[amluto]

To me, datacenters, especially for AI (which tolerates an extra hundred ms of latency quite well) seem like an unusual form of development. Many forms of development have similar downsides: they destroy green space, they can be noisy, they compete for energy resources [0], etc. On the flip side, though, most new developments add substantial value: jobs, tax revenue, increased industry around them, local availability of their outputs, etc.

Datacenters are weird: they come with the negatives (although, per unit energy consumption, they’re relatively innocuous), but they seem to be missing most of the regional benefits. They don’t create many jobs. They pay little in taxes. They don’t actually produce anything that makes anyone else want to further develop the surrounding area. Their outputs offer little benefit to the community that wouldn’t be provided nearly as well by a datacenter somewhere else.

So I actually get why it makes sense to support development but oppose new datacenters. Or to have an added tax on datacenters so that at least some economic benefit is captured.

As an interesting contrast, a colo facility is a bit different: I want to have colo datacenters near by so that I can use their services. But this isn’t what the new development is about.

[0] Some of which are finite in a given region or are preferable not to use. (But don’t forget that more energy use = increased production in the long term, which can be a very good thing.)

[oersted]

I'm confused, do they really have such an impact? They are of course big and expensive, but surely most datacenters are relatively innocuous in terms pollution and general disruption to the area compared to any regular heavy-industry site, right? Please let me know if I'm wrong, I'm not sure.

EDIT: I do get it, it is mainly about local benefit not really about pollution or disruption, even if they are loud about that because it sounds better. Local municipalities should definitely charge significant land rents or zoning fees so that the community benefits. China has been very successful at this for decades.

EDIT: Very rough overview after some research, using Colossus 1 as reference, which is among the biggest GPU deployments. Not thoroughly verified, but it'll be around the right ballpark.

- Electricity: 150 MW live, with another 150 MW planned/studied. That is like adding a medium electric-arc steel mill or large chemical plant to the grid. A small standard power-plant can generate about that much.

- Land / space: About 217 acres and 785,000 sq ft. Footprint-wise, that is like a large factory campus or logistics park; much smaller than a mine, port, refinery, or industrial farm, but far beyond a normal commercial warehouse.

- Water: Roughly 1.3–3M gallons/day in public estimates. That is comparable to the consumptive water use of a small-to-mid steel plant or a large industrial cooling site; not refinery-scale, but locally significant.

- Air pollution: The servers are not the dirty part; the issue is on-site gas turbines/generators. That makes it more like a small gas peaker plant than a steel mill or chemical plant. Colossus 1 reportedly used up to 35 gas turbines before grid connection.

- Noise: Mainly cooling equipment, substations, batteries, turbines/generators. More like living near a substation or small power plant than near a mine, port, or metalworks.

- Traffic / logistics: Heavy during construction, then relatively light. Much less disruptive than a port, mine, farm, steel plant, or refinery, because there is no constant flow of ore, scrap, fuel, chemicals, crops, containers, or waste.

- Heat: Nearly all consumed electricity becomes heat. At 150–300 MW, the heat rejection is industrial-scale, closer to a small power station / large process plant than normal manufacturing.

[vitally3643]

The noise pollution is quite significant in the immediate area, and the heat output notably raises outdoor temperatures in a surprisingly wide area. That also of course ignores air and water pollution caused by the increased demand on electricity generation (or jet turbines spun up in the parking lot).

But the point is they suck up land and resources for no material or economic benefit to the local population. There's absolutely no reason to build these things in or even near cities. They can be built in the middle of nowhere where they don't bother anyone with zero impact on the services they provide.

[amluto]

I’m sure there is real data and real math available.

For heating: imagine a 1km^2 campus. That’s 1e6 m^2, and peak daylight is around 1kW/m^2. So peak daylight on the campus is about 1GW. (Wow, just covering the whole campus with solar panels would be pretty awesome!) If you put a 1GW datacenter there, that is equivalent to full daylight, with zero albedo, 24/7. Hmm, I’d rather live at a considerable distance from just the dissipated heat.

As for noise, there is no substantial noise emission inherent to the operation, so I expect it largely comes down to how hard the facility tries to mitigate accidental noise and how well local regulators enforce sound measurement and control. Consider a high-end Noctua or similar fan, compared to a super cheap fan of comparable RPM, flow, and static pressure — a 30dB difference in emitted sound is entirely plausible. The datacenter has some switching noise from handling its 1GW of power, but it also has lots and lots of fans and pumps.

