Ep 46: The Hidden Cost of AI on Cities
Seyi (00:09)
Hello, hello, hello Reza, how are you? I am doing very well. Today it is another episode of Future Forward.
Reza (00:11)
I'm good Seyi how you doing?
Yeah, Shay. This is our 47th episode, Shay, and when we release this episode, it's gonna be one year that we've been doing Future Forward. And so I just wanna mark this milestone. ⁓ Yeah, yeah, this is awesome. We were just saying before this podcast started how we kept talking about wanting to...
Seyi (00:32)
That is fantastic. You know what, I'm clapping for that one.
Reza (00:42)
do a podcast. And then once we got started, were like, what was holding us back? We're having so much fun doing this. But yeah, Shay, today is going to be, today is, you're the expert on this episode. We're going to talk about the hidden cost of AI and really talk about energy infrastructure. How does it keep up with these AI energy demands? And you've really become an expert.
Seyi (00:43)
Yeah.
What took us so long?
Yes.
Reza (01:08)
on this and your industry insights will help us unpack how cities, utilities, planners can navigate this challenge that we're facing where we have a lot of growth in digital and AI and how does it intersect with sustainable energy use? ⁓ But yeah, before we begin, we should welcome our listeners, Jay.
Seyi (01:25)
Yeah.
We should, we should. So all of you who are listening to us right now, thank you. I'll just start with that. It's been a year of so much fun for us. I truly, truly, we both truly value the fact that you consistently show up to listen to Future Forward. To our new listeners, welcome. If this is your first episode, you meet us at our one year anniversary.
Reza (01:36)
Yes.
Seyi (01:55)
Thank you for joining. To our old listeners, we are just appreciative of you consistently showing up and listening to a conversation that Reza and I had been having for years before we started to record. We would meet up almost every week at a coffee shop and talk about just cities.
the city we both live in, the sustainability of that city, how to build thriving communities in that city. And we will take the lens of history on a certain topic.
the present day conditions and then strategic foresight about the future. And at some point within the one year, we came up with a few laws based on deep research as well. We came up with 21 laws that sort of explain and contextualize all the ideas we share on future forward. So thank you for joining us.
Reza (02:50)
Yeah. So Shay, this one, we're going to talk about this AI energy nexus. And you've been on a number of panels where you've talked about utilities, AI, things that you have learned. But before we do that, Shay, you've been working at this intersection of AI and energy. And I'd love to hear about how did you get to this place? How did your career lead you to this particular
of becoming an expert on this.
Seyi (03:18)
Yeah, so I'll keep it pretty brief, but I have now been in the energy slash utility industry for 23 years, which is mind blowing to me. I started after post-grad in the UK, working at a thousand megawatt power plant, combined cycle gas plant. worked in operations and it was just so amazing to me when I worked back.
there back then that the work we were doing was providing electricity to about 600,000 people in the London and Greater London area. Just the impact. I worked in the control room on the plant floor.
and in the office which was where the power was converted to money essentially. So I did a bunch of the shift planning, outage planning and the nitty gritty of running a power plant and
Reza (04:06)
Mm-hmm.
Seyi (04:16)
There was a point we needed to build some software to trade our electricity. So I got pulled into the team and that sort of moved me into the intersection of power generation and technology on the software solution inside.
Reza (04:32)
Hmm.
Seyi (04:33)
which is around when I then moved to the US after spending about eight years or so working at that intersection, moved to the US, started a company that was sort of a backend for utilities that presented data in a way that end-use consumers could digest and make decisions on. We sold that company. I worked.
at a fund that was investing in the clean energy transition and then did a bunch of consulting, actually traveled the world a bit, which was a lot of fun. and the focus was on the implementation of clean energy in the traditional utility framework. ⁓
Reza (05:17)
Yeah.
Seyi (05:18)
did Europe, did some parts of Africa, I'd never been until I started doing the work. So that was really fun for me. Did Mexico for a little bit as well. And then I ended up now where I'm at with a renewable energy developer, the third largest renewable energy developer in the country.
on the engineering team but focused on innovation. What's changing about the industry? How can we be at the forefront? I'll pause here and say the things I share on this episode podcast are my own representation of what's going on, not my, not the company's, but.
Reza (05:44)
Mmm.
