← The HC Commodities Podcast
The HC Commodities Podcast
Energy & AI with the IEA’s Thomas Spencer and Siddarth Singh
00:52:00
Today, we return to energy and AI. What will AI mean in terms of global energy consumption up to 2030 and beyond? And could that be offset by efficiencies and transformations generated by AI in our daily lives? Could AI even accelerating the energy transition? To do that, we are discussing International Energy Agency's latest paper - Energy and AI (https://www.iea.org/reports/energy-and-ai). A comprehensive review and deep dive in AI’s impacts on our world energy map. Our guests are the lead authors, Siddharth Singh and Thomas Spencer. Both are also part of the team that produce the World Energy Outlook annually.
Loading summary
Transcript56 lines
- [00:00]
A
Foreign.
- [00:06]
B
Welcome to the HC Commodities Podcast, a podcast dedicated to the commodities sector and the people within it. I'm your host, Paul Chapman. This podcast is produced by HC Group, a global search firm dedicated to the commodities sector. Today we are talking energy and AI. What will AI mean in terms of consumption and what will AI deliver in terms of efficiency and even accelerating the energy transition. To do that, we are discussing International Energy Agency's latest paper, Energy and AI, published in April 2025. I'll put links in the show notes and our guests are the lead authors, Siddharth Singh and Thomas Spencer. Both are also part of the team that produce the World Energy Outlook annually. As always, you can really support the show by leaving a positive review on the platform you're listening on. And as always, I hope you enjoy the episode. Sid Thomas, welcome to the show.
- [01:08]
A
Great to be here. Thanks for the invite.
- [01:10]
C
Thanks for having me.
- [01:12]
B
So we're talking energy and AI. We're talking around your recently published paper, the IEA paper. We'll put links to it in the show notes, which is essentially a holistic look about how AI is going to both change energy consumption as well as impact efficiency. And you know, I guess this is quite a well worn trope at the moment. We're going to dig into which is this sort of exponential requirement for power in data centers that are used to power this AI. We're going to look at that. Are those sort of assumptions and is that sort of public narrative true? But also there's a much broader world of AI which we're going to get onto, which is also going to impact energy consumption. So great paper, guys. Maybe we can start with you, Thomas, and just, can you just give us some sense of, in terms of AI, where are we right now in terms of market capitalization, the amount of investment that's going into it, and what impact we've already seen on energy requirements.
- [02:10]
A
It's pretty clear that we're in a period where AI is advancing very rapidly and so we're seeing new announcements almost every day and it's quite difficult to keep track of everything that's going on. But I think the big picture is that AI is rapidly becoming more capable and is starting to be used in a whole variety of economic activities and also in, in our daily lives. If you look at, you know, investment in AI, maybe the best proxy is sort of investment in data centers, because most of the investment that is happening today in data centers is being driven by expectations for future electricity demand coming from AI. So we're looking at about US$500 billion that is going to be invested in new data center projects this year. That's obviously a huge amount of capital if you look at the market capitalization of AI related firms. So prior to the market downturn that happened in the beginning Of April, about 60% of the market capitalization increase that took place since the launch of ChatGPT was from firms that had a value proposition related to AI. And that's, that's a huge amount of market capital growth that has come from those firms. It was almost US$12 trillion. So things have come down a little bit since then obviously. But it is a sort of, you know, testament to expectations that AI is going to transform our economies and bring a lot of economic value.
- [03:49]
B
We had Adam Myrick on sort of talking about the sort of the micro level, the individual data center and how it consumes energy and how it's constructed and some of the choices around locating. But I found it fascinating reports that have actually the breakdown of data centers by country and then can you give us some sense of their energy consumption where that stands today?
- [04:10]
A
In terms of the breakdown by country you have about slightly more than 40% of installed data center capacity globally is in the United States. So it is by far and away the market leader. It's almost double the next market which is China with about 25%. And then you have Europe with about 15% and the rest of the world accounts for the rest. So there's quite a strong market concentration in a few leading economies, China, the United States and Europe. In terms of energy consumption, things are split roughly the same way. So global data centers account for about 415 terawatt hours of global electricity consumption last year and about 190 of that was, was in the United States. So that's how things break down as of today.
- [05:04]
B
Great. And Sid, can you put that 415 terawatt hours in sort of context of global energy consumption? And then can we start and this is, you know, where the hard bit starts, which is what are kind of the, you know, what are your forecasts on how this will look over the next decade plus and then we can dig into some of the uncertainties around that which you know this is a key bet people are making at the moment about how much power is going to be needed to satisfy these data centers. But there's significant uncertainty either side we'll get into. But where, where is sort of the forecast tending to at the moment? And also can you contextualize that consumption as a sort of percentage of global Energy.
