Infineon key visual in RGB

Supporting AI power demands sustainably: the role of renewables

We’re bringing you the latest from the world of semiconductors – straight to your ears! From quick takes on trending applications to deep dives on product innovations, our experts give you their take on the tech behind the tech.

Podcast

Feb 18, 2025

Podcast

In this episode of Podcast4Engineers, host Peter Balint explores the intricate demands of powering AI with guest, Fanny Bjoerk, Director Global Application Marketing for Datacenter Power Distribution. They discuss the exponential rise in power requirements of hyperscale data centers driven by the AI revolution, and the critical role of renewable energy in meeting these demands sustainably. The episode delves into the challenges and solutions for integrating renewable energy sources, such as solar and wind, and the importance of energy efficiency and reliable power delivery for the future of AI infrastructure.

Listen on Spotify Watch on Youtube

Speakers

In this episode of Podcast4Engineers, host Peter Balint speaks with Dr. Fanny Bjoerk, Director Global Application Marketing for Datacenter Power Distribution at Infineon.

Peter Balint

Host:

Peter Balint has shaped visual and audio narratives at Infineon since 2021. He’s a video producer with 20 years of experience and has produced podcasts for the past 10 years. Over his career, Peter has interviewed speakers from all over Europe, bringing high-quality media production and engaging conversations to the forefront of his work.

Fanny Bjoerk

Guest:

Dr. Fanny Bjoerk is the global marketing director for data center power distribution at Infineon. Fanny joined Infineon in 2003 and has previously held positions in marketing and program management for power devices, e.g., silicon super-junction MOSFETs, IGBTs and SiC devices. Fanny holds a PhD degree in solid-state electronics and a M. Sc. degree in electrical engineering from the Royal Institute of Technology in Stockholm, Sweden.

Additional information

More episodes

Guest: Hyperscale data centers today have power demands of 100 to 200 megawatts. This is compared to 5 to 10 megawatts it used to be before the boom in AI use happened. Renewables is the most important trend, the biggest lever when it comes to quickly driving decarbonization.

 

Host: This is the Podcast4Engineers, the podcast you just have to listen to if you're interested in what's going on in the semiconductor market. I'm your host, Peter Balint, and today we talk about We powering AI and specifically what roles do renewables play in powering AI. And with us today in the studio is Fanny Bjoerk. She's the Application Marketing Director here at Infineon. Welcome. Thanks for coming in today.

 

Guest: Thank you. Pleasure being here.

 

Host: And my first question right out of the box is, how much power is actually needed? We've in previous episodes said that AI is consuming so much power, but can you quantify this for us?

 

Guest: So newly built hyperscale data centers today have power demands of 100 to 200 megawatts compared to 5 to 10 megawatts it used to be just very few years back before the boom in AI use. So already today, the data center power went up by 20 to 50 times. And yet we are only at the very beginning of the AI revolution. And exactly how this now will unfold, a lot is uncertain. But at the same time, all the power figures that are discussed in the industry now show that in the coming years, so-called AI factories rather than traditional data centers will be needed. And power demands of surely several hundred megawatts, but even 1 gigawatt or 2 gigawatt data center halls are, will be the new arrival in the next coming years. The sheer power of this is like city size of power robots.

 

Host: And your role here at Infineon deals with renewable energy as it relates to these hyperscale data centers. Can you tell us a little bit how that looks or how can renewable energy really play an important part in the data center?

 

Guest: So already today it plays a significant part based on market analyst studies, like for example, S&OP, Global Commodity Insights, they did a survey with the top 20 hyperscalers in the world in 2023, where already then, the renewable share was 50% over the non-renewable share for global electricity demand for data centers. It's already there, and many data centers today, they already have onsite generation with a solar farm. for example, next to it. At the same time, with these very high-power increase needs now to enable the next AI platforms and the next level of AI progress in the world, it's not as trivial to match, let's say, the availability of sudden wind with having a stable baseload for an AI data center.

 

Host: What's the reality of shifting a little bit more renewable energy towards powering a data center, for example?

 

Guest: It's a lot about timing and/or baseload, but also timing or peak loads. The question is, to what extent can battery energy storage systems mitigate this when there is no availability of sun and wind. Most likely it's going to continue with a combination of onsite generation, having solar farms and wind towers basically on premises of the data center site, but in combination with— also renewable energy coming through the utility grid from longer distance generation. This is already happening today with power purchase agreements, so-called PPAs, also driving a lot of this 50% share that I mentioned is roughly the renewable portion already today. Renewables is the most important trend, the biggest lever when it comes to quickly driving decarbonization. Nuclear energy as another clean carbon-free energy source will take longer time. It's not going to be able to connect to data centers within this decade on large scale. Even it's, of course, a promising way of further powering data centers with clean energy in the future. Another clean energy source is geothermal energy that again now shows a lot of promises, was also shown by the International Energy Agency recently. But this also takes longer time. This is not within this decade. And as we talked about earlier, the next steps in AI progress the next level of AI platforms, it's going to happen already in the next 5 years. There is only renewable energy in principle as a sustainable energy source.