Seyi (05:59)
I've ended up on a bunch of doing a bunch of work, but also in a lot of conversations and panels and presentations, looking at this intersection where we've been talking about AI now for the last however many years. And one of the things you and I would discuss Reza was the fact that AI is really just the conversion of energy into work.
Reza (06:23)
Mmm.
Seyi (06:23)
And
I remember we would have that conversation. And that's been the premise. That's the history for me personally, and the intersection of AI and energy usage just sort of culminates in all the time I've spent at technology, at the intersection of technology and utilities so far.
Reza (06:42)
Yeah. So that's a great intro, Shay. And I'm very curious about all these stats that we keep seeing about heavy power demand from AI. And you have some thoughts about why are AI workloads so energy intensive? That's a good place for us to start. And how is that changing the data center and a utility landscape?
Seyi (07:05)
Yeah, so the point I just made, which I believe that line came from you Reza, where you were, you drew me like a flow diagram of before, we used to use energy to create sort of food and then humans would create work and knowledge.
Reza (07:25)
Yeah.
Seyi (07:26)
And now we're just using that same energy and directly passing it to machines, data centers, chips and racks and converting it into knowledge. So we're sort of taking a step out of the way there. And for all of the AI companies, the real equation is the more power you can get,
Reza (07:33)
Yeah.
Seyi (07:52)
the more models you can train, the more data you can crunch, for lack of better way to put it, and convert into knowledge, intelligence, which is the second part of the AI phrase. And so the electricity consumption to create knowledge,
Reza (08:02)
Yeah. Yeah.
Seyi (08:13)
And the exponential growth of electricity consumption to create knowledge is really the conversation we're having at that intersection. The training of...
complex models through the use of thousands of graphic processing units is significantly raising power requirements. One of the stats I believe was around the fact that we currently use about 500 terawatt hours today. Sort of, these are rough numbers, but.
Reza (08:48)
Mm-hmm.
Seyi (08:52)
By 2030, because of AI, we will be just a little over 1,000 terawatt hours. That's a doubling. That's a doubling. And it will all go towards just purely crunching.
Reza (08:59)
Wow. Double. Almost double. Yeah.
Seyi (09:09)
information to convert it into intelligence. So that's, that's the big driver. data centers themselves have evolved. So we currently use a bunch of graphic processing units in data centers. The goal is to move towards more neural processing units for AI and graphic processing units allow us to get the AI models we need, but they weren't built for.
Reza (09:12)
Yeah.
Got it.
Seyi (09:38)
AI. They were built to render games and we've now force fitted them to allow us to get the AI. And what does that mean? We went from
Reza (09:41)
Yeah.
Seyi (09:52)
using data centers that were about 30 megawatts of electricity. Remember I said I worked at a thousand megawatt power station before data centers would consume about 30 megawatts of power. But today,
we're now seeing 200 megawatt facilities everywhere. That is almost the entry point, especially when you talk about the hyperscalers, the Facebooks, Metas of the world, the OpenAI's of the world. They're looking at 200 megawatt data centers. Microsoft is working on a gigawatt, a thousand megawatts, the size of
of the power station I worked at. Microsoft is working on a data center that will use the same power that I said was serving about 600,000 people. So the size and density of our data centers have grown. That's the second driver. And then the third is that we need to cool these data centers as well.
Reza (10:46)
child.
Yeah.
Seyi (10:56)
And that is also energy draw, we're converting energy to cool the data centers. And then the fourth is just this push towards electrification of everything, transportation. So I'll pause there, but those are the four main ones with the real one being the one I spent the most time on, data centers and AI.
Reza (11:10)
Yeah.
Yeah. Yeah.
So that's a really good setup for this next section that I want to talk about, which is we're trying to understand the impact on the urban infrastructure. This is going to create strain. And one of the things that you've shared, Shay, is that Ireland now sees 21 % of its electricity going to data centers. That's pretty staggering, right?
Seyi (11:29)
Yeah. Yeah.
Yeah.
Reza (11:41)
type of pressure mean for urban grids in cities, primarily, because that's what we care about.
Seyi (11:46)
Yeah, yeah,
which is our area of concern here. So yes, you are correct. That stat is mind boggling. It's there was in 2015, 10 years ago, only 5 % of Ireland's electricity was used for anything related to data centers. Now it is 21%, which is a 470 something percent increase in consumption on a grid.
like the US's grid that wasn't built for this much load period. The infrastructure was built decades ago. It is what is and was crumbling before AI.
came in. We have transmission lines that were built based on a centralized transmission grid or distribution grid. And now when you think about these data centers and the little 200 megawatts in one place, in one location, we're talking about almost what would be considered a mini grid unto itself as part of exactly, exactly. And so that growth, unfortunately,
Reza (12:54)
concentration.
