That's incorrect. LLM training models are expected to run through GPUs at a pace that even hyperscalers don't want to admit they are depreciating them incorrectly. You can look this up, Microsoft is already getting questioned on their accounting for this hardware. State of the art GPUs will have one purpose - AI. My pushback on 99's post is that the hardware for that scale of AI use, even without buildout pushback, pushes the cost of AI to insane levels.

JB99 said:

---

Urban Country Boy said:

---

JB99 said:

---

500,000ags said:

---

The data centers being built have nothing to do with hardware needed for the internet or cloud.

NEW BIG BUILDING NOT FOR INTERNET. IT FOR AI.

---

I wouldn't say nothing. They have to connect those data centers to the internet with massive pipes. As time goes on more people will interface with their LLM of choice vs. Going to company websites. The websites will become more LLM friendly vs. User friendly exposing their data in a way LLMs use it natively. There's alot of overlap. Alot of SaaS companies are embedding AI into their applications as well. I think we'll see alot of co-location between AI and corporate data for efficiencies as time goes on.

---

News for you. Everything you do is connected by "pipes" to everything you interact with. Everything you do is already going through a data center. They know everything already. Why do you think you get ads on your phone after just talking about something?

---

Why is that news for me? Your argument is that AI has nothing to do with the internet or web traffic. It most assuredly does and will play a much bigger impact very very soon. It will be the dominant way people search the web, buy things, etc...

---

This. My buddy just spec'd out and sold connectivity for a DC going in just north of Houston. They requested 5tb/s pipes between the DC and both Houston and Dallas with ability to expand it later.

500,000ags said:

---

That's incorrect. LLM training models are expected to run through GPUs at a pace that even hyperscalers don't want to admit they are depreciating them incorrectly. You can look this up, Microsoft is already getting questioned on their accounting for this hardware. State of the art GPUs will have one purpose - AI. My pushback on 99's post is that the hardware for that scale of AI use, even without buildout pushback, pushes the cost of AI to insane levels.

---

I am telling you what i see daily when I am in data centers, no one has once debated the costs, the tenants make decisions based on dozens of variables, lead times tend to change plans rapidly. Will they all eventually be 100%, possibly, but no one reslly knows. The same can be said for one of your precious legacy data centers, no reason a tenant couldn't change what they have deployed, it happens all the time.

You quoted some developer, who has absolutely nothing to do with what is being deployed.

Myopic, pure and simple. You're debating if I understand that maybe AI is only 80% of compute at any given point, when the underlying reason it is getting built is 100% AI. The hardware is expensive and power hungry by many multiples, so yes I can assume the use case. That has nothing to do with anything, in the grand scheme, but it's the part you want to debate because it's the part you understand (and want to lecture over).

Based on all these comments about the internet and debit cards, I just can't believe we haven't gone dark without all this new data center buildout.

I don't care if you believe me, a developer, or whoever the hell you want. You are dug in and no amount of evidence will change your mind.

I am telling you what I see, real world experience. You want data centers to be a boogeyman because you have talked yourself into it, and for some reason you don't think there is still massive growth in cloud/storage.

Not sure why you are so dismissive, but you seem like a pretty miserable person to be around and I'm done with your nonsense.

Very little of what you just said is true. I am not entrenched at all. You want to make the argument that current data center projects are necessary for existing use cases. For some reason you don't want to acknowledge that current data centers are materially different in the hardware they include, the resources needed, the scale of funding, and the approach to approval and buildout. The free market is telling us all we need to known. Valuations are high, financiers are flinging money, inputs are skyrocketing, people are protesting, etc. But yeah, it's all in our heads.

They only ones who care this much about measuring something are the ones who are embarrassed about the size of their own, whatever that may be.

Its not a great look, but here we are.

Shhh, don't tell him.

500,000ags said:

---

Very little of what you just said is true. I am not entrenched at all. You want to make the argument that current data center projects are necessary for existing use cases. For some reason you don't want to acknowledge that current data centers are materially different in the hardware they include, the resources needed, the scale of funding, and the approach to approval and buildout. The free market is telling us all we need to known. Valuations are high, financiers are flinging money, inputs are skyrocketing, people are protesting, etc. But yeah, it's all in our heads.

---

None of what you have listed changes the fact that the building is a giant warehouse with power, cooling, and compute, and no one is protesting because of the scale of funding. Protests are because of power, water, they're an eyesore, and God knows what people are making up.

