Red state residents lead growing rebellion against data centers that Trump loves

ThatRobGuy said:

That's why I mentioned that as a comparable example, the NIMBYism we normally see is with regards to building Section 8 or affordable housing adjacent to people's neighborhoods.

This is just a different flavor of the same mentality.

Click to expand...

Sure it is, sure...

ThatRobGuy said:

"I don't want my home's value to drop by 30%, so they need to build that apartment complex somewhere else"
and

Click to expand...

Why would an apartment complex lower the value of my house?
If it did, why would I care? It would (potentially) drop the assessed value on the property tax. That sounds like a good thing.

ThatRobGuy said:

"I don't want them building that datacenter near my house, it'll ruin the scenery and make my utility bills go up by 30%"

Click to expand...

Datacenters have warehouse-like profies (and need at least modest access to good roads as there is a lot of heavy equipment to install). They go in flatish places, not on the riverbank or the side of a mountain. Neither are they 12 stories tall and block the view.

As for the "cost" to neighbors, the apartment might impact values (if it would at all) within a few blocks. The electric costs cover cities, counties, and regions.

The benefits of these "AI" data centers are what? Slower, less useful google searches? An assistant the summarizes my emails? People addicted to chatting with non-real "people"?

The extra people might attract new business I might like to patron, or employees for a business I might own, or a new bus stop so I don't have to drive to work (or home from the bars).

It looks to me like one of your items has minor impact very locally and some benefits, and the other one has wider ranging impacts and no discernible value.

Hans Blaster said:

Sure it is, sure...

Why would an apartment complex lower the value of my house?
If it did, why would I care? It would (potentially) drop the assessed value on the property tax. That sounds like a good thing.

Datacenters have warehouse-like profies (and need at least modest access to good roads as there is a lot of heavy equipment to install). They go in flatish places, not on the riverbank or the side of a mountain. Neither are they 12 stories tall and block the view.

Click to expand...

I would guess that second-gen DCs would be tied to the earth more, literally, using geothermal cooling since otherwise it uses so much water!

Hans Blaster said:

As for the "cost" to neighbors, the apartment might impact values (if it would at all) within a few blocks. The electric costs cover cities, counties, and regions.

The benefits of these "AI" data centers are what? Slower, less useful google searches? An assistant the summarizes my emails? People addicted to chatting with non-real "people"?

Click to expand...

We produce more data in a few months than it used to take decades to create. That has to continue to grow or the whole system collapses under its own weight, I guess.

Hans Blaster said:

The extra people might attract new business I might like to patron, or employees for a business I might own, or a new bus stop so I don't have to drive to work (or home from the bars).

It looks to me like one of your items has minor impact very locally and some benefits, and the other one has wider ranging impacts and no discernible value.

Click to expand...

Enron and Theranos’ love child.

One man's minor impact is another's significant event. Having seen manufacturing dry up in towns, it falls into the category of "at least it's something." It's an old story, and there's many a forgotten town that died when it's only industry closed.

SimplyMe said:

If that is what is read, you appear to have a comprehension problem.

Click to expand...

Thank you for your insult.

Uh-oh - looks like a CEO is feeling put upon.

Sam Altman would like to remind us all that humans use power and water too! Even worse, it takes longer for us to learn things than AI!!

Hans Blaster said:

Why would an apartment complex lower the value of my house?
If it did, why would I care? It would (potentially) drop the assessed value on the property tax. That sounds like a good thing.

Click to expand...

Depends on the kind of units... and the current conditions of the "host city"

The NIMBYism that occurs in that realm typically centers around the idea that if heavily subsidized housing pops up a few blocks away, it will introduce some negative elements (that typically go hand-in-hand with subsidized housing)

Hans Blaster said:

As for the "cost" to neighbors, the apartment might impact values (if it would at all) within a few blocks. The electric costs cover cities, counties, and regions.

The benefits of these "AI" data centers are what? Slower, less useful google searches? An assistant the summarizes my emails? People addicted to chatting with non-real "people"?

The extra people might attract new business I might like to patron, or employees for a business I might own, or a new bus stop so I don't have to drive to work (or home from the bars).

It looks to me like one of your items has minor impact very locally and some benefits, and the other one has wider ranging impacts and no discernible value.

Click to expand...

It sounds like you're under the assumption that the bulk of AI is just the end-user/user-facing AI that people recreationally use online. (like people hopping on to ChatGPT and asking questions, asking it to make a funny picture, or self-driving Waymo cars)

I think most people are probably ignorant of how much they use (and enjoy) benefits of AI without even realizing they're using it.

