We need to win in AI and to do that we must have data centers. The solution I believe is for the people building them to get creative.

1. Build them in the industrial part of town. I'm from Michigan, there are neighborhoods in our cities filled with manufacturing firms stamping steel and making all kinds of noise with few houses. Yes the real estate can be more expensive and sometimes needs pollution removed but there are usually willing economic development departments willing to help.

2. Make the data centers bring their own power.

3. Find ways to creatively help the community. Saw pictures of a data center recently where they created two huge public swimming pools that are open all winter long, There is a power plant on Lake Michigan where they heat all the sidewalks. Imagine waking up in the morning and not having to shovel or spread ice before going to work.

4. Find ways to repurpose unwanted buildings. Detroit wants to tear down two of the five towers of the Renaissance Center which is on the Detroit River. One of the towers would have the first two floors occupied by the University of Michigan which would offer training classes on technology to the community. The rest would be a data center for the university. Power would be two gas turbines on the roof. The other tower would be a partnership with Detroit Public Schools that would offer a dormitory for all the school age kids living on the street. Educate these children from 6-18. Most American cities have at least one empty skyscraper that could be repurposed as a vertical data center.

5. Repurpose old shopping centers as data centers. South of the Mason Dixon line where solar has a higher ROI you could cover the entire parking lot with solar cells. You could offer free or nearly free shaded parking, maybe even let campers have extended stays.

▲

breakingstuff 7 hours ago | parent | next [-]

It blows my mind that you don't hear more about repurposing old malls and other abandoned large properties as data centers. Also, the fact that hyperscalers don't think more about how their buildings could add more benefit to the community they are in shows just how tone death they are. If they don't change their approach soon, no way capacity is going to catch up to demand.

	▲	pibaker 3 hours ago \| parent [-]
		> you don't hear more about repurposing old malls and other abandoned large properties as data centers The building is not the hard part. The hard part is getting enough electricity to run GPUs 24/7. Old malls' electric connections are not powerful enough for that, so you are going to either spend money on new infrastructure anyway, or park a few natural gas turbines in the parking lot.

▲

tzs 4 hours ago | parent | prev | next [-]

> We need to win in AI and to do that we must have data centers

What exactly do you mean by "win in AI"?

▲

add-sub-mul-div 7 hours ago | parent | prev | next [-]

"Winning" in AI could mean standing by while other countries race to atrophy the minds of their citizens.

▲

rapsey 7 hours ago | parent | prev | next [-]

Bring your own power means natural gas turbines or diesel generators. Both of which produce incredible noise pollution.

▲

bigstrat2003 7 hours ago | parent | prev [-]

> We need to win in AI

No, we do not. There's no prize to be won, nothing of value to be gained.

	▲	Larrikin 7 hours ago \| parent \| next [-]
		Both sides of this argument have not justified their stance
	▲	moron4hire 7 hours ago \| parent \| prev [-]
		AI Inevitablism. This is a concerning thing. Folks talk about AI as if it is a forgone conclusion. But it has yet to be demonstrated. I'm stuck between a rock and a hard place right now. I work at a company that claims its people are the source of its great work output, yet the key stakeholders for my particular project are constantly beating the "AI, use AI" drum. I've been trying to design a product that enhances our analysts abilities. A middle ground where the subject matter experts use AI to do the boring, manual labor kind of work that doesn't enrich anyone and just leads to our organization burning out junior analysts with overtime they'll never get compensated for. But my stakeholders keep beating that drum. "AI can do this work from front to back." To be clear, it can't. We've done the research to figure out that any sense that an LLM applied to the kind of work we do is only a dilettantism. It looks good if you are skimming the output, but drilling down deep there are massive problems. But that story, "AI is good now. What did you try last year? What model did you use? It can do so much now." Is pernicious. First of all, the models today I don't see producing anything functionally better; they just dress it up in better language. Second, that's not an actionable software engineering plan! "Oh, just wait a year, the AI will get better". Sure, it gets better at not completely shitting the bed before you coax it into doing a particular job. But it hasn't been getting better at being actually insightful, actually delivering on what our people with very deep experience can do just by rote, just by asking them, "what do you think of <insert competitor>'s capacity to deliver X compared to our ability to do same?" I feel like I'm living in crazytown. I evaluate AI capability much more than what my stakeholders do and they keep telling me "more AI!" If it weren't for my mortgage and my kids and my junior devs I'm desperately trying to protect, I would have quit months ago.