An LLM informs me that 1kW of acoustic energy radiated into free air (no surfaces) is 79dB SPL, Z curve (unweighted), at a range of 1km. So if 1 part per million of the datacenter’s power consumption ends up as noise, it’s loud. There are all manner of corrections needed. For example, your ears’ sensitivity is much lower at non-peak-sensitivity frequencies. But the data center isn’t in free air, and the effect of the ground, the atmosphere under appropriate circumstances, and the height of the datacenter could easily dramatically increase the intensity at longish distances.

A lot of this boils down to large datacenters using immense amounts of power and that power being something you would prefer not to have redirected at you in any form.

[gruez]

> If you put a 1GW datacenter there, that is equivalent to full daylight, with zero albedo, 24/7. Hmm, I’d rather live at a considerable distance from just the dissipated heat.

Doesn't this comparison fall apart when you consider that the heat is going to be dissipated across a large area? The reason why stuff gets so hot during the day is that the surrounding areas also get heated, so there's nowhere for the heat to escape other than up. If only 1sq km is getting heated, and that's getting dissipated, the effect is going to be far less than your implied comparison of 2 suns. To bring it back to your example, a patch of payment in the sun (ie. 1kW/m^2) can get scorching hot. But in absolute terms it's less than the output of a space heater (typical one is 1.5KW), not even enough to keep a room warm.

[amluto]

Yes, absolutely, but it will depend on where one is, the surrounding geography, etc.

A hairdryer pointed at me from 1m away running 24/7 will make me notably warmer. A hairdryer 20m away is probably unnoticeable. A 1GW datacenter is a lot of hairdryers. At some scale, there is probably a buoyancy effect such that cooler air will be drawn in near the surface, get heated, and convect upwards, so I could even believe that a monster datacenter would cool some surrounding areas. (The sun inland from the west coast in the US has this effect during certain seasons, and AIUI this is a good part of the reason that the areas very near the coast tend to be dramatically cooler than inland areas in the spring and early summer.)

[Aurornis]

> and the heat output notably raises outdoor temperatures in a surprisingly wide area

There was a study posted here on this exact topic a few weeks back. The most they could measure was a little over 1 degree C near the datacenter.

> But the point is they suck up land and resources

Land use for a datacenter is kind of negligible. Even the largest data centers are barely a rounding error compared to all of the other commercial and industrial operations around me.

Like it's not even close.

> They can be built in the middle of nowhere where they don't bother anyone with zero impact on the services they provide.

For what it's worth, all of the data center projects I've looked up near me are being built in remote or industrial areas. It hasn't stopped the protestors, who are arranging for bus transport to get to the sites because they're so far away.

[jeffbee]

That paper was 100% garbage. It hinged entirely on the observation of 1 huge industrial park in Bajio, Mexico, which contains one AWS data center and about 3000 auto parts factories. The before/after breaking point of the effect was during the time that the park went from bare ground to industrial park. It has nothing to do with the data center.

[Aurornis]

Interesting. So the minimal heat rise observed was from the total transformation of the entire area and they blamed it all on the datacenter?

[jeffbee]

Yes. And just on the arithmetic it should be crystal clear that no data center is anywhere near energetic enough to heat the countryside for miles around. The effect comes from the man-made surfaces facing the sun instead of natural ground cover. Only the sun has the energy to do this.

I used the paper's data to investigate some of their claims. The top figure shows the temperature in the area surrounding Google's Oklahoma data center, in the middle of the figure. I think you can clearly see that the effect is totally dominated by the nearby coal power station and its huge, black pile of slag.

https://observablehq.com/@jwb/data-center-temperature-effect...

The scene is roughly here:

https://www.google.com/maps/@36.2154925,-95.3281217,13968m/d...

Edit: fixed notebook access. Sorry.

[oersted]

But are they really at the same level of disruption as any steel, chemical, nuclear, gas, manufacturing, recycling... plant? A mine? A port? Industrial farming?

I get it though, they don't mind if it benefits the local community, that's the issue. I suppose it does make a lot of sense for at least the local municipality to charge steep ongoing land rents or fees for the zoning license. And I suppose that requires national coordination, otherwise they'll just go the next town over, which is exactly what is happening right now.