My only point that I have tried to make a dozen times is that these have been around for decades, at scale and in rural areas. Over 4000 data centers in the US currently, that didn't pop up over night. You can't do anything about the cost, like it or not it's a race and speed costs a lot of money.

I'd ask AI to summarize this thread but we probably need a few more data centers to wade through this crap

GAC06 said:

---

I'd ask AI to summarize this thread but we probably need a few more data centers to wade through this crap

---

I agree. 100%.

Saying 'data centers have existed for decades' is like saying cars have existed since 1885, technically true, but a Model T and a Tesla aren't the same. Scale and speed of change is the point. 4,000 data centers built over 30 years is a very different story than hundreds of gigawatt-scale facilities being rushed through in 3.

The ROI for AI is very good. The productivity gains are huge. The economics for making money with these hyperscalers like Google and Anthropic is very good. The problem is they are all in an arms race with each other and other countries. Any money they make has to be re-invested immediately to build the next model. Companies are putting out new models every 3-4 months. So they won't be profitable for a while because they have to keep building or get left behind.

500,000ags said:

---

Saying 'data centers have existed for decades' is like saying cars have existed since 1885, technically true, but a Model T and a Tesla aren't the same. Scale and speed of change is the point. 4,000 data centers built over 30 years is a very different story than hundreds of gigawatt-scale facilities being rushed through in 3.

---

Agree about the scale and speed, but functionally they are the same as they were decades ago, only difference is size and what is inside. Still have the same basic requirements, but there have been monsters being built in rural areas for years as well.

TAMUallen said:

---

Data centers are going to be such an issue for electricity grids

---

It's the next dot.com but in a different way

500,000ags said:

---

Saying 'data centers have existed for decades' is like saying cars have existed since 1885, technically true, but a Model T and a Tesla aren't the same. Scale and speed of change is the point. 4,000 data centers built over 30 years is a very different story than hundreds of gigawatt-scale facilities being rushed through in 3.

---

Agreed, the form factor is the same but the scale is different.The scale introduces new engineering problems.

That IS NOT true. What goes inside are GPUs that are, by several multiples, more power hungry to handle parallel processing. Everything from back up power to ventilation still needs sorting on a case by case basis. And frankly, I don't think most local leaders understand developer representations.

Bold statement. Product orgs definitely are trying to monetize their ability to push new work. GTM still identifying playbooks that increase win rates. G&A trying to work around proprietary data. I'm in tech and the productivity gains are real, but we will have to see post-fundraising. IMO.

Urban Country Boy said:

---

YouBet said:

---

Urban Country Boy said:

---

Anyone here that makes a comment about how they hate data centers are hypocrites. You really want to protest? Shut everything you do down. No Cell phone. No debit or credit card transactions. Walk into a bank for your balance. Only write checks. Dial from you home land line. Play games on your Atari.

And shut off your comments on a data center run site

---

We can maintain all of the tech we've already had for decades without hyper scaling DCs to handle...

---

Seriously, you were good with 14.4 dial up?

---

No need. My post was self-explanatory. You made an extreme absolutist and irrelevant comment that I responded to.

500,000ags said:

---

Bold statement. Product orgs definitely are trying to monetize their ability to push new work. GTM still identifying playbooks that increase win rates. G&A trying to work around proprietary data. I'm in tech and the productivity gains are real, but we will have to see post-fundraising. IMO.

---

Yep. Coding was the first real gains. It makes sense because code is native to AI. Intent is clear, it's deterministic, etc... To port AI into additional domains is ongoing but a bit more complicated due to the fuzzy nature, the governance, risk, etc... We are dealing with this at work. It's clear you can make the automation and Agents, but doing it at scale introduces new challenges. There's more tools and harnesses coming out, but adoption is lagging. It will come.

500,000ags said:

---

That IS NOT true. What goes inside are GPUs that are, by several multiples, more power hungry to handle parallel processing. Everything from back up power to ventilation still needs sorting on a case by case basis. And frankly, I don't think most local leaders understand developer representations.

---

It is true, it's just scale. And like I have said 10 times, it's not all GPUs.

Not unless you have a natural gas pipeline leading straight to it

Still at it, I see. When I looked, from 2010-2018, most data centers were pulling 20-100 MW. Goal now is for 1,000 MW. Can you not see the difference? You keep calling that scale, but that drastically impacts power, cooling, etc.