It's heavily baked into a lot of services and business verticals (that most people enjoy, but don't even realize is using AI)

FireDragon76 said:

The difference is that Section 8 actually helps vulnerable people. It's unclear how much public benefit comes from large data centers. A quarter of OpenAI's compute is used to produce videos that mostly amount to AI slop, and OpenAI itself admits it's business model doesn't produce real revenue.

I make some songs using Claude and Suno but most of the songs on Suno are little more than slop with little thought or care rooted in real artistic sensibilities. Alot of AI generated content is like that, especially when people use AI to replace actual effort. I don't produce a song just because I'm bored, but because I'm genuinely inspired to make something, I have a vision of what that should look like in the end. I use a simple Claude or DeepSeek model to generate a rough draft based on a prompt or discussion about my artistic goals or some theme or idea, then I go over the actual text and use a model again to propose changes to a few lines or verses that don't feel quite right (sort of like a thesaurus), so that there's a sense of coherence and depth to the lyrics. Many people on Suno don't even bother doing that, you can get songs that appear fluent but have little artistic merit or grounding in anything substantive, and people churn these out one right after the other.

For much of general LLM use, it's similar. Most hardcore users right now aren't necessarily businesses, but private individuals looking to produce fanfic, pulp fiction, and smut, which is probably one reason OpenAI's userbase took a hit recently (it instituted more content controls in the wake of lawsuits, and will no longer generate highly explicit or violent content).

Click to expand...

People interacting with random chatbots to ask questions or use it for recreational purposes is by no means the crux of AI usage

JustaPewFiller said:

Uh-oh - looks like a CEO is feeling put upon.

Sam Altman would like to remind us all that humans use power and water too! Even worse, it takes longer for us to learn things than AI!!

Sam Altman would like remind you that humans use a lot of energy, too | TechCrunch

"It also takes a lot of energy to train a human."

techcrunch.com

Click to expand...

While the CEO picked a particularly poor, tone-deaf point of comparison, I think the overall sentiment of comparing energy use to other things is a valid point of context for people to keep things in perspective.

The one I would have invoked:


Streaming Netflix: ~0.077 kWh per hour = 77 Wh/hour
LLM chat prompt: ~0.24 Wh per prompt (using a middle of the road model)

Meaning: an hour of Netflix streaming has roughly the same footprint as 300 typical prompts to an AI chatbot.

...and the author of the IEA pieces was trying to suggest that the power usage from streaming services "wasn't so bad".

Although, given the timing of that IEA piece (late 2020), it's possible that it was trying to make people feel better about the choice to stay cooped up inside watching streaming services all day during the lockdowns given the "streaming's not so bad" vibe of the article.

ThatRobGuy said:

Depends on the kind of units... and the current conditions of the "host city"

The NIMBYism that occurs in that realm typically centers around the idea that if heavily subsidized housing pops up a few blocks away, it will introduce some negative elements (that typically go hand-in-hand with subsidized housing)

Click to expand...

So just a lot of ignorant assumptions and prejudice against poor people.

ThatRobGuy said:

It sounds like you're under the assumption that the bulk of AI is just the end-user/user-facing AI that people recreationally use online. (like people hopping on to ChatGPT and asking questions, asking it to make a funny picture, or self-driving Waymo cars)

I think most people are probably ignorant of how much they use (and enjoy) benefits of AI without even realizing they're using it.

Click to expand...

I've seen the "features" the add to apps I use (like Zoom) add AI "features". Never touched one. I always skip past the AI summary on my google searches. (Slow and don't trust. I want a link, silly search engine.)

ThatRobGuy said:

It's heavily baked into a lot of services and business verticals (that most people enjoy, but don't even realize is using AI)

Click to expand...

"Business verticals". LOL. Sounds like powerpoint blather.

ThatRobGuy said:

Americans Use AI in Everyday Products Without Realizing It

Nearly all Americans use AI products, but many don't realize it. Most say the government and businesses are equally liable for reducing its potential harms.

news.gallup.com

Click to expand...

Read that article. Don't use em.

ThatRobGuy said:

While the CEO picked a particularly poor, tone-deaf point of comparison, I think the overall sentiment of comparing energy use to other things is a valid point of context for people to keep things in perspective.

The one I would have invoked:

https://www.iea.org/commentaries/the-carbon-footprint-of-streaming-video-fact-checking-the-headlines

Streaming Netflix: ~0.077 kWh per hour = 77 Wh/hour
LLM chat prompt: ~0.24 Wh per prompt (using a middle of the road model)

Meaning: an hour of Netflix streaming has roughly the same footprint as 300 typical prompts to an AI chatbot.