- [05:44]
C
Oh, sure, of course, Bob. So 415 terawatt hours of electricity consumption by data centers globally as of 2024, which is about 1.5% of total electricity consumption globally, which is quite small. But this is also one of the sectors that's among the fastest growing, but it is not the single most fastest growing sector in any case. So we have, for example, industry, electric, transportation, so EVs and so on. We have appliances in particular, air conditioners and those related to heat pumps, for example. So these are some of the sectors that are seeing a faster growth in terms of electricity consumption over the course of this decade. So we find that electricity consumption from data centers are expected to double by the end of this decade to reach about 950 terawatt hours. So by 2030, more electricity will be consumed by data centers globally than the whole country of Japan today. So we are effectively going to reach Japan in terms of consumption. But it's important to note here that of course, not all of this consumption from data centers is artificial intelligence. So we actually tried to break it down a little bit, and we found that component of data centers that often relates to AI related demand. So these are accelerated servers, those types of computation and equipment that is generally associated with the artificial intelligence, but not entirely. That component is expected to actually quadruple. So it rose fourfold between now to 2030. So indeed, the biggest reason for this growth of data center electricity consumption is coming from AI related accelerated servers. And which is why we have been trying to break down this growth a little bit more, which is why we're seeing, for example, data centers themselves, which a hyperscaler data center today may consume as much electricity as 100,000 households, but the largest under construction could consume as much as 2 million households. So really, the size of data centers is growing, and therefore its role in the electricity system as a whole is also becoming more prominent.
- [08:02]
B
Yeah, and I guess there's also that backdrop of the story when you look at a global picture in your report. It's very much, obviously the developing world, greatest increases in power consumption. Electricity consumption will come in things like air conditioning, transportation, just those communities becoming more online and middle class. This is a really significant story, as you point out, though, for the developed world, where probably without data centers, there's not much of a case for electricity growth, demand growth. And actually we had a long period of demand stagnation that's suddenly, you know, we're seeing power demand grow again at the same time. And a story relatively true across the Western developed world is A lot of its power infrastructure is 50, 70 years old and all in need of, of upgrading as well. Can you, can you just lean into that story? Why is this such a significant story for the developed world?
- [08:56]
C
Indeed, I think, you know, it gets really interesting when you start looking at individual countries because although, you know, as we discussed, at a global level, it's only 1.5% and even after it grows, it does not, it still remains around the 5% mark. So why then is this such a big story? It's a big story because I think when we kind of drill down to, especially the United States, the world's largest market for data centers, we come about with some pretty striking numbers. So in the United States, from now until 2030, over 45% of the growth of electricity consumption comes from data centers. We find that in the year 2030, more electricity will be consumed for processing data at these data centers than would be by all the heavy industries put together. So this is more than iron and steel, cement, aluminium, chemicals, all of that put together. So it's quite striking how the nature of the American economy, of course, catering to this technology growth, sees the rising role of data centers within it. And of course, the United States is also a big country. So if we actually drill down to specific locations, such as say Northern Virginia, we find that already this year, you know, over 20%, 25% of electricity consumption is just by data centers. We find that to be the case in other clusters as well. So we did this geospatial analysis of all data centers globally. We found a few very interesting facets and characteristics of these data centers. So we found that firstly, they tend to be very clustered around each other, which is not surprising because they of course rely on the presence of network infrastructure, of other digital infrastructure, and of course, skills. But this is quite unlike other energy consuming infrastructure. For example, iron and steel mills. But these steel mills may, you may have one per district or one per large geography, whereas data centers tend to cluster around each other. The second is that data centers tend to be close to cities, you know, of course, to ensure there's low latency and people are able to have a very quick response to their queries on AI or related to streaming and so on. So really these data centers tend to be very clustered and these clusters is where the energy sector challenges really lie.
- [11:29]
B
Yeah, fascinating, Thomas. So how is, can you, you know, I guess 945 terawatt, you know, we're talking 2030 nukes there, you know, big ones to supply that. I guess I might be Wrong on that calculation. But can you give us some sense of this? You know, taking, taking what Sid has just said about that, those challenges in locating these, but also the growth of them, how is that power demand going to be met on a sort of macro scale? And what is that then? The cascading impacts on the infrastructure required.