They will get no tax guarantees for 5-10 years. Then when the time comes to hit them with tax, they will threaten to close and move and they will extend the tax breaks.

500,000ags said:

---

Still at it, I see. When I looked, from 2010-2018, most data centers were pulling 20-100 MW. Goal now is for 1,000 MW. Can you not see the difference? You keep calling that scale, but that drastically impacts power, cooling, etc.

---

First, you are missing several years, they didn't just go from 20-100MW to a GW, it took years. They are also way more dense now, used to need over a million sq ft for 100MW, much more power can be utilized on that footprint now.

That is also the definition of scale, what was once a 100MW building is now a GW campus spanning multiple buildings. There are a ton of older sites in rural areas, microsoft in Quincy, Washington, and Cheyenne, Wyoming, AWS in Boardman, Oregon, google in Pryor Oklahoma, etc etc etc, these are all millions of sq ft of space and several hundred MW of power. Just because they dont have super spooky gpus doesnt mean they aren't energy and water hungry.

This entire thread is tragic.

Personally, I can't wait for the data center build out bubble to burst

It's going to be Broomfield, Colorado, 2001 all over again.

You'll be dead before that happens. We consume too much.

This thread has a lot of people talking past each other. For Texags, it is a pretty big disappointment.

Burdizzo said:

---

This thread has a lot of people talking past each other. For Texags, it is a pretty big disappointment.

---

This thread, like many, has people looking for confirmation bias and ignoring facts that fly in the face of their perception of reality, which to them should be a universal truth to all.

Unless we stop "consuming" data and need for cloud storage, data centers will be needed. Social media, internet, business use, phone photo storage, AI, saved Aggie football games, ……

I'm dealing with front side of DC builds. Infrastructure mainly - water/sewer/storm. The public perception is an issue owners/developers need to address early and publicly. Yes they consume energy and initially water to fill the loop. What I've seen these outfits pay for that public in some places are not aware of is crazy. Upgrades of water lines, treatment plant upgrades, roads, years of school funding, energy capabilities and capacity. They just really suck at getting ahead of all that.

Now I'd not want to live right next to one. But being out in country farmer John's kids gonna sell off once he's gone.

HollywoodBQ said:

---

This entire thread is tragic.

Personally, I can't wait for the data center build out bubble to burst

It's going to be Broomfield, Colorado, 2001 all over again.

---

I don't think there is a build out bubble because they are limited by the power shortage and backlog of gas turbines for power. They don't have a way to power all of the data centers they want to build. Many announced data centers are being delayed or the plans being abandoned.

There probably is an AI investment bubble though.

Thaddeus73 said:

---

H20 in Texas is a problem, and these things require a lot!

---

Aren't the cooling systems closed looped? In all of the big systems I seen on government installations they are mostly closed loop.

What I see are hundreds of air conditioners mounted on the roofs.

Pre-AI, there were ~8,000-10,000 data centers and ~1B square feet of data centers in the U.S. They consume ~150B gal of water and ~300 TWh of power per year. Not including those which we built abroad.

But but but Daaaave the $/sf and w/sf required by AI data centers are higherrrrr. Should we include domestic and outsourced manufacturing, shipping, storage, etc centers as well? . Or maybe we should just shut down the industrialized society if buildings and technology are bad.

Anyways you can shout at the clouds bc they're coming whether you like it or not.

bmks270 said:

---

HollywoodBQ said:

---

This entire thread is tragic.

Personally, I can't wait for the data center build out bubble to burst

It's going to be Broomfield, Colorado, 2001 all over again.

---

I don't think there is a build out bubble because they are limited by the power shortage and backlog of gas turbines for power. They don't have a way to power all of the data centers they want to build. Many announced data centers are being delayed or the plans being abandoned.

There probably is an AI investment bubble though.

---

Related to this, Elon was discussing this at Davos this year and said at some point this summer we will have chips sitting on shelves because there aren't enough DC's available to use them for all of the planned AI build out. If he's correct, then we are going to hit a wall this year on DC's anyway.

It can easily be argued AI investment is a bubble. The sheer amount of circularity in that space deems it to be so.

I agree with Burdizzo(?) that these things are going to happen in some form or fashion pretty much no matter what so we have to figure out a way to force them to be good neighbors.

Data centers are getting started up on diesel generators while waiting for gas turbines. (Artificial) Life finds a way.