...and the author of the IEA pieces was trying to suggest that the power usage from streaming services "wasn't so bad".

Although, given the timing of that IEA piece (late 2020), it's possible that it was trying to make people feel better about the choice to stay cooped up inside watching streaming services all day during the lockdowns given the "streaming's not so bad" vibe of the article.

Click to expand...

It's a fair point.

Maybe regarding the AI Data Centers and jobs its down to this.

Will an AI Data Center help with jobs in a rural area? Yes, some. But probably not as much as something more traditional like some type of manufacturing. But, it is what is and that is where we are at.

Most anything that comes to a rural area that brings is jobs is also going to have some type of offsetting factor as well be it pollution, energy costs, etc.

Maybe it boils down to people aren't convinced these are going to produce enough economic benefit for the area to offset the downsides?

JustaPewFiller said:

It's a fair point.

Maybe regarding the AI Data Centers and jobs its down to this.

Will an AI Data Center help with jobs in a rural area? Yes, some. But probably not as much as something more traditional like some type of manufacturing. But, it is what is and that is where we are at.

Most anything that comes to a rural area that brings is jobs is also going to have some type of offsetting factor as well be it pollution, energy costs, etc.

Maybe it boils down to people aren't convinced these are going to produce enough economic benefit for the area to offset the downsides?

Click to expand...

I would say it's largely going to depend on the rural area.

If the particular AI facilities, in general, are going to be used to facilitate the replacement of the jobs the people in that area already have, then their concerns would be understandable. (like if the new AI facility is specifically for facilitating self-driving trucks in a city where 40% of the men in town are truck drivers)

However, if it's an area where the jobs have already dried up, and AI has no impact on the trajectory of that, then not so much. (like if we're talking about coal mining jobs or something like that, those aren't coming back - with or without a new AI data center in town)

Hans Blaster said:

I've seen the "features" the add to apps I use (like Zoom) add AI "features". Never touched one. I always skip past the AI summary on my google searches. (Slow and don't trust. I want a link, silly search engine.)

Click to expand...

Hans Blaster said:

"Business verticals". LOL. Sounds like powerpoint blather.

Click to expand...

Hans Blaster said:

Read that article. Don't use em.

Click to expand...

Again, it's not just the user-facing web-enabled applications.

Have you received mail?
Have you shopped in a retail store?
Have you ever used GPS?
Do you have any credit cards that have fraud detection enabled?
(the list goes on)

If the answer to any of those is "yes"...hate to be the bearer of bad news, but you've indirectly used AI. USPS uses AI as part of their routing logistics, retail stores use it for supply chain logistics, GPS uses it, Chase/Citi/Amex use it as part of their fraud detection algorithms.

ThatRobGuy said:

Again, it's not just the user-facing web-enabled applications.

Click to expand...

You said that already. I read your stupid article.

ThatRobGuy said:

Have you received mail?

Click to expand...

Not efficiently. Mail service had gotten real lousy in the last decade. Is AI to blame for that?

ThatRobGuy said:

Have you shopped in a retail store?

Click to expand...

A few times per year.

ThatRobGuy said:

Have you ever used GPS?

Click to expand...

Not recently.

ThatRobGuy said:

Do you have any credit cards that have fraud detection enabled?
(the list goes on)

Click to expand...

Not that I know of.

ThatRobGuy said:

If the answer to any of those is "yes"...hate to be the bearer of bad news, but you've indirectly used AI. USPS uses AI as part of their routing logistics, retail stores use it for supply chain logistics, GPS uses it, Chase/Citi/Amex use it as part of their fraud detection algorithms.

Click to expand...

Haven't they been using computer optimizations and just-in-time logistics for a while now? The CapEx and power usage for these data centers can't be driven by companies doing logistics. There isn't enough money in it for them.

Hans Blaster said:

Not efficiently. Mail service had gotten real lousy in the last decade. Is AI to blame for that?

Click to expand...

Sorry your experience has been so poor, but reports would suggest that your experience must be an outlier


For those getting package deliveries from USPS, sorting facilities that leverage it can process things 10x faster with higher accuracy.

Hans Blaster said:

Haven't they been using computer optimizations and just-in-time logistics for a while now? The CapEx and power usage for these data centers can't be driven by companies doing logistics. There isn't enough money in it for them.

Click to expand...