Seyi (13:01)
the grid is not prepared. And this is a topic that I wish we were talking a lot more about, but the data center companies, the technology companies are trying to find ways around the crumbling infrastructure by...
Reza (13:04)
Mm-hmm.
Mm-hmm.
Mm-hmm.
Seyi (13:20)
islanding off the power they use for their data centers away from the grid, which is absolutely not the way you treat what is a public good. Utility infrastructure is a public good.
Reza (13:33)
Yeah,
because you mentioned this old model of power infrastructure is just collapsing. This is what you're talking about,
Seyi (13:38)
Exactly. This is exactly what I'm talking about.
so we're putting the technology companies and AI companies are putting demands on the grid, but trying not to be a part of upgrading and updating the infrastructure. And what you will end up with is collapse if we keep taking the approach we've been taking so far.
Reza (14:05)
And what does collapse
mean? what, yeah, what?
Seyi (14:08)
Yeah,
collapse literally means power outages, Reza. Outages at a scale we haven't really experienced before. And...
Reza (14:11)
outages.
Yeah.
I think we talked about it in a prior episode where we talked about brownouts. Yeah.
Seyi (14:20)
We did. We did exactly.
And, and because we now have these hubs of data draw and by the way, these data centers also have a lot of need for water for cooling the facilities. We are concentrating resources in locations within a grid that is collapsing. And what it will do is.
break everything else around it if we don't pay attention or find a way and this is where the policy is coming to get the infrastructure, the data center companies and technology companies to participate in the fixing of the grid. So I won't jump there. I know we're going to talk about strategic foresight eventually here.
Reza (14:49)
Yeah.
Yeah. Yeah. Yeah.
So the next part that I want to talk about, is, which is we are beginning to intimate at, which is about these trade-offs and tensions that you're bringing up. there's a big tension over here. Cities want to attract technology investment. They want AI jobs, but the energy demands, this could overwhelm, they're already strained infrastructure. And so how should city leaders think about this trade-off?
Seyi (15:13)
Yeah.
Yeah, so...
I'll start. made the statement on a panel a few weeks ago and one of the other panelists had to chime in because I was in his view derailing the conversation and I'll make the statement again here. think because of the decisions we're making to focus our attention on creating more power on the grid for technology.
Reza (15:44)
Yeah
Seyi (15:58)
and losing sight of the fact that we used to generate power to serve people in cities. We've taken our eye off that goal and we're consequently chasing after the conversion of power into intelligence and ignoring the original human intelligence.
It is the real fundamental issue we have here. We will end up with collapse if we keep doing that. And I think the main thing cities, there are some cities, don't get me wrong. The Virginia area of the US is seeing a lot of data center activity and AI activity because think about it, the fiber that comes into the US comes from there. So.
It really just boils down to plunking data centers right at the point of the network that you can then distribute things out. but there are one or two towns and leaders in some towns. There's this lady who is on the city council. I can't remember her name and I forget the name of the
town now. She is one of those standing firm in the need for decision makers, policymakers to recognize this, the failing infrastructure and the impending collapse and shift our eyes back to what do the people need in these communities that we're stopping, that we've stopped paying attention to. Yeah.
Reza (17:32)
Yeah, because this is like, who's bearing the cost of doing this, right? It's the communities that aren't benefiting from this. so you're prioritizing these data centers that are turning power into knowledge over the people that make that community thrive. So there's this real tension that's being created where...
Seyi (17:38)
Yeah. Yeah.
Yeah.
Reza (17:57)
You know, like you said, you have your eye on these data centers, but you're taking your eye off of the community that you're serving.
Seyi (18:04)
It's so true. We're focusing on the reliability of power for technology at the expense of reliability of power for humans. it's, it's, and I think
Reza (18:10)
Yeah.
That's a really good way of putting it.
Seyi (18:19)
electricity and AI data and data center power usage in this case actually represents something much bigger Reza. It's this allocation of resources and how our leaders and how business is deciding who gets what resources.
Reza (18:28)
Yeah.
Yeah.