- [12:01]
A
945 terawatt hours, 950 terawatt hours by 2030 is a big number for sure. But we need to remember also that the power sector is not structured around annual consumption. The power sector is structured in order to meet peak demand, because demand needs to be met at every time of the day, 24, 7, at every location in the grid. So I think it's interesting to sort of translate that annual electricity consumption from data centers into their contribution to peak demand. Right. Because investment will need to be made to meet peak electricity demand going forward. And we can take the example of the United States again. So as of today, data centers account for about 6% of peak electricity demand across the continental United States. By 2030, that could be as much as 15%. So that's a doubling of data center contribution to peak demand. To put it in perspective, the entire industry sector of the United States, so that's all of the sectors that Sid mentioned, plus, you know, automobile manufacturing, plus every other manufacturing industry, contributes about 20% of peak demand in the United States. So we're talking within the space of a few years, data centers moving from 6 to about 15% of peak demand in the United States. That's a huge increase in a, in a pretty short period of time. And as you very rightly mentioned, it's coming at a time where energy infrastructure in advanced economies like the United States, but also elsewhere in Japan and Europe is aging, is being retired and replaced as part of natural replacement cycles. And so this is a big challenge for these economies. And I think the answer that we come up with in terms of how this demand can be met, I would put it in terms of two taglines if I can. It's an all hands on deck situation. We need contributions from all sources. But it's also a smarter is faster situation. In other words, we need to be really intelligent about how we integrate data centers into our electricity systems. Because if we don't do it in a smart way, if we don't make our electricity networks as smart and flexible as possible to account for this dramatic increase in a new source of load, then there's a real risk that there will be bottlenecks in terms of how quickly we can connect these projects to the electricity system. In terms of sources, we see around 50% of the incremental growth in electricity generation needed to meet data center demand coming from renewables. This is driven by a couple of factors. So first of all, you know, renewables are quick to build and relatively cheap in many places. So take for the example of Texas. You know, 80% of capacity additions over roughly the last five years have been wind and solar PV, plus battery storage. The other factor contributing to the role of renewables is the tech companies that own and operate data centers. They typically have quite strong sustainability commitments, so they're actively looking to produce, to procure lower emissions electricity. But of course, you know, we will also need dispatchable sources of electricity in the near term. In places like the United States, that's likely to be natural gas. So we do see an increase in natural gas consumption coming from, from data centers in the longer term. So let's say post 2030 we see a role for novel technologies, for example, small modular reactors. The tech sector is one of the drivers for the growing interest in small modular reactors. We're tracking about 25 gigawatts of forward commitments to small modular reactor projects from the tech sector. And this is one of the major major drivers for the project pipeline for small modular reactors, but also increasing interest in advanced geothermal technologies that benefit from the fracking revolution. To open up new geothermal resources, the.
- [16:40]
C
Energy and resources sector is experiencing unprecedented change. To help navigate this change and capture.
- [16:46]
B
Its opportunities, HC Group launched Enco Insights.
- [16:50]
C
A global advisory network dedicated to the sector providing senior advisors and subject matter experts to investment and infrastructure funds, law firms and corporates. Encoinsights leverages HC Group's 20 years of.
- [17:05]
B
Connections in energy and commodities to give.
- [17:08]
C
Clients the expertise they need when the.
- [17:10]
B
Stakes are high and insight matters.
- [17:13]
C
Learn more@encoinsights.com.
- [17:17]
B
So I want to move on to those sort of uncertainties. There's just a couple of questions. It strikes me they're obviously, and we've gone into it in previous podcasts as a demonstration sort of the permanent load needed by some data centers and others that have more discretionary compute that feeds into whether you can take advantage of renewables, etc. First question would be as these, as these data centers cluster and as they sort of wrangle with local, you know, with the, with existing infrastructure, existing market legislation that can make it quite tricky to integrate these. You pointed out the risk of doing that, you know, as at the same time we start to get these SMRs come online. Yes, Your the predictions around power consumption could be the same, but is there a sense that some of this power consumption sit sort of, I guess, in, in front of the meter or which, whichever one's the correct term, I. E. On site, in the facility and not actually be hooked up and you know, a couple of like kind of, we just all share a couple of SMRs and we're fine.