What would you estimate the usage breakdown (in terms of power) is between user-facing "recreational" AI & prompting vs. companies embedding in in their various subsystems for BI purposes in an intensive manner?

I showed in a previous post that the "casual AI use" is a minimal impact. If you watch Netflix for an hour, you've used more power than if I type 300 Q&A prompts into an AI bot interface.

It's when companies (like mine) are having it do the work of a half dozen database administrators that it's racking up the heavy hits.

ThatRobGuy said:

Sorry your experience has been so poor, but reports would suggest that your experience must be an outlier

Click to expand...

Not me. USPS relaxed their 2-3 day delivery target for a letter from any point in the CONUS to any other a few years back.

ThatRobGuy said:

NVIDIA Provides U.S. Postal Service AI Technology to Improve Delivery Service

NVIDIA today announced that the United States Postal Service – the world’s largest postal service, with 485 million mail pieces processed and delivered daily – is adopting end-to-end AI technology from NVIDIA to improve its package data processing efficiency.

nvidianews.nvidia.com

For those getting package deliveries from USPS, sorting facilities that leverage it can process things 10x faster with higher accuracy.

Click to expand...

That's an NVidia PR piece, and I am familiar with their exaggerations. But more importantly, they speak of deploying it to 200 postal facitilities (regional sorting operations. Each one handles 1 or more 3-digit zip code regions for all zip code with the same first 3 digits). These are not in data centers, but in warehouse-like facilities where mail is sorted for whole regions.

ThatRobGuy said:

What would you estimate the usage breakdown (in terms of power) is between user-facing "recreational" AI & prompting vs. companies embedding in in their various subsystems for BI purposes in an intensive manner?

Click to expand...

I don't know. This massive "AI data center" build out (and we should stop calling it that and use the actual technology names) is not being driven by a rack of servers in a logistics warehouse. (Not sure what BI is. I know what BO is.)

ThatRobGuy said:

I showed in a previous post that the "casual AI use" is a minimal impact. If you watch Netflix for an hour, you've used more power than if I type 300 Q&A prompts into an AI bot interface.

Click to expand...

Don't know why you're trying to guilt me with that. I don't use either.

ThatRobGuy said:

It's when companies (like mine) are having it do the work of a half dozen database administrators that it's racking up the heavy hits.

Click to expand...

Since I don't know what a database admin does...

Hans Blaster said:

I don't know. This massive "AI data center" build out (and we should stop calling it that and use the actual technology names) is not being driven by a rack of servers in a logistics warehouse. (Not sure what BI is. I know what BO is.)

Click to expand...

Hans Blaster said:

Since I don't know what a database admin does...

Click to expand...

"BI" is Business intelligence.

When companies send massive amounts of data into it to analyze sales/trends/etc... determine which sales pitches and promotions performed the best, analyzing sales & retention by zip code...

And having AI do that and spit out detailed results and accompanying charts, tables, graphs in an hour vs. waiting for a week team of BI/SQL developers to get their arms around that massive amount of data and write reports (that a sales person will then have to manually translate that into charts/graphs in powerpoint)
Or...
Things like customer service contact centers doing quality audits an uploading gigabytes and gigabytes of call recordings or chat transcripts in to get sentiment scores and analyze which approaches are working the best vs. having a team of 5 or 6 picking a handful of call recordings at random to manually listen to them to audit

...those are the things that are the resource intensive AI operations, not someone bored on their couch asking ChatGPT "Which cast member on Happy Days was the oldest?"

Hans Blaster said:

That's an NVidia PR piece, and I am familiar with their exaggerations. But more importantly, they speak of deploying it to 200 postal facitilities (regional sorting operations. Each one handles 1 or more 3-digit zip code regions for all zip code with the same first 3 digits). These are not in data centers, but in warehouse-like facilities where mail is sorted for whole regions.

Click to expand...

The sorting software (with an AI backend) would be in data centers. (much like the databases that supported their earlier software wasn't actually housed in the sorting facilities themselves)

Very few big outfits are leveraging "on-prem" servers these days. Most have their stuff hosted in Cloud services across multiple availability zones for redundancy.

JustaPewFiller said:

Uh-oh - looks like a CEO is feeling put upon.

Sam Altman would like to remind us all that humans use power and water too! Even worse, it takes longer for us to learn things than AI!!

Click to expand...

Whatever will we do with these inefficient and superfluous 'humans'.

ThatRobGuy said:

"BI" is Business intelligence.

Click to expand...

That's one of those "oxymoron" things, right?