Seyi (18:40)
beyond just power. It's the same with water. It's the same with land use because these data centers are taking up space in communities where in some cases you have huge homelessness. Let's not even talk about that. We're converting some of the old buildings that were business locations. And I see that up in Round Rock, Texas here. We're converting some of the
Reza (18:43)
Right.
out.
Seyi (19:07)
big buildings that could have become multi-unit homes, for example, into co-location data centers where only about six people work, but it's taking up 60, 70 megawatts of electricity and a bunch of water that would have otherwise gone to serving the people in those cities. We can't keep doing that.
Reza (19:29)
Yeah,
that's a really good way of making that contrast. And Jay, but you've also talked about solutions, right? And this is the strategic foresight portion of the episode. So as you look forward five years out, first off, I have two questions here. One is where do see this going? And then the second question is going to be about
Seyi (19:43)
Yeah.
Reza (19:54)
policy changes that could be made, but where do you see some of this surge in energy demand going? Are we gonna hit these bottlenecks, these brownouts? How are we gonna manage through it? And then what are the policy implications or choices we should be making?
Seyi (20:09)
Yeah, a few things and the policy and the technology slash innovation sort of go hand in hand here. One of the policy recommendations that I believe need to come in place is just a little bit more flexibility in or yeah, flexibility in how we
allocate land and demand to places in cities. It's so fascinating. We end up with these hubs.
immediately. And I get it. It's about efficiency. It's about having all the resources in one location and one area. And then it makes it more convenient for new data centers to come to that area. But I'd suggest that there should be a distribution, geographic, but also resource utilization in the policy making so that
When Round Rock, and this is where some collaboration needs to come in, when Round Rock decides, you know what, we're letting a few data centers here, make that decision in collaboration with Hutto, Buddha, and pick Gerald and some of the other towns around here, be flexible in
collaborating with your neighbors is maybe the, so that you all make joint policies. And I know you have one of our laws that addresses this. So please share that now, yeah.
Reza (21:33)
Yeah.
Yeah, well,
it's that regional cooperation, right? It's the one that where we talk about for cities to thrive, regional cooperation is enhance the sustainability. So sustainable cities cannot exist in isolation. They must coordinate with surrounding municipalities and rural areas for truly sustainable regional development, which is what you're talking about. These various cities in this region need to cooperate together.
Seyi (21:58)
Yeah. Yeah.
Reza (22:01)
And this is based on research that Wheeler did arguing about the importance of regional approaches for urban sustainability, because there's this inefficient use, right? This inefficient resource usage if you're not doing it in a sort of in a regional cooperative way, so.
Seyi (22:08)
Yes. Yes.
Absolutely, absolutely. And then on the technology side, thank you for sharing that because the law really ties in immediately there. On the technology side, a second sort of idea is that the nuance of AI is that you have sort of three buckets of work that you're doing. There's the model training, we call it compute.
Reza (22:21)
Yeah.
Seyi (22:39)
but there's the model training, there's inference, and we want to compute model training inference. And I believe there's another one, but it's failing me now. My point is, depending on the different type of work,
Data centers and AI companies should also embed some flexibility into their demand such that if you don't have to train your models with power being used at a data center that is in the middle of the city, for example, geographic load distribution should be a consideration so that you're not taking
power from a city at a point when people just need AC. For example, for example. Absolutely, absolutely. Consideration, yeah. Which the utility has done for a while. Exactly, exactly.
Reza (23:24)
Yeah, so it's like time of use, right? Time of use, yeah, ⁓ off peak. demand side management is such a common thing that utility does.
Seyi (23:38)
Exactly. So that is one that I absolutely think should be factored in because the data centers and the AI workload requests almost always come in the form of we just need reliable power. I think the tier four data centers require 99.9999 % reliability.
of the power that's coming in. Do you though? Maybe there's a flexible time of use, geographic load distribution thing that can happen there. And I feel strongly about that one because what can also happen, which ties to the second part, a lot of the AI data centers use diesel generators as backup. One, we should.
stop that honestly because it's just not sustainable. You can't claim net zero ambitions and then burn diesel to power your AI. It's actually quite backward when you think about it. It's such a prehistoric way of your burning fossil fuels to
Reza (24:33)
Yeah. ⁓
Yeah.
Yeah.
