I drive a lot. I have yet to see any signs saying please bring the data centers. Always 100% against.
Are you for or against Data Centers?
- Thread starter ZZman
- Start date
There is zero upside to a data center near you. A whole bunch of temporary travelling workers come in to build the thing. They might hire some local contractors but not many. They fill the local hotels for a while. When its done they leave, employ a handful at most. Your water bill goes up - they literally sent me a letter saying connection fee was going up 30% and water itself 20% to pay for "capital improvements".
Compared to a factory or similar coming in and employing a bunch of locals permanently.
Yes, a heat exchanger that pulls air through a cooled coil is pretty much how they all work. I am sure there is some way but I can't think of a commercial method of cooling that doesn't use liquid to gas phase change to provide cooling. DX does it. Heat pumps do it. Centrifugal Chillers do it. From Munters Oasis Standard:
"An air-to-air heat exchanger transfer heat from the data center air to a separate scavenger airstream, which never enters the white space."
Likely so they can charge more for your heat exchange coil because it has a special name.
I think that this is a bit of a fable, although this trope is often used in defending the disruptive aspects of data center development.
I am personally in accord with the poster above who wrote that advances in hardware and software will soon render these massive constructs obsolete and that many will be dark shortly after or even upon completion.
Remember Moore's Law? Think of how it applies in this analogous case.
Think of the capital destroyed when the data center boom implodes that might have been put to productive uses.
This is also a risk to the economy as a whole, with a deep recession likely to follow.
I have been following the economics behind the AI boom and where the money comes from and goes to. Based on what I see so far:
1) Money for the data centers build are going partly into construction, power equipment, cooling equipment, and finally the chips.
2) The money for the chips are mostly going into the chip companies like Nvidia, Micron, Samsung, SK Hynix, SanDisk, ASML, TSMC, etc. Their stock prices go up
3) These data centers are paid for by the cloud companies like Microsoft, Google, Meta, Oracles, Amazon, etc. A lot of those were funded by their cash flow as they were swimming in cash in recent years. Instead of buying back shares they are start to raise capital as well to pay for these data centers.
4) Some of those were rented to companies like OpenAI and Anthropic. Some were used by themselves to train their own models.
5) Some of the customers of these AI model companies are paying for these models usage (tokens) by increased productivities. If you use Palentir as an example you can see how that significantly improve the US military capabilities, and a lot of businesses were able to improve their performance by using these models. At my work we have people using these to do simple tasks that would have taken a couple months and finish them within 1 afternoon. Unfortunately, some companies even layoff employees since they got more productive, yet some were forced to compete by using more AI and layoff people to pay for it (some AI work is currenty more expensive than human).
6) There are a lot of demand for US based investment due to geopolitical reason, and US being the safe haven currency, and some want to stay invested in the US but don't want to buy Treasury or mortgage because they got burnt before, so they buy AI stocks. Very recently there was a cross over at one point that Microsoft can borrow money cheaper than US government based on bond rate, although for a very short time. Let that sink in a bit.
Will this be a bubble? I'm sure it will be, but I am also sure we are at a time when AI model is really improving a lot of companies' productivity. This may commoditize a lot of service out there, or help inflate away the US debt crisis, or increase unemployment, or crash the world economy, or a combination of them all.
I think the infrastructure for these data centers would still be used by non AI even if AI bubble burst. It may not be immediately, the companies that build them may lose money, the rent they could charge may crash, but there would be demand for them at the right price. For now, we have too low of an interest rate to prevent people from taking a chance on them.
Yes, if you go by all the signs and protestors, and listen to those that are interviewed on the news, you would think there is no one in Northern Utah that is in favor of the data center. But I've been asking my friends, those at the gym, people at church, and I have yet to find anyone that is opposed to it.
It should be no surprise that the loudest side on an issue always gets all the attention. It doesn't sell any news time when you interview someone and they say, "I'm ok with the data center. It's got to go somewhere, and why not out in the desert?" It's a lot more exciting to interview those that are yelling, and threatening to boycott local businesses, and vote out the county commissioners. Those in favor aren't going out and putting up signs.
It is totally unrealistic to judge the community attitude towards this issue by who is spending the most money to put out signs.
I'd note that lots of times that money comes from outside agitators.
A lot of the resistance in development opposition in real estate are there to prop up the bargaining power and negotiation tactics. I've dealt with public hearing before and have seen some "community leaders" coming out of nowhere a few towns away show up to support one side or another for a clause they have nothing to do with.
I am genuinely concerned about electricity and natural gas price due to the AI boom, and maybe a bit about water usage only if they are near me (Seattle or Montana water is not something that impacts me regardless).
I am genuinely concerned about electricity and natural gas price due to the AI boom, and maybe a bit about water usage only if they are near me (Seattle or Montana water is not something that impacts me regardless).
They are large and very noisy in off putting strange droning frequencies . Some epileptics claim the sound gives them seizures , some folks who are nearby claim lost sleep and nausea because the noise keeps them from sleeping.