- [18:15]
A
There are a few projects that have been announced in the United States in particular that look at behind the meter generation. So this is not connected to the grid. And the data center would be the sole consumer of the power generated by the generation source. So there's a couple of projects looking at behind the meter gas generation. Generally speaking, the tech sector and data center operators, you know, they're exploring behind the meter generation, but they're not so keen on it for a couple of reasons. If everyone sources their own power generation and owns and operates behind the meter generation, then it can be quite an expensive solution because you're not benefiting from the broader economies of scale of the electricity grid and you're not benefiting from the security that comes from being connected to a broader electricity grid. And it's a big risk for the data center operator because, you know, you're taking a capital risk on both the data center itself, which is extremely capital intensive, and also potentially on the power generation project that will power it. So yes, there's exploration of this, but at this point in time, what we see is a general preference for grid connected power supply and for signing power purchase agreements with, with utilities. And we have to remember also that, you know, the tech companies, many of them are very big, very wealthy and very sophisticated, but they're not power generation companies. They generally don't own and operate the generation projects that power their data centers. They don't necessarily have the expertise to do so. And so they prefer to procure from utilities.
- [20:02]
B
Yeah. And the utilities want it connected to the grid.
- [20:05]
A
Exactly.
- [20:05]
B
Yeah, exactly. Okay, great.
- [20:07]
A
Exactly.
- [20:08]
B
So and then. Okay, Sid, turning to you. So second segment to this is really the uncertainties about this predictions around consumption. And then I want to move up. The latter half of this podcast is talking about actually when you look at energy consumption overall, the picture is not so clear what AI does because there's lots of positives and gains in efficiency. But just turning to that uncertainty, I guess there's a story where this could be a sort of underestimation and there's a story where it could be an overestimation and kind of, I guess my take on it reading between the, you know, the paper and reading between the lines is kind of, you've made an overestimation is one of efficiency, technology changes. Everyone recognizes that these chips are incredibly power hungry. There's a, that's a significant cost. That is now the major cost alongside obviously sustainability goals already mentioned and that actually we might see significant gains in energy efficiency of data centers themselves. On the flip side, there's I guess this security narrative which we started the podcast talking about the US having sort of 45% of the data centers. And maybe six years ago Europe would have been quite happy to use US data centers. You know, that that picture is changing as well as people want serenity over their data and capabilities and the clouds that operate their vital infrastructure and all the rest of it. And that means you could get more, I guess, data centers, what are the sources of uncertainty? Do you agree with that kind of binary analysis? I assume you don't and kind of, you know, what should we be looking out to as what are creating uncertainties?
- [21:42]
C
So Paul, indeed, when we started out with our analysis, we quickly ran into a problem. You know, we found that there are so many challenges in just understanding where this demand is coming from and even understanding today's state of play. You know, forget the future. Even here Today, as of 2025, what is precisely the data center related electricity consumption of which what part of it comes from artificial intelligence? It is very much a black box. So, you know, in many jurisdictions around the world, there's no clear reporting of the exact consumption from data centers. Data centers also tend to be of different types. So of course there are those colocation ones or hyperscaler ones. So the big companies tend to be better at reporting. But then you may have small servers in a particular building that just basically goes unaccounted for. We also ran into the problem of not understanding how the technology companies are actually seeing a surge in demand for, for their software, for their products. So we do not really know how much energy has been consumed in the training of models, but also the inference of those models. And of course there were challenges in terms of understanding what kind of uptake you'd see beyond generative AI. Generative AI is of course the more public facing and popular types of AI. But then at the back end of all of this is are companies that are using AI to optimize their systems internally, even in the energy sector, for example, to optimize oil and gas exploration or to integrate more renewables into the system and so on. So all of that just led us to this big question of what exactly is the energy demand today moving into the future. Naturally, there are greater layers of complexity added to this. We understand from the experience of the last several years how the efficiency gains in hardware and software have been improving. But it's anyone's guess where that could take us into the period of 2030 and beyond, especially with new technologies, new materials. So it's not just a case of the physical servers itself becoming more efficient, but also the software related optimizations. There's a reason why Nvidia today has more software engineers than the people who actually design the physical, you know, the servers itself or the chips itself. So to incorporate for all those types of uncertainties, we work with a scenario based approach. So our base case, which is where we were talking about, you know, the 415 terawatt hours in 2024 increasing to the 945 or 950 terawatt hours in 2030, that is based