ThatRobGuy said:

When companies send massive amounts of data into it to analyze sales/trends/etc... determine which sales pitches and promotions performed the best, analyzing sales & retention by zip code...

And having AI do that and spit out detailed results and accompanying charts, tables, graphs in an hour vs. waiting for a week team of BI/SQL developers to get their arms around that massive amount of data and write reports (that a sales person will then have to manually translate that into charts/graphs in powerpoint)

Click to expand...

Do these "AI"s do statistical analyses that do not otherwise exist? Are they not just running it on the same data sets through the same algorithms? (Total sales, total sales by zip code, Year-over year changes, etc. are all standard things. No need for an AI to "estimate" them.)

Then it looks like the "AI"s do the "understanding", chart making, and write reports.

Frankly that sounds like a *bad* idea. Perhaps the C-suite guys were just nodding along to the presentation and not understanding it, but the sales people (and VP of Sales) making the presentation would. This seems like a *bad* thing that no one understands the data used to make decisions. (Note to self: Switch from stocks to a sandwich heavy portfolio.)

ThatRobGuy said:

Or...
Things like customer service contact centers doing quality audits an uploading gigabytes and gigabytes of call recordings or chat transcripts in to get sentiment scores and analyze which approaches are working the best vs. having a team of 5 or 6 picking a handful of call recordings at random to manually listen to them to audit

Click to expand...

We're going to let a machine judge "sentiment"? That will go well.

ThatRobGuy said:

...those are the things that are the resource intensive AI operations, not someone bored on their couch asking ChatGPT "Which cast member on Happy Days was the oldest?"

Click to expand...

Trick question. Some of them are dead. (If you're talking about the "youths", then that is almost certainly Henry "The Fonz" Winkler. I think that guy was a 30-year-old "teen".)

ThatRobGuy said:

The sorting software (with an AI backend) would be in data centers. (much like the databases that supported their earlier software wasn't actually housed in the sorting facilities themselves)

Click to expand...

That's not how the NVidia PR sheet reads.

ThatRobGuy said:

Very few big outfits are leveraging "on-prem" servers these days. Most have their stuff hosted in Cloud services across multiple availability zones for redundancy.

Click to expand...

While there has been some noticeable expansion data centers for "cloud services", the data centers in the thread are specifically the high-energy density GPU-powered "AI" data centers.

Hans Blaster said:

Frankly that sounds like a *bad* idea. Perhaps the C-suite guys were just nodding along to the presentation and not understanding it, but the sales people (and VP of Sales) making the presentation would. This seems like a *bad* thing that no one understands the data used to make decisions. (Note to self: Switch from stocks to a sandwich heavy portfolio.)

Click to expand...

I think that's where there's some misunderstanding with regards to how it's leveraged.

The AI workflow that are processing it have already had detailed instructions injected in to the organization specific customized model.

Meaning, it's not just a guy uploading massive CSV files and asking a chat bot a generic question like "tell me stuff about this data"

There's a human upstream injecting the model with detailed business rules, KPIs, etc... so it knows how to make the decisions.

The AI model can just get its arms around the massive data interpret where to join things together, can identify bad records and all that stuff quicker than a SQL report writer ever could.

People are still thinking of AI in terms of people asking chatbots things, and robots moving stuff around in a warehouse. That's really only a small piece of the pie in terms of AI usage and computing power usage. The bigger footprint is stuff happening behind the scenes that nobody ever sees.

Hans Blaster said:

We're going to let a machine judge "sentiment"? That will go well.

Click to expand...

Amazon's Contact Lens and Lex are quite proficient at it.

We have a client (that I won't name) that has a pretty large contact center ecosystem, and is handling about 30,000 calls per month. (ranging from quick 5 minute phone calls, up to calls spanning a half hour or more)

Their internal Quality/Verifications team only consists of about 20 people.

Even if that team was doubled in size, there's no way they'd be able to listen to and analyze a large percentage of the calls. So before, they were relegated to randomly auditing wav recordings of phone calls, combined with pulling up ones where there was a customer who called in to complain about the rep they spoke with.

And it's not a blind trust situations. It's not as if orgs are spinning these things up without scale up testing.

The proof of concept phase for those situations is almost always send a few small batches of calls up to it, and then compare what the tools rated it as compared to what the human rated it as. In over 95% of the calls, the review outcomes were in the same ballpark, and for the 5% that weren't, some of those differences were actually the human making a mistake or not catching something that happened on the call.

FWIW:

What is a Data Center - Types of Data Centers

And:

What Is an AI Data Center? | IBM