How Edge Computing Cuts Data Center Cooling Costs

Lucas: So there's a number that's been stuck in my head all week: cooling can account for up to forty percent of a data center's total energy use. And when you think about the edge — those small, distributed nodes sitting in a factory or a cell tower — you realize they sidestep a huge chunk of that entirely. Luna: Forty percent? That's massive. But isn't edge compute still running in some kind of enclosure? They still need to keep the silicon from melting. Lucas: Right, they still need cooling, but the scale changes the game. A central data center might have thousands of servers in one building, and you have to cool the whole space. An edge node might have a handful of servers, sometimes just one or two, and you can cool it with a fan or even passive heat sinks. The economics and the engineering are just fundamentally different. Luna: So it's not that edge compute doesn't need cooling, it's that the cooling problem is smaller and simpler. But what about the big data centers? They're not going away — they're still the backbone. Lucas: Exactly. And that's where the innovation is happening. I was reading about a facility in Sweden that pipes its waste heat into the local district heating system. So the data center is literally warming homes in the winter. That's not just efficiency, it's a revenue stream. Luna: That's clever. But not every location can do that. And I've heard about data centers in the desert using water for cooling — that feels like a resource conflict. Lucas: It is a real tension. There's a facility in Arizona that uses treated sewage water for evaporative cooling. It's a smart reuse of water that would otherwise go to waste, but it still uses a lot of it. And that's a big reason why edge compute is attractive in arid regions — you avoid the water problem altogether. Luna: So the edge is a way to dodge the water bill and the energy bill. But what about the performance side? Don't you lose something by not being in a massive, optimized facility? Lucas: You do trade some raw density, but you gain latency and resilience. A node on a cell tower can process a request in milliseconds, and if the central cloud has an outage, that edge node keeps working. It's a different trade-off, and for a lot of use cases — think autonomous vehicles or industrial robots — that trade-off is worth it. Luna: Speaking of trade-offs, I've been reading about immersion cooling — where you dunk the whole server in a dielectric fluid. That sounds wild, but it's supposed to be way more efficient. Lucas: It is. With immersion, you can get heat transfer that's orders of magnitude better than air. That means you can run servers at higher densities, and some companies report cutting cooling energy by over ninety percent compared to traditional air cooling. But there's a catch — it's messier, it's harder to service, and it's not always practical for an edge node. Luna: So immersion is more of a central data center play, not edge. But it's still a way to cut cooling costs where the big clusters live. Lucas: Right. And that's the broader point: the industry is moving toward a mix. You have hyperscale data centers adopting immersion and free-air cooling in cool climates, and you have edge nodes that just don't need that heavy infrastructure. The result is that the average cost per unit of compute is dropping, and cooling is a big part of that story. Luna: You know, I've been thinking about our own setup here — we record this podcast on a single laptop, and it barely gets warm. We're kind of running an edge node in our studio. Lucas: Ha, exactly. And that's the thing — not every workload needs a data center. A lot of the time, local compute is more than enough. And that's what keeps this show going, by the way. It's a small operation, and a couple of dollars a month from listeners genuinely makes a difference. If you've gotten something useful from these conversations, you can support us at buy me a coffee dot com slash fexingo. It really does keep the lights on. Luna: Yeah, and we keep it ad-free because of that. Every little bit helps us keep digging into these topics. Lucas: So, back to those numbers. There's a concept called power usage effectiveness — PUE — which is the ratio of total facility energy to IT energy. A typical data center might have a PUE of one point five, meaning fifty percent extra energy for cooling and other overhead. But the best-in-class facilities are getting down to one point one or even closer to one. Luna: And edge nodes — what's their PUE? I can't find solid numbers because they're often just part of a cell tower or a factory. Lucas: That's the thing — they don't even measure it the same way. An edge node might have a PUE of one point zero five because it's just a box with a fan. The infrastructure is shared with the tower or the building, so the overhead is minimal. That's a huge advantage when you're deploying hundreds of them. Luna: But wait — if you have hundreds of edge nodes instead of one data center, aren't you multiplying