I was catching up on the September 1st Links with Friends episode, and Wendell brought up the EPA looking to exempt data centers from air pollution disclosures. He made a point that data centers traditionally haven’t been massive local air polluters, but that’s changing now that some sites are spinning up their own on-site power generation (like Musk’s natural gas generators).
This brought up an old memory. During the early days of Tesla Superchargers, there was so much “barbershop talk” about how the power grid couldn’t handle it and the major cities in California were going to grind to a halt. That apocalyptic scenario obviously never happened. I tried researching the actual grid impact back then to separate fact from fiction, but just got totally lost in the weeds.
Now, with the huge boom in AI and data centers, I’m seeing the exact same type of panic. Since I do small work in IT and run a small homelab, I want to cut through the sensationalism and understand how this infrastructure actually works at scale.
The problem is, Googling this right now is a difficult. I’m just hitting a wall of AI-generated summaries and SEO fluff, and it’s tough to find a good primer before diving into heavy articles.
I’d love to get a better grasp on the actual logistics of standing up a data center. Specifically:
Location and Efficiency: What actually makes a modern data center efficient? What factors dictate whether a specific location genuinely makes sense for a massive power draw?
Legit Builds vs. Cash Grabs: Wendell mentioned municipalities using water rights to block builds. When a data center is proposed in a populated area, how can you tell if it’s a well-planned infrastructure project that makes sense for the region, versus investors just trying to cash in on AI funding/subsidies with zero regard for local resources (like water consumption and power strain)?
I’d appreciate any insights, or if anyone has some solid resources to read up on this so I can go down a proper rabbit hole.
What makes you think that efficiency is even a goal to begin with?
AI datacenters are very power-intensive. Maybe if you use the waste energy for teleheating, you could argue that a datacenter is efficient.
Every W you create inside a datacenter, you also have to get rid of. If you do that with an AC, you probably need 1W for every 1W created. That is why most datacenters, half the power consumption is for cooling.
So probably the most efficient are datacenters that don’t need ACs but only use fans to cool. ungleich.ch is one such company. To quote from their webpage:
How does a data center work without cooling?
Instead of using high-density racks and putting them next to each other, we use the available space of the old factory buildings. Instead of hosting 40 servers on one square meter, we host one server on 4 square meters. We can do this because factory halls tend to be huge. And in Glarus, they are also available.
You probably can’t and it is also not my job. That is IMHO the job of the region and also the state. We had people complaining in Texas about BTC mining noise, which is IMHO a Texas issue. And I say that as someone who dislikes BTC.
It is not the fault of private companies that you have no consumer protection or regulations. Don’t hate the player, hate the game.
Regarding the grid: in theory, datacenters could provide a more stable grid. Just like with BTC mining, where the miners sometimes made most of their money by shutting down their systems.
If a datacenter goes online, the total consumption grows, heck the total revenue grows, and if you also have a contract to shut them off to shave off demand spikes, everyone should be happy. However, it seems like at least in Texas, that they got ripped off by miners because of bad contracts.
So my answer is: it depends And unfortunatly I can’t provide you any good sources of information on that topic.
This scenario did not happen because EVs did not take over the roads completely like was talked about. They are still a small fraction of total vehicles and the rate at which superchargers were installed was slow enough that it did not overwhelm the grid, and the grid has been upgrading and changing slowly as well so things are accommodating. In the end it was more doom and gloom talk than reality, because the talk of EVs replacing all vehicles in a year or two was never going to be realistic. California made a law that all vehicle sales in the state must be 1/3 EVs by 2026, 2/3 by 2030, and 100% EVs only by 2035. We will have to see how that is enforced at the end of this year and what the sales numbers actually were for the state as a whole.
In doing a search right now, it appears only 15.9% of vehicle sales in CA this year have been EVs, a long way off from 1/3. I also just saw this:
The federal government has since rescinded California’s authority to enforce the ban, though California is actively challenging this decision in court.
Yeah id agree that efficiency is a kind of made up meaningless metric right now. Efficient per what? Unit of imaginary work they do?
The powercolapyse was a real thing, the solution was not connecting the new gigawatt datacenters to the grid at all. They just built lines of their own power plants right on site with the datacenter.
Water and noise are the next insurmountable problems. Watts in = watts out and cooling 150kw per rack DC footprints are insanely thirsty. Maybe not as thirsty as alfalfa fields, but you cant eat tokens, so… That said, the deafening noise inside the datacenters require hearing protection to just walk around. Outside its obnoxious and nobody would want to live near them.
I guess I just want to be more informed. When developers say “we’re not like the other guys, we’re power efficient”, I realize I don’t really know what that means to them. For me, it’s just about filling the gap between what they say they’re doing and what they’re actually doing, so I can understand the reality better instead of just taking “super energy efficient water cooling” marketing at face value.
That is totally fair. For me it’s less about hating what these companies do, and more about trying to critically think about their motives and the loopholes they might use to get there. I definitely don’t claim to be an expert, I’m just trying to see if being efficient is actually a genuine goal for them or if it’s just something they say to get approved and after approval what ways they circumvent regulations.
I get that, thank you! The idea that data centers could actually stabilize the grid by shutting down during demand spikes is an interesting angle to consider, never considered that to be a viable way to reduce the impact.
Thank you for the insight, that makes a lot more sense. The whole doom and gloom narrative seemed a bit off to me during those early EV days, but it’s interesting how now that it’s being brought up again, it seems to just get disregarded. I appreciate you breaking down why the EV scenario played out the way it did. Makes a lot more sense
I guess that opens up another question: how are they even measuring efficiency, and what is the actual value of that efficiency when they claim it?
I actually wasn’t aware that some of these gigawatt data centers were just building their own power plants on-site to avoid the grid entirely. It’s really hard to tell where these power connections are being made just from a quick Google search, which makes it even harder to understand the real impact.