 

Host: And when we talk about renewables, what are we talking about here? You mentioned solar and wind.

 

Guest: Solar is very prominent. It's doable. If there is land, it's possible to build a solar farm and to power 1 gigawatt The corresponding size of the solar farm is roughly 8 square kilometers. So, 2.8 times 2.8 km. It's a huge solar farm, of course, but if there is land, it's possible. Wind is also a prominent source, but maybe mainly for long-distance generation. And then there is also fuel cell energy, which is another renewable energy source. If the hydrogen is generated by renewable energy sources. But fuel cells are gaining quite some attention now regarding data center power needs.

 

Host: We've already talked about energy needs, but what about the efficiency needs? This is probably a huge issue as well.

 

Guest: Yes, of course, efficiency is a big concern, especially when increasing power so many times like we are talking about now compared to before. As we have seen at the Open Compute Global Summit conference in San Jose last October, hyperscalers talk typically about 96% and 96.5% being feasible for the power distribution part. So just to get from the utility grid or the onsite generation to the server rack portion inside the data center. And with 1 gigawatt of power, if the efficiency is 96.5%, then you are close to 40 megawatts of power that you have to take care of as waste energy. Or reused in a smart way, which is also a big discussion now, how to reuse the heat generated. But efficiency is a big concern. At the same time, such a big step in increasing power also opens now for efficiency improvements by new solutions, because new power architecture solutions become interesting and also feasible to work on simply due to the higher power need. And this opens for further efficiency improvements. And here, of course, power semiconductors are a big part in this because any power conversion stage used in the whole chain here for getting the power from the energy stores down to the AI GPU chip, in any power converter stage, power semiconductors are at the very heart of this. And no power stage can operate more efficiently than what the power semiconductors allow. And here at Infineon, we have the track record here of consistently innovating and bringing out the most advanced, latest, and energy-efficient semiconductor technologies. Primarily now in InN bandgap materials like gallium nitride and silicon carbide.

 

Host: So, Infineon has a lot to offer in terms of efficiency in power moving forward, but they also have an advantage with customers. Infineon is close with their customers, and it would seem to me that there's some feedback that we can gain in trying to develop what kind of products we have in the future. Can you tell us a little bit about that?

 

Guest: Yes, we work on several examples of scale developments now of new solutions for power infrastructure suitable for AI data centers. And let me maybe start with the— from the start of the energy source, how to now power an AI data center with a clean energy source. Here, Infineon is helping our customers and in the and also the hyperscalers to improve on LCOE, so the levelized cost of energy is a measure how to quantify the cost-effectiveness and scale-effectiveness of an energy source. Here we help in improving this by rolling out new high-voltage components, in this case in silicon carbide, to enable the solar farm to provide to increase the voltage, which is very beneficial for the LCOE. And then that's just the first power conversion stage to connect it to the energy source. But then there are typically 6 power conversion stages. Let's take the example here of photovoltaic energy, so solar, a solar farm. First, there is needed a DC-DC stage that converts the different DC voltages coming from the solar panels into one stable DC voltage. That's the first one. The next power conversion stage needed is the DC-AC to convert it into an AC distribution voltage. And what comes after that is another 4 stages to subsequently convert down the voltage to 0.8 V, which the AI GPU finally needs. It's 6 power conversion stages, starting at a source voltage of 1,500 V, but the trend is going higher now, 2,000 V, 3,000 V, because it improves the levelized cost of energy of solar farms, all the way down to 0.8 V. And Infineon has a broad portfolio with all the voltage levels and form factors to maximize efficiency in every power conversion stage here.

 

Host: Let's take a look at what we know today and see if we can apply this to future, or do you see this so-called self-contained data center with its own source of power? Is this actually possible? And when I say power, I mean renewable energy, of course.

 

Guest: Yes, based on what I know today, I think it's doable. We will see such role model data centers. Of course, they will have to be built where the renewable energy source is available or can be built, like these 8 square-kilometer solar farms I mentioned. And another factor that I see is that it will also require a bit of flexibility, say, from the hyperscaler, the data center operator, to also only go for the big demands of AI power needs when sun and wind are available. If there is a flexibility willingness, it's definitely doable.

 

Host: Okay. So that we can end then on a positive note, which is that this is possible and it's in the future. It's just a matter of time. Okay, good. Well, thank you so much for coming in today and sharing your knowledge with us and with our listeners.

 

Guest: Thanks for having me.

 

Host: And for our listeners, I say thank you for tuning in today. And we are always open to your ideas about future episodes or questions. And you can reach us at wepowerai@infineon.com. Thanks again, and we'll see you soon.