on our understanding of actual shipments of the servers and the under construction data centers and all the reported activity. As we see it. We did a massive data collection exercise and we tried to work with the industry to create a data set that's quite unique and therefore we feel that it's the most robust understanding of the energy consumption today and in the future. But then we also added a few, few other scenarios which looked at these uncertainties and try to, you know, explore what would happen, for example, if there was a stronger case for the uptake of, of AI, if there were fewer energy sector related challenges. So, you know, there are currently many data centers around the world that are, that are in the queue to get connections, you know, for electricity supply, if those kinds of challenges are overcome, if the uptake of AI is even more than anticipated, if some of the other competition and other related challenges are overcome, we find that in our lift off case, the electricity consumption from data centers could be 30 to 50% higher over the course of 2030 to 2035. We also have a high efficiency case where we find that if we were able to double down on efficiency improvements both in the hardware software as well as the AI models and inference itself, we could cut down on electricity consumption without reducing the actual use of AI and rollout of data centers themselves. And finally, we have the headwinds case which looks at a world where many of the challenges that the tech industry is facing are not really overcome. So, you know, as an example, the energy sector is unable to meet the rising demand from data centers. In fact, we did an analysis to See where the energy sector lies in balance when it comes to this growing demand for energy from data centers. And we find that about 20% of capacity additions from now to 2030 are at risk because the energy sector may not be ready in terms of creating that infrastructure, the long distance transmission lines, the other ancillary electrical components and infrastructure that's needed for data centers to actually come online. So 20% are at risk because the energy sector may not be ready. So we try to look at those types of challenges and we also try to incorporate an assumption where some of the artificial intelligence related generative AI especially applications are not really taken up and these companies don't really find the same type of applications as is being anticipated over the horizon to 2030. And that is the headwinds case where consumption from data centers could be about 1/3 less compared to the base case. So yes, we did try to look at all these various scenarios. The idea is to give the readers to give the industry something to peg their expectations on. So we don't want everyone to be shooting in the dark and neither we want to provide a, a certain kind of forecast which is not what we do. The idea is never to present a single view of the world, but to present different alternatives which can help peg our expectations.
- [28:00]
B
Could you just talk to that, just, just that energy sort of, that data security piece. I mean the world is, is rapidly changing in, in how countries, how politics think about security. In other words, what do you think? You know, are data centers going to proliferate by sort of security block, by even country? And actually we, we might even see overcapacity in that sense. And I don't know if that's been a consideration.
- [28:26]
C
We did try to see how countries have been approaching their own data sovereignty issues. Certainly there are several countries where you have data localization requirements. So big tech companies especially have to move certain servers closer to the sources of consumption, especially in the specific jurisdictions where these kinds of laws apply. But having said that, we do see that, you know, large technology companies in particular, they tend to use the data centers that are under their purview, often happen to be in the United States, especially for things like model training and so on. So we find that yes, there are certainly growth areas coming out of these types of policies. We find that there are new centers or new clusters of data centers that may emerge. But really a large share of the upcoming pipeline of data centers is concentrated around existing infrastructure, which also points to how clustered this industry tends to be and how reliant it is on existing telecommunications and digital infrastructure. And I mean, in general, this also points to what countries can and cannot do if they want to see more data centers and technology infrastructure in their own backyards. We try to do an analysis to see globally what countries have the types of electricity supply that is needed to ensure that data centers receive the electricity that they need. We find, for example, that the outages that exist in the grids of several countries, especially in countries which have extremely high outages, they can be as high as in the hundreds of hours per year in many of the emerging economies. Also, it's in the several dozens of hours per year, which compares to, if you look at advanced economies, which would be in the minutes or sometimes seconds annually. So that type of uncertainty, of course, doesn't work for the technology industry. So it is really something for the, for the governments to think about. It's not just about data localization, but about all this ecosystem of energy supply, that infrastructure, as well as skills that will together ultimately determine where the new centers of data centers are.
- [30:59]
B
Thanks for that. Okay, so let's move to the second half of the story, which is really about what AI itself will do to the energy sector in particular, but more broadly, energy consumption as a whole and the potential huge efficiencies that it could unlock across the sector and society more broadly. Thomas, can you sort of set the scene for that and talk about, you know, where, where we start to see these impacts that could be made?