In China, local municipalities were very profitable for decades just from selling or renting land to industrial deployments. It had a big impact on the local tax burden and they were significant net contributors to the national budget, instead of the other way around.

[smw]

I think the idea that they raise outdoor temps over a wide area is a myth spread on facebook. The physics doesn't work out.

[swiftcoder]

Just adding a square kilometre of parking lots raises the surrounding temperatures measurably, so I don't see how building a datacenter that size is going to have less impact

[expedition32]

Picture a town of 50k people with crumbling infrastructure and suddenly a tech bro wants to build his data center megasite there.

And these tech bros don't REALLY want to build out the infrastructure. All they want is tax cuts, cheap land and free power/water.

The locals ofcourse are supposed to prostrate themselves before the Emperor for this wonderful opportunity.

[toyg]

Heavy-industry sites are also extremely discouraged across Europe, outside of very specific zones. If anything, the current shift is about bringing datacentres in that same category.

[gruez]

>If anything, the current shift is about bringing datacentres in that same category.

But datacenters are hardly that? Sure, maybe whatever musk's doing with on-site gas turbines might qualify, but it's hardly representative of datacenters, which are probably closer to light industry or warehouses in terms of local impact.

[thefz]

Noise pollution is huge and water usage is through the roof which could worsen the effect of droughts where water is already scarce. Not to mention the imbalances they generate on power distribution.

[jmyeet]

It's worse than that. Data centers aren't a net zero for the area, they're a net negative. They use up water, arguably pollute said water [1], they jack up the price of everybody's electricity (because everyone else ends up paying for the extra infrastructure), can cause pollution directly (eg xAI runs highly-polluting gas turbines in a city through a legal loophole of them being "mobile" [2]) and aren't the quiet, unseen facilities proponents make them out to be (eg [3]).

On top of all that they typically get massive subsidies and tax credits. Why? Because the DC might go somewhere else, allegedly. Where? Nobody wants it. Everybody knows the politicians approving all this are getting bought or just coerced.

I'd love to see a single example of where one of these data centers was welcomed by the community or somehow a net positive.

[1]: https://www.bbc.com/news/articles/cy8gy7lv448o

[2]: https://earthjustice.org/case/xai-illegal-gas-power-plant-da...

[3]: https://www.nbcwashington.com/news/local/northern-virginia/n...

[morpheuskafka]

A warehouse or factory would have more jobs, but would also bring massive truck loads to the local roads and corresponding pollution. The low staffing of datacenters means that one they are built there is little transportation impact.

[znpy]

> A warehouse or factory would have more jobs,

It would add low value-added jobs though (unskilled labor). Datacenters add high value-added jobs (skilled labor).

Only looking at the headcount is shortsighted imho.

[oblio]

Aren't DC remotely managed for almost everything? How does a DC in Paris, Alabama benefit the locals if 99% of the well paid jobs for it are done in the Bay Area?

[eqvinox]

> Datacenters add high value-added jobs (skilled labor).

The highest skill job an AI datacenter adds locally is electrician or climate technician. And not many of those.

[thefz]

And there is almost zero value generated locally.

[tartoran]

The tax will address that.

[sieve]

> Datacenters are weird

In the same way that most public utilities are: train stations, railway lines, airports, garbage dumps, sewage treatment plants, military bases and a hundred other things. The negatives are concentrated to the locality and everyone else reaps the benefits.

I get it if you wish to put a 99% self-sufficiency condition (water/power etc) but everything else reeks of luddism and nimbyism.

[suddenlybananas]

>train stations, railway lines, airports, garbage dumps, sewage treatment plants, military bases

All of these create a ton of jobs in the local area and many of them provide massive advantages to the local area (except railway lines if you're not near a station I guess).

[amluto]

I think you're missing my point.

A sewage treatment plant may or may not be stinky, but I very much want to live in range of one so that my sewage gets treated. Similarly, living near a railway line means I can ride the train. Living near a landfill means that my trash can be managed at reasonable expense. Living near a military base adds huge potential economic value and nominally keeps me safe. Living near a giant AI inference or training facility? Meh -- I could use its services just as easily if it were 1000 miles away.

