AI, Energy and Climate: IEA’s New Energy and AI Report: Thomas Spencer and Siddharth Singh

To stay up to date on all the news that you need to know, there's no better p...

And there's no better way to enjoy the DSR network than by becoming a member.

Members enjoy an ad-free listening experience, access to our Discord community, exclusive content, early episode access, and more. Use code in DSR 26 for a 25% off discount on sign up at the DSR network.com. That's code in DSR 26 at the DSR network.com/by. Thank you, and enjoy the show. Welcome to the AI Energy and Climate Podcast, a special series from the DSR network hosted by David Sandalot, inaugural fellow at Columbia University Center on Global Energy Policy.

Join us as we talk with leading experts to explore the intersection between these critical issues that will impact the future of each and every one of us.

I am David Sandalot, this is the AI Energy and Climate Podcast. The International Energy Agency in Paris just released a new report on energy and AI and it's a treasure trove of information.

“The report, which is called Key Questions on Energy and AI, builds on a landmark study released by the IEA last year, updating and expanding its analysis.”

The new report is global in scope and looks at how much energy AI is using where that energy is coming from and how AI is transforming the energy sector. It's an important resource, providing authoritative information and diving deep into topics including the role of natural gas, nuclear power and batteries and AI data centers. The strains caused by the clustering of data centers in certain locations, the future of physical AI such as robotics and autonomous vehicles and much more. The report was prepared by a large team led by two senior energy experts at the IEA, Thomas Spencer and Sid Hart Singh.

I recently had a chance to talk with the two of them about their new report. I hope you enjoy our conversation.

“Thomas Spencer and Sid Hart Singh, it's great to see you. Congratulations on the release of your new report called Key Questions on Energy and AI and welcome to the show.”

David, great to be with you. Thanks for the invitation. And thanks David, it's a pleasure to be here. Well, well, thanks for joining us. So your report covers a wide range of topics. You look at how much electricity data centers are consuming. Where that power is coming from, what impact the AI is having an energy systems and a variety of related topics. What would you say, what are some of your headline conclusions? You know, David, that's a dangerous question because it could become a long answer.

If you ask an open leading question like that to two researchers, they will bang on for a long time, but we'll try and distill it down.

“You know, I think the first key message from the report was really that AI is still moving incredibly rapidly.”

It's not slowing down and that's having implications for the energy sector. So whether it be in terms of, you know, the increase in capital expenditure by hyperscalers on data centers, progressing AI itself with the rise of AI agents and increasingly energy-intensive modes of AI or improvements in energy efficiency.

It's just really moving incredibly rapidly. And then, you know, we see this and this was the second key message coming out of the report.

We see this increasingly coming up against physical constraints. And some of these are in the energy sector, you know, in particular long-way times to connect to the grid, but also increasingly in energy technology supply chains and we can get into that a bit more if you're interested whether it be in terms of turbines or batteries or or transformers. But also, you know, physical constraints in the manufacturing supply chain for some of the IT equipment that needs to go into data centers.

And so when we add those two together, we see a huge pressure for innovation to get around these bottlenecks.

So I think a key finding from the report was that AI and data centers are als...

You know, we put in the report while still an energy-taker AI is increasingly also an energy-maker.

You know, the tech companies are becoming active participants in electricity markets through their procurement strategies, and really reshaping the sort of market and technology mix in the electricity sector. And so if we zoom out a bit and think about the broader implications of this, this really does offer an opportunity for, you know, the countries and companies that that consists on this to modernize the electricity system and to drive forward innovation in electricity technologies. And that's a big opportunity.

There's so much to dig into there Thomas. Let's just start with how much power data centers are using.

“And I think you say about 1.5% of total global power demand, if I recall, is being used by data centers.”

You'll elaborate on that and how that's distributed globally. It's not some pretty interesting observations on that. Sure. So David, as you rightly mentioned, about 1.5% of global electricity consumption currently goes to data centers. This is about approximately 500-about hours as of last year.

You also rightly point out that it's actually not very spatially distributed. So we did this geospatial analysis of all the commercial data centers globally. That's over 11,000 of them.

“And we found that there are a few interesting geospatial trends that we observe firstly.”

They tend to cluster around each other. There are obvious network effects and advantages that come from data centers being around infrastructure that caters do their needs including infrastructure for telecommunications, for electricity and so on. Data centers also tend to cluster around cities. So unlike a lot of other energy consumption infrastructure, which could be more spatially distributed.