My issue with data centers is multi fold but ignoring all that at a minimum the tech is changing so quickly that we likely can be “behind the curve” and cheaply, easily pull ahead at the last minute with smaller, cooler, cheaper, more efficient, faster tech.
There is no benefit to the spend on a nation full of bleeding edge ewaste that will be junk in 3-5 years as its outmoded.
This is ignoring the heat, wasted drinking water and the potential societal damage.
Worse, pioneers and inventors almost never make back investment, it’s always the commoditization phase years later that makes money.
The numbers being thrown around are under 3M gallons per day.
Just watch, after the bubble eventually bursts and we move on to much more efficient tech, the public will be stuck with these data centers and will have to pay for the cleanup, if we even attempt any cleanup. They will probably be left to rot away.
Ahh, but what am I saying? It’s for our own good. Can’t make an omlet without breaking a few eggs, right?
Ahh, but what am I saying? It’s for our own good. Can’t make an omlet without breaking a few eggs, right?
So the same as approximately 50,000 people, give or take.
If you were a farmer in 1910, would you have been better off buying a new Model T for $750.00 or taking a Mercedes for free?
My point is that these huge AI intended data centers may simply cost too much to run to allow for cheaper uses, not to mention the loss of capital involved in marking them to market.
Given the amounts of capital involved, these may be losses with economy wide impacts.
With so many players betting on the same set of potential users, oversupply seems inevitable and that will impact the value of all of these huge and costly constructions.
When "more efficient tech" shows up they likely would just swap the computers inside with the more efficient stuff if possible, but if not, the building gets torn down and rebuild. Data centers get rebuild once every several years as need change. They don't get remodeled they get torn down and rebuild.
Machine has different power, cooling, and weight requirement over time and they are not unchanged like human within a generation (you don't breathe and run twice as fast after you gain 10 years of work experience). The older data centers sometime sat as idle capacity and fire up during peak seasons (Xmas shopping for example), and when that's not even worth the space it sat on they get torn down and rebuild. Same for some FAB as process changes over time and different kinds of machines and support equipment can't be upgraded to keep up. In the end it is really just like a big box store or shopping mall with a lot of power and cooling. It doesn't take a lot of work rebuilding it or prevent it from turning back into a farmland. The problem is usually the landowners not wanting to turn an industrial zone back into agricultural because of profit reason. If you want to turn it back into farmland you can always start a go fund me page and buy it off, tear it down, and just let it sit to do nothing. It is probably easier to build single family homes on it afterwards though, if you can find enough people wanting to buy them for a profit. Data centers are not hazmat that needs expensive cleanup. They are just like offices with bigger weight support structure and a lot more air conditioning. E-waste are moved away and recycled just like a typical office.
I don't think that's apple to apple comparison. If AI boom stop, they would stop ordering AI specific chips and keep using what they already bought and let them slowly depreciate, wear out, and removed when end of life. Farmers back then buy and own their individual equipments and if their need change they are stuck holding on to obsoleted equipment. This is more like what would happen to Hertz or Enterprise during a boom and bust cycle in the auto industry. They are just financing / leasing cars and then renting them out to others. Let's say all of a sudden nobody order AI chips, TSMC would redirect capacity to make other chips (say AMD CPU for laptop or Apple's chips for iPhone), this is why TSMC is always doing ok and Intel is doing bad (they got stuck with a factory making chips that they couldn't sell, and their accountants wouldn't let them upgrade or build new fab until they can sell their chips, a catch 22).
Most non individually owned AI equipments are leased, either short term or long term. If one day suddenly people change from ChatGPT to Claude for example as we have seen recently, whatever intended for OpenAI could suddenly be lease cancelled and they can be leased to others like Meta (as we have seen recently some of Oracle's capacity for OpenAI got redirected to Meta).
Are there going to be boom and bust cycles? Semiconductor industry always have them, but we already know that and expect that. Just watch out for the cycles when you buy their stocks and when things get too bad the companies merge and layoffs happen, then reduce capacity and let price stabilize, and back to booming again in a few years.
SNDK was $60 about 3 years about right at the end of a downturn after splitting off from WDC, it is now $1340. It was $80 in 2006 and then in 2008 at the bottom down to $10. Nothing new here, just don't hold it for 30 years and expect stability. It is not Coca Cola or Costco.
Well, here's a new one involving Pulte Homes - yep, that Pulte:
A distributed AI Node on your home - awesome!
Here's the reporting on it with more information:
https://www.realtor.com/news/trends/nvidia-pultegroup-span-date-center-backyard/
A distributed AI Node on your home - awesome!
Here's the reporting on it with more information:
https://www.realtor.com/news/trends/nvidia-pultegroup-span-date-center-backyard/
I can already see gangs of thieves stealing Nvidia GPUs out of these server boxes. This great idea will likely die as quickly as it was born, maybe even quicker.
