Here's the number nobody's fully added up yet: every gigawatt of data center capacity announced this year comes with a hidden bill — the cooling, the power delivery, the sheer electricity these buildings burn just standing there, doing nothing but keeping themselves cold. Get the design wrong, and that bill lands on somebody. The developer eats it in wasted capital. The utility eats it in strained infrastructure. Or you eat it, on your monthly power bill, if nobody along the way was careful about who pays for what. So what actually separates an efficient data center from a power hog? Today on Concrete Compute, we're going to find out from the people whose entire job is answering that question. It's a quiet day on the wires, so we're doing something different — a Buildout 101 on efficient data center design, straight from the Department of Energy's own playbook. Welcome back to Concrete Compute, your daily brief on the AI infrastructure buildout. It's Tuesday, September 1st, 2026. Let's get into it.

So let's start with who's actually talking here. The guide comes from the Department of Energy's Federal Energy Management Program — that's the office inside DOE whose whole job is squeezing waste out of federal buildings — working together with the National Renewable Energy Laboratory, the government's flagship renewable energy research shop. Together, back in July of 2024, they put out a best practice guide for data center design. And here's the detail that actually caught my attention: they didn't just list a grab-bag of efficiency tips. According to the guide, FEMP and NREL developed, quote, 'an approach for optimizing data center sustainability, listed in order of importance,' end quote. What does that phrasing actually imply? It means somebody sat down and decided that some efficiency moves matter more than others — you don't just throw every trick at a building and hope it helps, you prioritize, because some decisions foreclose others down the line. And FEMP's own website spells out why it bothers with any of this: it says FEMP, quote, 'encourages agencies and organizations to improve data center energy efficiency in accordance with' the OMB's Smart Cloud Strategy and M-16-19 Memorandum, end quote. And separately, it states, quote, 'data centers offer a tremendous opportunity for energy and cost savings,' end quote. Now, I'll be straight with you about what I've got in front of me today — this excerpt doesn't hand me the full ranked list, so I'm not going to pretend I can read you FEMP's top ten efficiency moves in order. That would be me inventing specifics the document in front of me doesn't actually contain, and this show doesn't do that. What I can tell you is the shape of the thinking, and why it matters way beyond federal buildings. Here's the thing about efficiency in this industry: it's not a nice-to-have side project, it's load management. Every watt a data center wastes on inefficient cooling or oversized power distribution is a watt that still has to come from somewhere — a gas plant, a grid interconnection, a transmission line that took years to permit. So when a federal lab spends time ranking efficiency strategies by importance, what they're really doing is triaging where the biggest megawatt savings live, and which decisions have to happen first because everything downstream depends on them. And that's true whether you're the government running a data center or a hyperscaler running dozens of them across multiple states. Now, why would the federal government bother writing this guide at all, rather than leaving it to industry? Because Uncle Sam runs plenty of data centers itself, and that same opportunity for savings applies to taxpayer dollars just as much as it applies to a hyperscaler's balance sheet. An inefficient federal data center is money out of your tax bill, the same way an inefficient private campus eventually shows up in your monthly power bill if nobody's careful about who pays for what. The guide's audience is federal agencies specifically, so some of what's in there is written for procurement rules you and I will never deal with directly — approval chains, budget cycles, agency mandates that decide who signs off on a new server room. But the underlying physics doesn't care who's signing the checks: power in, useful compute out, everything else is waste, and waste has a cost that somebody, somewhere, is going to be handed. And that's the part worth sitting with for a second, because it means efficiency isn't really a separate topic from the gigawatt arms race you hear about every other day on this show — it's the quiet variable sitting underneath all of it, determining how many of those announced megawatts actually turn into useful work once the building is finished.

So here's my read on why a document like this is worth your morning today. This industry has spent the better part of two years obsessed with one number — gigawatts announced — and almost no airtime on the number that actually determines whether those gigawatts get built on time: how much of that power gets wasted before it ever reaches a GPU. My view from the engineering and economics is that time to power is the buildout's real bottleneck — not chip supply, not money, but transformers, interconnection queues, and substation permits that take years to clear. Efficient design is the one lever in that whole chain a developer controls completely, on their own timeline, without waiting on a utility or a grid operator to say yes. You can't will a transformer into existence faster. You can absolutely design a building that needs fewer of them, or that sizes them correctly the first time instead of overbuilding and then discovering the load never materializes. And that's really what FEMP's order-of-importance framing is getting at, even in the short excerpt we have today: efficiency isn't a checklist you run through after the design is locked, it's a set of decisions made in sequence, because getting the big structural stuff wrong early means no amount of after-the-fact tinkering saves you. You can bolt on better fans later. You can't retroactively change where you put the building, or how the airflow moves through it, without tearing the thing apart. Now, I want to be honest about the limits of what I can tell you today. This is a federal best-practice guide, not breaking news — there's no fresh number, no new project, no company to hold to a standard here. So the skepticism I'd normally aim at a splashy gigawatt announcement doesn't really apply, because there's no press release inflating anything. The question isn't whether this document is hype. It's whether anyone in the private buildout is actually reading it. Because for all the money getting poured into this industry, a two-year-old, free, publicly available federal guide on how to not waste power is exactly the kind of unglamorous document that never makes it into a keynote. Nobody puts 'we read the FEMP guide' in a press release. But the developers who quietly follow that order-of-operations discipline are the ones who need fewer megawatts, wait less time in the interconnection queue, and hand their utility — and their neighbors — a smaller bill at the end of it. That's the beneficiary-pays principle that should guide how infrastructure costs are assigned, just approached from the design side instead of the financing side: the cheapest megawatt is the one you never had to build in the first place, and good design is how you get there before a single kilowatt-hour gets billed to anyone. And ask yourself this — why doesn't every developer lead with efficiency in their pitch to a county commission? My guess is it doesn't sound as impressive as a gigawatt figure in a press release, but a commissioner deciding whether to approve five hundred acres in their county should be asking about it anyway, right alongside the water use and the tax abatement, because efficiency is the thing that determines whether the promised power draw ever actually shows up as promised — or whether the building ends up drawing more than anyone planned for, and the neighbors find out the hard way. Time for the Hype Check. On substance, I'd put this one at an eight. There's no hype to check, really — it's a government engineering document doing exactly what it says on the label, laying out efficiency priorities in order of importance. The only reason it's not higher is that the excerpt in front of me today doesn't give me the actual ranked list, so I can't tell you exactly where FEMP thinks the biggest savings live. That's the document I'd want next. So if a fuller version of that ranked list ever surfaces on this show, and the very first priority turns out to be something structural — building shape, cooling architecture, load matching — rather than a bolt-on fix, that tells us the industry's biggest efficiency gains are still sitting in decisions made at the drafting-table stage, before a single shovel goes in the ground.

Now, before I let you go — if today's slower pace was useful, an explainer episode like this is exactly the kind of thing that's easy to miss if you're not already following the show. So if you got something out of it, follow Concrete Compute wherever you listen, and tomorrow's episode will be waiting when the news picks back up. This has been Concrete Compute, an AI-voiced podcast, created and built by a real human using today's cutting-edge technology. Nothing you heard on this show is financial advice. I'm Brian Lampert, and I'll catch you all tomorrow — take care!