For example, aluminum smellters or steel plants, data centers can consume as much electricity as them and still be close to each other and close to big cities which are already big centers for energy demand. So as a result of that, even though that 1.5% may seem small today, in areas where these data centers are clustering, it can be as high as 25 or 30% of electricity consumption going to data centers. That's where the challenge really lies. And we find that this clustering trend actually isn't slowing down.

So all the data centers that are currently in the pipeline, which means that the ones that are either under construction or the ones that have applied for initial permits but have not started construction yet, they will create new clusters as well. So the clustering phenomenon of data center only increases as we go forward from here. And these data centers of course are becoming larger and larger. So typical AI hyposcaler data center today might consume as much electricity as 100,000 households.

This increases to, you know, well over 2 million households for the largest ones that are under construction today.

Yeah, I thought one of the very valuable parts of your study is that it's global and you look everywhere and the figure of 1.5% of global electricity demand seems small, but initially not obvious why this phenomenon would be getting so much attention in some localities. If it's such a small percentage of global power demand, and you really bring out the importance of clustering. And although it's 1.5% of global power demand overall for data centers in number of geographies, it's very significantly more than that,

and putting significant strains on electric threads. You have very interesting discussion I thought about power density and the implications of technological advances in that area.

Something I've never seen before, which is the statement that with the newest generation of server racks that are coming out of the NVIDIA,

in particular with the there are a Ruben architecture.

“You find that the amount of power that is being used by a server rack about the size of a refrigerator, which it equals the amount of power being used by 65 households, which is I think a pretty striking binding.”

I wonder if you have anything to add on that front. I think David this is a good example of a trend that is being driven by AI that will impact on energy technologies and energy innovation.

First of all, why are data centers and AI data centers in particular becoming...

That's really driven by the needs of AI.

“So the largest models today are too big to fit on a single chip.”

They need multiple chips to work in tandem in order to train them and run them. And those chips need to be spatially proximate to each other in order to reduce data losses and data latency. So they can really work as a single computing unit. And that means you're getting more and more chips packed in together close together in these latest AI server racks. And so we show in the report that you know in 2020 before charging BT, the average power density of an AI rack was roughly 15 kilowatts per rack.

15 kilowatts is roughly equivalent to the power draw of about five. Evans running at the same at the same time. The next generation is going to move up to 600 kilowatts per rack. And the size of the rack hasn't changed. And as you mentioned, 600 kilowatts is, you know, nudging up towards the peak power draw of 65 households. And this really has implications for the power delivery architecture within within data centers.

We need to move up towards higher voltage delivery. And in some cases, even moving from alternating current to direct current voltage within the data center. And that has implications for the kind of transformers that are needed. But also for technologies like power electronics. So power electronics are semiconductors that are not used for information processing, but semiconductors that are used for power control and power quality.

So we can use semiconductors to step up the voltage of power distribution step down the voltage and so on. And as these data centers become more and more power dense, we will need these technologies to become more and more sophisticated. So this has implications, you know, it could help to commercialize technologies like solid state transformers, which can be very interesting elsewhere in the energy sector. Whether it be in terms of, you know, the next generation of distributed energy grids with batteries and solid state transformers at the distribution level.

But it also has implications for supply chains, because some of the power electronics that data centers need.

I semiconductor based on their based on critical minerals like like gallium, the supply chain for which is is very concentrated in a single country, namely China.

“So I think one of the big big findings from our report is that we need to pay more attention to this electricity technology supply chain.”

And the innovation stresses that data centers are playing on it, but also the kind of geopolitical risks that this could create. It's made a difference. Do you want to know what's in your community, but don't know how? You can now get to GOFANTMI.com and start your own spin-off business. Your next spin-off business doesn't have to be in the school or the kids. You can only start in less minutes in a GOFANTMI spin-off business.

The most important thing is that the human or the project is on the ground. GOFANTMI is the 4th spin-off business platform for money spin-offs, which are important.

If you are interested in your work and support yourself, you can do it right away. If you are creative, local, or your life is important, then it's important. There is a reason why GOFANTMI from million and support will continue and will continue from spin-off business. GOFANTMI makes many people a different difference and helps people to come together. Start in your own GOFANTMI spin-off business.

“GOFANTMI spin-off business platform for money spin-offs is the most important thing. You can now get to the point where you can do it.”