Seyi (24:54)
create more intelligence than we've ever experienced. That was the other thing, you know? ⁓ But I think we can move to batteries as grid support, which then allows those data centers to become resilient, flexible load on the grid. So what am I saying? If we're having a super hot day,
Reza (24:59)
Yeah.
Yeah.
Seyi (25:19)
and you need power to go from the grid to homes, but the supply on the grid is not enough. The grid can call on the backup batteries from the data center.
Reza (25:35)
Mmm.
Seyi (25:37)
to use that power on the grid. That is not an option that is being taken advantage of as much as is possible currently. And I think that would be amazing in its... So you turn this demand-based asset into a grid-supporting, grid-firming, resilience-building asset. Yeah.
Reza (25:40)
Yeah.
Yeah, resilient. yeah.
This is great, Seyi. mean, this has been really eye-opening for me. And if you've gone from talking about the incredible demand that data centers are putting on the grid and the incredible load and the reasons why, we talked about the impact that it's having on our urban infrastructure.
which could lead to, you talked about the collapse and the brownouts and why these pressures are really something we need to pay attention to. And then it was really revealing how you talked about the trade-offs with how we're sort of, we have our eye on data centers, turning this power into knowledge, but we're forgetting about the humans that are part of our communities that are gonna be impacted if we lose sight of
what their needs are and what our community needs are. And I really like how you put the policy and the technology opportunities, which, you know, with the policy one, you know, we picked our law, which is the regional cooperation, but the technology opportunities as well for us, not just with, you know, how
AI can become more efficient, but how these data centers and their energy uses can be done in ways that are not as impactful. I sense an optimism of where this goes. There is a huge opportunity that comes with AI. We have these mixed feelings about it, but it seems like with our human ingenuity, we're figuring our way through
Seyi (27:15)
Absolutely.
Reza (27:28)
how we can tackle this challenge that we have ahead.
Seyi (27:31)
I totally agree. Thanks for summarizing and framing that. The point I'll make, which highlights the optimism I have about this, is that with AI, we can actually design the grid we want. I genuinely believe with AI putting in some constraints that suggest, hey,
Reza (27:46)
Mm.
Seyi (27:56)
Just as an example, we should never take power from a human.
Reza (28:01)
Yeah.
Seyi (28:01)
to
serve a data center as an example. And we put those constraints as we build the models and then design the grid of the future that we want. We actually can do that now with the power of AI. And why wouldn't we? Now we have that opportunity. So I'm pretty optimistic. I caution about the problems, but I'm also optimistic that we will figure it out.
Reza (28:10)
Yeah.
Yeah.
Yeah.
Yeah, yeah. This has been great. Shay, any other things that you want to share before we close on this topic?
Seyi (28:34)
No, no, just, I think I'll do what you typically do, which is suggest a call to action for people. As you drive around your city, your town, look for these non-descript gray buildings now that are popping up in a lot of cities. In some cases in the middle of the cities, look for those and...
Reza (28:51)
Yeah.
Seyi (28:57)
be curious enough to figure out is that a data center? What impact will it have? Engage with the civic leaders to understand because a few cities are.
chasing the money that comes from allowing these data centers to come in at the expense of their citizens. But they need to hear from their citizens, their residents that we probably need to have a little bit more consideration before we put this in. So our listeners pay attention and participate is the message.
Reza (29:13)
Yeah.
Yeah, I love that call to action shade. That's a great one. We like our readers, our listeners to think about the implications of what we talk about. you know, like we do, we also ask our listeners, like if you have a mailbag that you want something to share, you're curious about something that we talked about, or you have other information that you should think that we should consider, please send that in. We're at hello at futureforward.fm.
And with that, Seyi, thank you to our listeners. We always ask you to share this with someone that you know will enjoy it. We also appreciate if you could like and subscribe, rate and review. That helps our listenership grow and folks find us. So with that, Seyi, on this one year anniversary, I'm so glad that I could...
Seyi (30:18)
and.
Reza (30:20)
get to ask you about this topic and get your expertise to share with our listeners. Here's to many, many more episodes to come.
Seyi (30:22)
you
Yes, and thank you for thinking of this as an episode that was an area that is near and dear to my heart. It's been for a long time. So I appreciate you Reza and to our listeners, thank you. And to Reza, thank you so much. This has been fun. I'm looking forward to the next year here.
Reza (30:45)
Yeah. All right.
Thanks, Jay. All right, everyone. Bye.
