Latest / The Operations Podcast with Fexingo: Process, People, and Profit in Modern Business / How a Data Center Cut Cooling Energy by 40 Percent
Transcript
- Lucas: You ever think about where your cloud files actually live? The photos, the streaming, the AI queries — they're all sitting in a building that's essentially a giant oven full of servers. Luna: Right. And those servers generate a lot of heat. So you need cooling, which takes energy — a lot of it. Lucas: Exactly. Data centers consume about 2 percent of global electricity. And roughly 30 to 40 percent of that goes to cooling, not computing. So if you can make cooling more efficient, you unlock huge savings — both financial and environmental. Luna: So what's the state of the art? Are we just talking better AC units? Lucas: Not anymore. The big leap right now is liquid immersion cooling. Instead of blasting cold air through the floor, you dunk the servers in a dielectric fluid — basically a non-conductive oil — that absorbs heat directly. One company called EdgeCore Digital Infrastructure did exactly that at a facility in Northern Virginia. Luna: Northern Virginia — that's basically the data center capital of the world. Ashburn, right? Lucas: Exactly, the so-called 'Data Center Alley.' EdgeCore built a 60 megawatt facility there and decided to pilot immersion cooling on a section of their compute floor. They started with a single row of servers, about 48 racks, retrofitted with immersion tanks from a company called GRC. Luna: So what were the results? Lucas: They cut cooling energy by 40 percent on that pilot row. Overall facility power usage effectiveness — that's PUE, the ratio of total energy to IT energy — dropped from 1.6 to 1.12. For context, 1.0 is perfect — all energy goes to computing. A PUE of 1.12 is exceptionally low. Luna: That's a massive improvement. What does that mean in dollar terms? Lucas: For a 60 megawatt facility, each 0.1 improvement in PUE can save roughly one point five to two million dollars a year in electricity costs. So going from 1.6 to 1.12 — that's a 0.48 improvement — translates to roughly seven to nine million dollars annually, just on the cooling side. Luna: Wow. But there's a catch, right? Immersion cooling isn't cheap to install. Lucas: Right. The upfront capital cost is higher. EdgeCore spent about 40 percent more on that row compared to traditional air-cooled setup. But the payback period was under 18 months because of the energy savings. Plus, the servers last longer — no dust, no vibration from fans — so hardware replacement costs go down. Luna: So it's a capex versus opex trade-off that favors the long-term. What about the AI angle? I know AI training runs generate insane heat. Lucas: That's actually the driving factor. AI chips like NVIDIA's H100 can draw 700 watts each, and racks can exceed 40 kilowatts — far beyond what air cooling can handle. Immersion cooling can dissipate over 100 kilowatts per rack. So for hyperscalers building AI clusters, immersion isn't optional — it's becoming necessary. Luna: So EdgeCore is ahead of the curve. But is this scalable? Can you retrofit existing data centers? Lucas: It's easier in greenfield builds — new construction. Retrofitting is trickier because you have to drain fluids, re-plumb, and deal with weight — immersion tanks are heavy. But EdgeCore's model is modular: they built the immersion row as a standalone pod that can be replicated. Their plan is to scale to 50 percent immersion across the facility by 2028. Luna: That's ambitious. What about the fluid itself? Is it safe? Lucas: The dielectric fluid is non-toxic, non-flammable, and has a high flash point. It's essentially a synthetic oil similar to what's used in transformers. But there are maintenance challenges — fluid needs to be filtered and tested, and if a server fails, you have to lift it out of the tank, which is a different workflow. Luna: So the operations team has to be retrained. That's a people challenge. Lucas: Absolutely. EdgeCore invested heavily in training — they sent technicians to GRC's certification program. And they redesigned the floor layout with gantry cranes to handle the heavier immersion pods. It's not just a technology change; it's a process change. Luna: I've also heard about two-phase immersion cooling, where the fluid actually boils and condenses. Any experience with that? Lucas: EdgeCore stuck with single-phase — the fluid stays liquid — because it's simpler and requires less maintenance. Two-phase is more efficient but more complex; companies like Microsoft have experimented with it. For now, single-phase is the sweet spot for most operators. Luna: So what's the broader takeaway? Will all data centers go immersion? Lucas: Not all, but I think for high-density compute — AI, HPC — it's inevitable. The hyperscalers — Google, Microsoft, Amazon — are all testing immersion or direct to chip liquid cooling. But for traditional cloud workloads, air cooling will still dominate for a while. The key is matching the cooling method to the density. Luna: It's a reminder that efficiency gains often come from rethinking the physical layer, not just the software. Lucas: Yeah. And that's a theme we return to a lot on this show — that operational innovation often hides where you least expect it. Speaking of which, if these conversations have moved your work forward in some small way, listener support is what keeps this show ad-free and independent. You can find us at buy me a coffee dot com slash fexingo. No pressure, just wanted to put it out there. Luna: It really does make a difference. We read every message. Lucas: Anyway, back to cooling. One more thing worth noting: EdgeCore also integrated ai driven airflow optimization in their air-cooled section — using sensors and machine learning to adjust fan speeds and vent openings in real time. That alone gave them an additional 8 percent reduction. Luna: So it's a layered approach — immersion for the hottest zones, AI for the rest. Lucas: Exactly. And that's the smartest play. You don't have to go all-in on one technology. You apply the right tool to the right heat density. The result is a facility that's both more efficient and more resilient. Luna: Seems like a blueprint for the next generation of data centers. Lucas: I think so. As AI workloads scale, cooling will be one of the biggest differentiators between efficient operators and everyone else. EdgeCore is showing that with the right mix of technology and process, you can cut cooling energy by 40 percent — and that's just the beginning.