses would be potentially places like Australia or Malaysia where there's a reasonable amount of data centers going in but they're not necessarily particularly strong EVA markets.

It's really interesting.

I mean, actually we just recorded a podcast with Colin Mcherrika, who heads up our advanced Transport analysis, and we were talking about EVO, which is the transport counterpart to NEO, So that's the electric vehicle out look, and we had a long discussion about how the US is going to be moving slower than other markets on electric vehicles.

But you know, it's interesting there's this pattern that some of those markets they might be getting some data centers, so there's going to be demand growth either way.

I remember a really interesting point that you made in a meeting when we were just talking about our data center analysis, because you know, there's this question that is sort of everyone wants the answer to all this data center demand, all this electric vehicle to Is it going to mean more renewables and more gas?

Is it going to mean more emissions?

And I remember you explaining this really nuanced point, which is that you can create a model that has all of those extra things layered in and you can see the difference to the power generation mix.

But it's not necessarily correct to just say that the difference to the mix is what is attributing that to those data centers or electric vehicles?

Can you just explain that?

Am I remembering this right?

You are?

And I mean there's a there's a chart in NEO which I think illustrates this point quite well.

So we with our economic transition scenario, we modeled in an additional scenario sort of behind the scenes, where we stripped out demand from data centers, and so we solved these two nearly identical scenarios.

The only difference is that missing data center demand.

And you can then look at those two sets of results and compare them and start to infer what the effects are of that data center demand.

And this goes into a chart which we have in the report where you see that potentially out to twenty thirty, about two thirds of the additional demand and for data centers is met by fossil fuels.

And you can't stop there though, because that's not the actual message.

That's what the chart shows.

But you sort of have to unpack what that sensitivity means, and I think it goes back to this concept of additionality.

So, first off, we're not saying that two thirds of the electrons that are going to go into data centers are going to come from fossil fueled plants at all.

That's not what we're saying.

I think we're probably saying the opposite I think we're looking at the sort of corporate ppa activity and there's like huge demand there.

So like these data centers at least normally are going to be powered more often than not by renewables, and that's not just a US trend.

And those I mean, the interesting thing about it as a market, and this is very different from electric vehicles, is it's so concentrated with a small number of very huge organizations that have considerable power in the market.

Yeah, and a lot of those organizations are investing heavily in renewables supply, so.

And nuclear and other things as well.

And I mean we also hear stories of them building on site gas generation as well.

So there's a lot of things happening.

But I think this concept of coming back to this concept of additionality.

So we're not saying that two thirds of the power that goes into these things is coming from fossil fuels.

What our analysis shows is that when you add this additional demand to the system, can we meaningfully increase the rate at which we build renewables in the next five to ten years And our analysis tends to indicate that we can't.

And so this is the important idea of additionality it we're going to build these data centers and nominally they're going to be powered by renewables, But that doesn't mean extra renewables built in the world.

It probably just means renewables that were already going to be built somewhere else now being built and attributed to data centers.

Does that means the rest of the system becomes a little bit more more fossil heavy.

And I think that's the way to understand the chart.

That concept of additionality is that we're not building many more renewables because of this additional demand, and the reason being we're sort of near or at the limits of what we think renewable supply chains can manage in the next five to ten years and what the grid could manage in the next five to ten years.

I mean, it's really interesting because I think where we get to here is on that boundary between our economically rational model and something that reflects the what happens in the fog of war.

And I could make the argument that when you say all those renewables would have been built anyway, but now they're going to be built with a PPA for a data center, you could make the argument actually they weren't going to get built anyway, even though it was the economically rational thing from a system point of view, because might just be the market design didn't favor that.

So I and we also have to keep that in mind, is that sometimes these things can act as of forcing mechanism to make maybe the outcome that should have happened anyway on an economic basis happen.

I think you're entirely right, and that's I was careful my language.

I didn't I used infer because the results don't tell you something's definitely going to happen.

It just gives you a hint at what might be what might be the reasons.

And I mean I've also made a similar argument before I said, Look, before we started talking about data centers an AI, we already were running to huge challenges in trying to connect the renewables that we were already going to build over the next five to ten years.

We were running into connection Q issues and some supply chain issues.

And this is before we started worrying about tariffs and whatnot.

So just connecting the amount of renewables that we thought was economically irrational in our modeling over the next few years was already a challenge, and we were already constraining the model there.

And to flip what I just said on his head, like, the amount of resources that the Amazons, the Googles, the Microsoft, and the Metas have in the world are enormous, so they can actually apply the capital and political influence required to in many cases solve some of these issues that might have persisted for longer otherwise.