- [31:26]
A
So in the report, we divided up this question into two parts. The first one looks at the potential to optimize existing energy assets, existing energy technologies, and existing energy infrastructure. And the second one looks at the potential of AI to drive innovation in new energy technologies. So looking at the first one, you know, the energy sector is extremely complex. It is becoming increasingly digitized, increasingly interconnected, and also increasingly complex. AI can really come to the fore in helping us manage energy assets on a dynamic basis. Let's take the example of the power sector and think across the value chain of where AI can come in. So in modern power sectors, with a growing share of renewables and an increasing penetration of electric vehicles or heat pumps or air conditioning units, introducing more variability of on demand, matching demand and supply becomes an increasing challenge. So AI can help, you know, number one, with helping to forecast supply from variable renewables more accurately. AI is becoming rapidly, really the cutting edge in terms of weather forecasting, with greater levels of accuracy at timescales of an hour ahead, a day ahead, 10 days ahead, 15 days ahead, and increasingly, and this is particularly interesting, you know, long range weather forecasts that could help to predict a dry spell, for example, for wind output for, you know, the month ahead. AI is becoming quite good at that as well. So weather forecasting is one aspect. And then if we think about the potential of AI to optimize how we use our electricity networks, AI can be used, for example, to monitor transmission lines in real time, to integrate information regarding how much electricity is being transferred down transmission lines, but also data related to the weather conditions, to the wind speed, to the temperature, which all affects how much electricity a transmission line can transfer. And then you could use your AI algorithm to boost transfer capacity on your existing transmission lines. And then coming to the consumption side, you know, we could in the future see AI being used to optimize and make electricity consumption more dynamic and responsive to the needs of the electricity system. So automating the charging of electric vehicles at times when there's surplus supply, helping electricity consumption in the commercial buildings to be more price sensitive, more dynamic, and therefore helping to match demand and supply. So taking just the example of the electricity sector, we see, you know, a plethora of applications that could be very interesting as this sector becomes more complicated, connected and digitized. The other aspect that I would just mention here, and sticking with the, with the electricity sector, is the potential for AI to contribute also to resilience. So we see increasing impacts of extreme weather and extreme events on the energy system. And the electricity system is particularly exposed because you have hundreds of thousands, indeed millions of kilometers of electricity lines which are above ground, which are exposed to wildfires, to storms and so on. And AI is providing some very interesting solutions that are starting to be implemented around using sensors, using satellite imagery, using drones to detect anomalies more quickly, more cheaply on our energy infrastructure. And so starting maybe to bring in also preventative solutions before there is a fault or correct a fault more quickly after it, after it happens.
- [35:37]
C
Hello, I'm David Hunt, founder and Managing.
- [35:39]
B
Director at Hyperion Search.
- [35:41]
C
Founded over a decade ago, Hyperion Search.
- [35:43]
B
Has helped organizations from major utilities to.
- [35:46]
C
Startups recruit their leadership teams and key.
- [35:48]
A
Individual contributors to accelerate both their growth.
- [35:51]
B
And the energy transition.
- [35:53]
C
Our three main verticals are renewable power.
- [35:55]
A
Energy storage and the mobility, the energy.
- [35:58]
C
Transition and the talent that delivers. It has been our passion since day one.
- [36:02]
A
To find out more, visit hyperionsearch.com or.
- [36:05]
C
Listen to my Leaders in Clean Tech.
- [36:07]
A
Podcast, available on all platforms.
- [36:10]
B
And we also see, I mean, even at the moment, there's a company that's been on this podcast, Vault Vision, that you Know, looking at just using technology, data, AI to under technology to understand power consumption in large systems like mines and I mean there's, there's so much efficiency gains there. Sid, maybe you can expand that from power into. Your paper talks about the impact on the oil and gas industry on obviously discovering energy, but also on fascinating on mobility as well. Can you, can you expand out where AI could actually lower energy consumption and or debalance it so that we don't need such high peak supply?
- [36:46]
C
Indeed, Paul, in fact, the oil and gas industry has been a pioneer in the use of supercomputers and artificial intelligence. In fact, they had no choice but to adopt the best computing that was available to them simply because the oil and gas extraction process in particular is so extremely complex and data intensive. So you have vast amounts of data that is generated as the prospecting of various fields happen, especially when it is deep water and these very complex types of fields. So firstly you have to kind of find ways to process all that vast amounts of data, but then beyond that you also have to find ways to analyze it very quickly and make decisions based on that. Of course it is to be noted that while artificial intelligence related interventions in the oil and gas sector can help oil and gas companies reduce costs, especially in the extraction processes, this reduction of cost does not necessarily translate to reduction in prices of the final products that consumers tend to purchase. Because of course that depends on things like the market dynamics at that point and taxes and VAT and those kinds of things. But just moving beyond that debate, I think it's very interesting to see how the oil and gas industry has actually used it and not just to reduce costs, but also things like monitoring the emissions of methane in real time and to be able to, you know, plug leaks as quickly as possible. So that's the oil and gas industry, but this is also true across industries, especially in the end use. You know, you pointed out mobility, so let's talk about mobility. We, we found some excellent case studies of how AI related route optimization is being done by freight operators, especially road freight operators, and that has been leading to significant reductions in energy demand. In fact, we find in our analysis that if all these various route optimizations and other interventions of AI in the transport sector were to be scaled up to the sectoral level, it could reduce as much energy as is consumed by 120 million vehicles. So 120 million cars to be precise. So that's the significant number of cars of the energy consumption associated with these many cars that can be reduced by scaling up such AI interventions. In fact, if I could just talk about one of my favorite interventions of AI, which people are probably less aware of, it is the reduction of contrails in flights. So you may have of course, seen when there's a plane flying at a very high altitude, you may find these condensation trails at the back of these, these planes. Now, scientists have estimated that up to 60% of the global warming potential from aviation actually comes from those condensation trails. And those condensation trails are actually formed when the planes are flying in a air pocket with very specific temperature and pressure conditions. Right. AI related route optimizations can ensure that these contrails are reduced by over 50% at very minimal cost, in fact, at very minimal cost and very minimal energy consumption rise. So only 0.5% more energy can reduce over 50% of the condensation trails in the aviation industry, which is quite striking. So it really goes to show how AI is able to solve problems that would be almost impossible to do manually.