[notarobot123]

The point is that most utilities like transport infrastructure and sewage treatment benefit local residents and the broader society directly. Datacenters are not utilities in the same way and have a much narrower social impact.

[gruez]

>Datacenters are not utilities in the same way and have a much narrower social impact.

chatgpt has 900M weekly active users, so a significant fraction of the world population. Given that the user base probably skews towards rich countries, its proportion among the local population is probably even higher. On the other hand what do you think is the proportion of people who take trains, especially in the US? What about steel plants?

[manuelmoreale]

A significant % of the population in rich countries, probably close to > 100%, benefits from transport infrastructure and sewage treatment and many other vital services. I don't see why 900M of free users on chatgpt is a relevant argument.

[gruez]

>benefits from transport infrastructure

So if I drive to work then it's fair game to be NIMBY when they want to build a light rail station? You might say "but even if you don't use it, it's reducing traffic so you're still benefiting!", but if those kind of arguments are allowed, you can make similar arguments about how AI will make the whole country richer because of increased productivity or whatever.

[notarobot123]

The _social_ utility of AI has yet to prove itself a net benefit. If it increases productivity great, but what about the displaced jobs, etc. Social impact isn't measured in active weekly users.

[vessenes]

Why don't they pay much in taxes? hyperscalers are pretty profitable.

[zdragnar]

The hyperscalers aren't making sales directly from the data center, so there's not much to tax other than the land value, the electricity they use, and the few employee salaries.

You'd need some sort of data ingress/egress tax.

[vessenes]

Interesting. Just checked some numbers. So Coreweave has like $3bn in operating cashflow last quarter. Your point is that after we are done with all the capex/interest rate deductions/etc and look at a steady state business for a given CoreWeave datacenter, that net income won't be marked locally to a given datacenter.

Is that true though? If a Delaware corp is operating in most states, I believe they must file as a foreign corporation in that state; I'm surprised GAAP would not require them accruing some of this income to the locations that provided the work to do the income.

[oersted]

The idea of having what equates to trade tariffs for data transfers sounds horrifying. But to be fair, we are kinda suffering similar charges already from all major cloud providers, and we seem to be okay with it... Still horrifying, but not entirely unprecedented I suppose.

[GuestFAUniverse]

Where? What do you tax? Per request?

They will shuffle most gains around to the place with the lowest taxes. E.g. by internally buying and selling (overpriced) services.

The only realistic tax is coming from the jobs that serve those data centres (builders, maintenance, that little IT staff left for on-site jobs). And those are rather low margin jobs.

[jfyi]

>What do you tax?

Physical infrastructure inputs and negative externalities.

For example: electrical grid strain, water table consumption for cooling, and local pollution/carbon footprints.

[sbayg]

Don’t forget security guards. These data centers will be a huge target for both thieves and saboteurs. Unless the ai promoters are right that AI will be a great thing for everyone and not at all a source of societal and political strain.

[criddell]

Tax per megawatt, decibel, and gallon of water. That would provide incentives to reduce power consumption, noise, and water use.

[notrealyme123]

Because company profits are booked at their European HQ (so Ireland or Luxembourg).

So there is little profit being taxed at the Datacenter/country Level.

[palmotea]

> Why don't they pay much in taxes? hyperscalers are pretty profitable.

I wouldn't be surprised if those profits are re-imagined as costs paid to some entity in a tax-haven.

Also there's different kinds of taxes. IIRC, local communities get their revenue from sales taxes and property taxes. A data center doesn't sell anything, so they probably get zero from that. I don't really know how they'd factor into property taxes, because they're a blight and I don't know how the locality would assess their value without comparable transactions.

[hylaride]

Many jurisdictions have lower property taxes for certain kinds of industrial/commercial zoning in the name of bringing in jobs, etc. You sometimes see clusters of businesses at border areas of jurisdictions for this reason. Sometimes it's other advantages. Bottled water companies often take advantage of cheaper municipal water for "industrial" use, for example.

[znpy]

> they destroy green space

the law addresses exactly this. it greatly overtaxes datacenter in green spaces and lowers taxes in former industrial areas.

Regarding whether it's a good development drive... I can tell you, most companies could save a shitload of money by buying a few pallets of machines and racking them in a... datacenter.

I see our monthly AWS bill, I highly doubt we'd be spending that in datacenter bills.