So we've gone from a situation where we didn't necessarily know how we were going to sort of muster the support we needed in these various areas to build this amount of renewables, and all of a sudden, we've got these huge players like pushing all the right buttons to try and get these things built as quickly as possible.

If we sat here in a year's time, I'm assuming that in the NEO twenty twenty six we will also have inc data centers, because we can't not include them unless in the next year or so everyone decides that actually, AI was such a terrible idea.

And you know, I highly doubt that's going to happen.

And one of the things as you were speaking that I was thinking about was this idea of on site generation and that's been discussed a lot as a potential solution.

I suppose I've got this question all in the wrong order.

I should have said, did we model on site generation this year?

And do you think we will model it next year?

So it's a good question.

So the short answer is no, So all our demand is grid connected.

That doesn't mean to say there aren't constraints that we can apply to the grid in the model, but all demand is grid connected.

I don't think it's a terrible assumption for data centers and AI.

So I mean, we've talked before about this five nines reliability constraints.

So we need ninety nine point nine nine nine percent uptime on many of these facilities, which are highly critical parts of the infrastructure that powers the modern world.

And I mean, when you do the math, ninety nine point nine o nine of all the hours in the year of cross the year gives you five minutes of downtime.

And so it's very hard to design a system that's off grid that only gives you five minutes of downtime with a high chance of likelihood.

I mean, I suppose there's being completely off the grid, which might create those issues.

But oh sorry, we're looking at this from the site.

Yeah, two separate things here.

So when it comes to on site generation, so that in theory is something that our model proposers, so we don't make a judgment call of where that generation has to sit in the grid.

So some of the generation that our model proposers could be on site generation for those facilities.

Okay, So actually, in a way you do factor it in kind of implicitly, but we don't make it explicit exactly.

So everything's grid connected, but we don't make a judgment call on where things have to sit in the grid for the whole thing to make sense.

And whether it's one four hundred megawatt gas plant or four hundred one megawat gas plants exactly.

I mean, I think for the reliability reasons we mentioned, it might be more likely to build more smaller gas plants for one of these things rather than one line one which is harder to maintain it that uptime level.

So it sounds like in a way you have got that covered.

And obviously we're making no promises here, but what do you think might be some of the things that we add in next years NEO.

Then when it comes to data centers.

I mean, I think we just we'll have a year's more data behind us, and we'll have a better view on what the short and then the long term looks like.

And one of my favorite data points this year is if you look at the data center demand that we put in the model, the fraction of that which is AI facilities at the moment, it's about two percent of old data center capacity is AI specific and the rest is just our general run of them, will data centers, backbone of the Internet, the cloud, et cetera.

So that's just two percent of the total.

And then data centers as an entire demand class are only about one point four percent of the entire global power demand.

And so so right now this is like a small fraction of a small fraction.

A small fractions for it.

And I love an analogy, Tom, so indulge me, Andre we go.

We haven't had enough to date.

So the way I like to think about it, it's, right right now, this demand from AIS just a very very small ripple on the horizon we're looking out to see and we can see this very very small ripple, and we really don't know in five or ten years time whether this is just going to be a small wave that sort of peters out or a tsunami.

That washes the power system away.

We really don't know at the moment, but there are some clues because, to continue this analogy, the coastline's not flat.

There are parts of this coastline which extend file into the sea, and we can look at these places today and see the challenges they're facing.

So, I mean, it's not a surprise we're going to talk about PGM now.

Basically the part of the US, one of the grids in the US where we see a hyper concentration of data centers.

Yeah, that's like forty percent of US state centers roughly, I think are in PJM.

Yeah, it's it's it's absolutely massive, And so you can look to those regions now and you can see the challenges they're facing and the struggles they're going through, and you can sort of think about the implications for the wider system further down the path.

So I'm I'm running with your rather doomsday like analogy.

Here, You're on the beach, there's a little ripple.

Some people get up and start running, and other people are like, really, you just need to chill.

There's nothing going on here?

Where are you?

Are you grabbing your towel and making it for the mountains, or are you just a bit calmer waiting to see what will happen, and whatever happens, you're like, I'm pretty confident we can handle this.

So if you look at our forecast compared to other research houses, we're more likely to be the ones staying on the beach ordering another dakery.

So we tend to be.

Or towards while the dakery is are cheap because no one else is around to buy them as well.

Exactly, So we tend to be the more towards the more conservative end of the demand forecast that you see for data centers and AI out there.