- [40:31]
B
Yeah, I like the AI doing the traffic lights and actually optimizing traffic and how much that could save. And you know, and as we all sit there early morning in a, at a red light when no other cars on the road. Right. You know, and I think that's the point. I mean, this segment I found really fascinating because it's just, it's impossible to predict at point, this stage how AI is going to cascade through society. Talking in the paper even about, you know, biomedicine and the impacts there and how that might change human behaviors and so forth, you know, and therefore impact on the, on the energy sector. You know, there's also the, what are we all going to do with our spare time? Are we all just going to be playing video games and consuming more power and all the rest of it? So it's a fascinating segment. The paper also talks about impacts on energy security and going into the critical minerals world, which we, we've spoken about a lot on this podcast and more to come, I guess. Thomas, when you put this all together and understanding that this is, as with all forecasting, this is rigorous in terms of the data sets, use the analysis done, but there's lots of variability, uncertainty in it. When you put it all together, does that increase power consumption as a result of data centers, is that offset by, by the potential efficiency gains? What does the overall picture look like in energy consumption as well as obviously greenhouse gas emissions?
- [41:51]
A
Things are very uncertain, as you rightly point out. But nonetheless, we think it's useful for policymakers, for businesses to have a bit of A guidebook in terms of where things may head. Because we can't get into analysis paralysis, as they say. We need to make investments, we need to keep moving forward. So when we, when we look at the aggregate picture, data centers are going to lead to an increase in energy consumption. In our analysis, the potential for AI to reduce energy demand, if, as Sid mentioned, you know, existing solutions are applied more broadly across entire sectors, then the decrease in energy consumption that we could have in our analysis would significantly outweigh that increase in energy consumption from data centers. But that, of course, is a big if. There are a number of important barriers. These include skills gaps. I actually found this aspect of the report one of the most fascinating. We worked very closely with LinkedIn to map AI related skills across sectors and across countries. And essentially what we found is that the energy sector is lagging behind in terms of acquiring AI related skills in terms of its workforce. That is a big barrier that companies keep talking to us about. When we look at the potential to roll out these solutions. There's also barriers in terms of digital infrastructure. So if we want, for example, AI to optimize heating and cooling in commercial buildings, well, the heating and cooling equipment needs to be digitized. Even in advanced economies. Only 50% of heating and cooling in the commercial sector is digitized and around 12% in emerging market and developing economies. So you need policy to come in and make sure that those smart meters are installed, that we have more controllable electricity consumption, so that the AI algorithms can do their work and help us to optimize consumption debt. And the other barrier that I would point to is regulatory and legal risks. So survey after survey of companies in the energy sector, but also outside the energy sector, point to concerns around data privacy liability as major barriers to adopting AI related solutions. And this is particularly the case in the energy sector where we're talking about infrastructure that is critical to national economies and indeed to national security. So, you know, no energy market operator wants to be on the hook for a blackout in any country, and particularly not if it is found out that that blackout is because, you know, to put it hyperbolically, you know, like, because the AI algorithm did something wrong. And so this is something where, you know, we will need policy interventions to create pilot projects, sandboxes where solutions can be tested, but also, you know, legal regimes, data regimes that give companies and regulators confidence that they can start to implement these solutions more broadly. The final thing that, you know, you talked about what we would be doing with our time if AI starts to take over jobs and so on. And you know, if we, if we think more long term, let's say beyond 2030, into the, into the2030s, it's quite uncertain, you know, what the rebound effects could be of broader AI adoption across the economy. If it boosts productivity and economic growth substantially, does that increase in economic growth lead to an increase in energy consumption? That's something that we don't really have the tools to answer, but something that we need to think about even on a more limited scale. You know, if we start to see significant penetration of autonomous vehicles already, you see, you know, companies like Waymo in, in San Francisco operating millions of miles per year with quite a good safety record. We're not quite there yet in terms of autonomous vehicles, but definitely they're making progress. So would that, for example, lead to an increase in energy consumption if people start shifting away from public transport where public transport is available? So there are a lot of uncertainties. But you know, the big picture when we step back is that our assessment is that the energy optimization potential of AI exceeds, at least in the near term, the increase in energy demand from AI. But it's by no means a given that that energy optimization potential is tapped.