And you make a great point about the water and noise. I’m honestly not sure if real solutions are being worked on for those issues, or if they’re just being actively avoided right now.
While cooling methods and operating efficiency of the data centers/gpu’s themselves is important, the power generation side is where the efficiency gains/losses are most noticeable.
A lot of gas turbine manufacturers offer mobile deployments of aero-derivative turbines that don’t require foundations in some scenarios and can be producing power in mere days after delivery. These are the instances where I think it shows the hastiness of a cash grab.
The grid in the United States has been neglected in many areas for decades and it is completely unprepared for the massive power draw of a data center. Single cycle power plants typically have an output of under 300MW per turbine so for a 1GW data center alone you need 4ish turbines. That’s why there are often dedicated power plants for massive data centers, the grid in most areas can barely handle the load of residential/commercial/light industrial as is.
The X AI data center i believe used around 50 mobile deployment turbines which do not have the same Nox output performance as the current generation of simple cycle turbines. Adding combined cycle (steam turbines that use exhaust heat to boil water) massively boosts the thermal efficiency of these plants but this takes significantly longer to engineer and implement. This is time that investors probably don’t want to spend.
That was the kind of example I was looking for. The consideration of those mobile turbines versus taking the time for a proper build really makes the “cash grab” aspect make sense.
The idea that the timeline for ROI drives them to bypass the cleaner, more efficient engineering gives a huge perspective on what they seek and what they want to do at the time. That provides perspective on my question on how to spot the difference between a real infrastructure project and a rushed deployment. Appreciate the detailed breakdown
Ohh while we are at that topic, the initially got the permission to fire up something like 25 of them. People asked “hey, why are there 50?”
X claimed that “don’t worry, only 25 are running”.
Experts said “that is bs, nobody has these expensive turbines on standby”.
So a private person simply put a thermal image camera on a drone and tadaaaa: 49 are running.
I unfortunatly don’t know what happened afterwards.
Gas turbines emissions are a serious issue. Another underrated issue is infranoise you can’t hear but is blearring at 80dB. And since its frequency is so low, you would have to build very thick walls to stop it.
IMHO you need good and strict regulations, and somebody willing to enforce these rules. If I were the mayor or whatever of that citiy (sorry, not from the US) I would force the X datacenter to shut down all but the agreed 25 turbines. If a police raid is needed for that, so be it.
Most of the efficiency claims are pr bullshit. However they define efficiency, its still a site requiring dozens to hundreds of megawatts of power and thats a number that local power grids are often not built to accomodate on a whim (having so many megawatts of spare capacity doing nothing is expensive, thats a bunch of power plants worth).
The risk of collapsing the local power infrastructure is very real in a lot of places.
This is the simple fact here, we are talking of facilities that can eat the whole output of a large conventional power plant, if not more. The time necessary to build said power plants is measured in years, they cannot wait that long during a gold rush phase of the market. So they are cutting all the corners they can and deal with the fines later
For example, its projected that by 2030 datacenters are going to double their current power demands and effectively use more power than every other sector combined. This is a completely insane thing to handle and makes the EV projections look like a fart in the wind
projects that electricity demand from data centres worldwide is set to more than double by 2030 to around 945 terawatt-hours (TWh), slightly more than the entire electricity consumption of Japan today. AI will be the most significant driver of this increase, with electricity demand from AI-optimised data centres projected to more than quadruple by 2030.
In the United States, power consumption by data centres is on course to account for almost half of the growth in electricity demand between now and 2030. Driven by AI use, the US economy is set to consume more electricity in 2030 for processing data than for manufacturing all energy-intensive goods combined, including aluminium, steel, cement and chemicals. In advanced economies more broadly, data centres are projected to drive more than 20% of the growth in electricity demand between now and 2030, putting the power sector in those economies back on a growth footing after years of stagnating or declining demand in many of them.
I am only going to speak on things I personally know about.
Whenever I have deal with a power or water company, they charge for everything. You want to build a house 5 miles away from the nearest power line, you will be paying per pole for the power company to bring in the power.
If you want to build a new warehouse and need XX MW of power and the local sub station can’t supply that, you will be paying for that upgrade.
Same goes for water.
It’s all negotiable, but it’s not free. If the power company is overselling what they can really produce, that is a separate problem.
Cities have noise and other ordinances along with other requirements on where stuff can be built. They also tend to have hearings on large construction projects.
Same with construction sites leaving the property, dairy’s and the small of cow crap not leaving the property. We have rules and regulations in place and I question on why this is not being enforced for datacenters.
I worked at a hospital that had it’s own natural gas power plant. A big one with 2 massive boilers and 3 turbines (for redundancy) along with backup generators. It was put in sometime in the 1990s when the power company wanted too much money to upgrade their power lines. It made noise, but the trucks in the loading dock made more. The main hospital building was 100 feet away and from inside, you could not hear it.
Parents have some property out in the middle of Nevada. A gas line that cuts across the corner of it is getting dug up and replaced with a much bigger line. It’s to run some turbines to generate power for some data center build. It won’t be on the grid. Someone is spending a lot of money to replace 100+ miles of underground gas line.
The demand for EV dropped off quite a bit after the initial early adapters jumped on. A lot are also charging from home with the 120V adapter.
Quartzsite, AZ has a Tesla charging station with 84 stalls. When I pass though the town and stop for fuel/food, I swing by and look at how many cars are charging. Out of 84 stalls, the most I have seen are 4 cars. There is another 10 or so charging stations across the street, and another 25 down the street some.
For an area that that is mostly a giant truck stop along 2 freeways out in the middle of nowhere, I would expect more people would be charging their EVs then that.
And unfortunatly I can’t provide you any good sources of information on that topic.