But put that in numbers because like our twenty fifty I think we still say data sense will be a pretty significant percentage.

Of demand in twenty fifty of power demand.

Yeah, we're we're going from about one point four percent of global power demand today by twenty thirty that's about three percent, and by twenty thirty five that's four point five percent.

If we keep going out to twenty fifty, it's almost nine percent of all electricity demand.

And to be clear, I mean we're seeing we're assuming a certain amount of electrification generally in kind of all economies, So a bigger percentage of a bigger volume, right, Yeah.

So the system is getting huge and yeah, and I think another thing is that, like originally I said, like some of this demand, there's more, maybe more of a spread across geographies than you might think.

It's not just the US only story, but I mean the US definitely is at the pointy end of the wedge here, and particularly that that PGM region we mentioned.

So, I mean those figures might be in twenty fifty, like eight point seven percent worldwide, but it's it's closer to about twenty five percent in PGM at that point.

Wow.

Okay, it's really interesting.

So it might be the own s beaches have another Dakari and another one.

Pick up your towel and start getting ready for change.

I think this wave is not going to hit all beaches equally.

Yeah, in case it.

Lost a thread of this metaphor, I mean that some regions and some markets will be affected by this more than others.

Just to be clear, Yeah, no, I think that's a good way to put it.

I think it's not panic stations everywhere.

It's also it's not also not an insignificant challenge, Like if we are going to keep building data centers, particularly at the pace that we currently want to build them, there are huge challenges in both in the short term and the long term to become a reality.

And I think part of the challenge of the modeling if you sort of oom out a little bit, and this is maybe more a story for sort of the more developed economies, but we've been doing neo for a while now, and if you look back at electricity demand over the last like ten or so years, I think we basically got out of the habit of having new demand.

We were quite comfortable with demand being quite flat and they're not being demand growth, and we just sort of were quite happy to sit there and twiddle our thumbs.

And if in many regions in particular because of basically energy efficiency gains, partly because of like the shift from industrial to service economies, you even saw falling demand.

And so I mean there's a regional variation here, but like basically, if you ignore evs and data centers, a lot of places, the more developed parts of the world would see falling demand over the next ten or twenty years, and then you're add in evs and it sort of evens out to a bit more stable, and then when you ad data centers on top of that, you're getting growth pretty much everywhere.

Now it's so interesting, And you know, I head up our global power market analysis where we focus a lot on the outlook for prices, and I think that and this was before I was heading this up, but before this wave of demand growth came into town.

The themes that we were talking about was like price cannibalization.

How much is price going to get cannibalized by these renewables, And that is still an important topic, but that is also predicated on this idea of kind of flat or declining demand.

Now you've got demand growth, It's like, yeah, price cannibalization is still a factor, but it might be offset by other factors as well, including the growth of demand.

So suddenly it's a lot more interesting and creates opportunities.

I think this is the key point is a lot of new power capacity will be renewable, and it maybe provides a more optimistic outlook for the economics of that power capacity as well.

So I would say that even if we're saying in our economic modeling that there might be more gas consumed, which from an environmental point of view is not good.

Of course, people in the renewable energy industry should be looking at this as and I think they are looking at this as an opportunity for them to make their mark.

And how it actually plays out in reality, whether it's majority of gas generation meeting that demand or renewables, it kind of depends on how people play their cards and strategize and make the right moves.

And so it's going to be really interesting to see all of that.

But the point is in the power sector, suddenly it's not just about keeping the lights on.

There's opportunity as well.

Yeah, and I mean it is, it's just more dynamic.

Is more exciting when you when you have to accommodate new growth, and I mean first level, it makes my job more interesting.

It's you've got to sort of that double whemmy challenge of like what are the new things we build and where do we canniblize and potentially replace or retire the old assets.

Well, for every new dynamic, that is another kink in one of your charts.

That is more the change is changing, and that is more questions from clients asking for a more detailed explanation of what is going on.

So Ian, thank you very much for joining today.

Nice Thanks Tom, See you on the beach.

See on the beach.

Today's episode of Switched On was produced by Cam Gray with production assistants from Kamala Shelling.

Bloomberg ne EF is a service provided by Bloomberg Finance LP and its affiliates.

This recording does not constitute, nor should it be construed, as investment ad vice, investment recommendations, or a recommendation as to an investment or other strategy.

Bloomberg Annia should not be considered as information sufficient upon which to base an invest decision.

Neither Bloomberg Finance LP nor any of its affiliates makes any representation or warranty as to the accuracy or completeness of the information contained in this recording, and any

Liability as a result of this recording is expressly disclaimed.