- [46:52]
B
I think it's fascinating because bear with me. So on the skill set side, our technology practice at HC Group, the search firm, is one of our fastest growing and has been for a long time. And you know, it is a struggle to find talent that wants to go into the energy industry. And part of that is because it's seen as the key driver of emissions. It's dirty, all this stuff, right? But ironically, I mean, the, the sort of, the meta statement that hits me from this discussion is that actually AI has turned up right at the perfect time to actually accelerate the energy transition. Right? We're not going to be able to manage these vastly more complex grids, decentralized grids, without it. That's going to enable deployment of decentralized generation of renewables much more quickly. All this efficiency gains that are there as well as actually it will be a key driver. And in some senses, actually it's a call to arms for the technology world that this is the place where you can make the biggest difference by coming to this sectors. That's what's truly, truly going to unlock all of this opportunity for everyone. And you know, in a virtuous circle of actually enabling us to have more data centers, more compute power more broadly, because we're using our power more efficiently and more effectively. I mean, it seems like AI is the shot in the ARM the energy transition needs.
- [48:13]
A
You know, in our report we said AI is not a silver bullet for the energy transition. We still need those, the same policies that we needed before to support, you know, electric vehicles or renewables deployment or nuclear investment or carbon capture and storage. You know, none of, none of what we have needed to do for some time has, has really changed. At the same time, it's certainly true as you, as you say, that I can provide, you know, a very interesting catalyst for some of the, or enabler of some of the changes that we, that we need to see. You know, one of the areas that we looked at in the report that I think is most interesting if we look longer term is the potential that AI accelerates in energy innovation. There's already been, I think, a sea change in how we see innovation in biomedicine with the breakthrough of Google DeepMind's AlphaFold winning the Nobel Prize last year and really offering quite extraordinary new tool for drug discovery where you can essentially model before testing on humans, before the laborious process of searching for the right molecule, you can actually go out and say, I want to target this particular biomedical mechanism from the get go. And we see a lot of innovation problems in the energy sector that are of a similar nature to finding a needle in a haystack. You know, it would be great if we had more efficient perovskite materials that could be easily manufactured, but we've actually produced, you know, experimentally only a tiny fraction of the total available perovskite materials that are theoretically possible. You know, we've produced about a thousand perovskites experimentally, but there are probably 10 million that we could explore. Doing that purely experimentally is far too slow. But if we could do it computationally, if we could do it with an AI model that would run through the possible combinations of perovskite materials and give us candidates that would be highly performing, efficient, not use materials that are rare or expensive, but is also relatively simple to manufacture. That could be a game changer. So I think sort of long term there's significant potential there to accelerate energy innovation. But even with that, you still got to manufacture all of these new technologies. You've got to scale industrial supply chains, you've got to bring them to the consumer, you've got to adjust regulatory policies and so on. So even with much more high performing low emissions technologies, we'd still have to do a lot of the same things that we've been needing to do for some time. So for us, AI is definitely an enabler of the energy transition, but it's not a silver bullet.
- [51:11]
B
Well, it's been absolutely fascinating. I've really enjoyed having you both on. And I encourage everyone to go and read the report. And again, I'll put the links in the in the show notes and, you know, hopefully have you both back on in a year or two and see where we stand, not only against the predictions, but also all those things, the unforeseen, the unknowns, unknowns that would undoubtedly have popped up in a world that's rapidly changing. And AI is a big part of it. So, Sid Thomas, thanks for your time.
- [51:37]
C
Thank you, Paul. It was a pleasure.
- [51:39]
A
Thanks, Paul. Great to chat. A really fun conversation.
- [51:44]
B
Thank you for listening. To find out more about HC Group, our global offices and our expertise in search within the commodities sector, please visit www.hcgroup.global.