the hardware and the cooling fans? The aggregate energy might be higher. Lucas: That's a fair challenge. It depends on the workload. If you're processing video feeds from security cameras, sending all that data back to a central data center uses a lot of network energy. Doing it locally at the edge can actually be more efficient because you're not moving gigabytes of data over the wire. The math isn't always obvious, but in many real cases, edge wins on both latency and energy. Luna: So it's about the data gravity — you put compute where the data is created, and that reduces the cooling burden indirectly by reducing network load. Lucas: Exactly. And there's a nice example: a manufacturer with a robotic arm that needs real-time control. If you run the control algorithm on a local edge server, you get sub-millisecond response times and you don't need a massive data center cooling system for that one function. It's a small box that sits in a cabinet, and it just works. Luna: And what about the environmental side? I've seen reports that data centers are a growing share of global electricity use — something like one to two percent. Does edge computing help or hurt that? Lucas: It's not a silver bullet. You still have the same silicon doing the same computations, so the total compute energy is roughly the same. But you can reduce the cooling overhead and the network energy. Some studies suggest that edge computing could cut overall data center energy by up to thirty percent, mostly because you don't need to cool and power a giant facility 24/7. But those studies make assumptions, and the reality varies. Luna: Let's bring it back to a concrete case. There was that Swedish data center — what's the actual name? I remember reading about it. Lucas: It's called Stockholm Exergi, I think — they partner with local data centers to feed waste heat into their district heating network. One of the best-known examples is the Bahnhof data center, which actually sits inside a former nuclear bunker under Stockholm. They've been using the heat to warm nearby buildings for years. It's a classic example of circular infrastructure. Luna: That's a great example. And it shows that cooling doesn't have to be a cost — it can be a resource. But that's easier in a cold climate. What about tropical countries? They're building data centers too. Lucas: That's where immersion cooling gets interesting. In a hot, humid climate, air cooling is inefficient, and water is often scarce. Immersion cooling uses a dielectric fluid that doesn't evaporate, so you can run closed-loop systems that recycle the heat. Some companies are even using that heat for desalination or other local needs. It's early days, but the potential is huge. Luna: And for edge nodes in those regions, you'd just rely on passive cooling or small fans, right? No need for chillers at all. Lucas: Right. A cell tower in the Philippines might have a small edge server with a heat sink and a fan. It's low-cost, low-maintenance, and it doesn't need a backup generator just for cooling. That's a huge operational advantage in remote locations. Luna: So the edge isn't just about latency — it's about simplifying the entire physical infrastructure. That's a compelling story for any business looking to cut costs. Lucas: And that's why you see companies like Amazon and Microsoft investing heavily in edge compute, not just for the speed but for the efficiency. They're building out distributed networks of small nodes that can handle workloads locally, and they're designing them to be energy-sippers. The cooling bill for a million nodes is a lot different than for a million servers in a data center. Luna: But let's be honest — there are still challenges. Managing and maintaining thousands of edge nodes can be a logistics nightmare, and security is harder to centralize. Is that worth the cooling savings? Lucas: That's the real trade-off. You're trading a big physical footprint for a distributed one, and that comes with its own headaches. But the cooling savings are often enough to justify it, especially for companies with high bandwidth needs. And as edge management platforms mature, those operational costs are coming down. Luna: So what's the one concrete takeaway for a business listener today? If they're thinking about infrastructure, what should they ask themselves? Lucas: I'd say: look at your workloads and ask which ones really need to be in a centralized data center. If a workload is latency-sensitive or produces a lot of data locally, moving it to the edge could cut your cooling and bandwidth costs significantly. It's not about replacing the cloud — it's about being strategic about where compute happens. Luna: That's a good note to end on. It's not either-or, it's both, and the cooling story is a big reason why. Lucas: And as the edge grows, we'll see more creative solutions for the data centers that remain. The future of cooling is a mix of immersion, heat reuse, and smart placement — and the edge is what makes it all possible. That's something I'll be watching closely.