Or to put it another way, a semiconductor, car or airplane factory wouldn’t really count as “infrastructure”, would they?
Then the public should make them look like fools and not pay for AI. But if people are paying for it, then they are probably capturing most of the value it produces. Paying Anthropic $20/mo is a trivial cost to recoup if it assists with your work, even at $60 or $100 it still makes sense for many.
I'm currently enjoying subsidized AI usage at relatively cheap prices from these same companies. It's cheaper than a cell phone bill and most people can afford it.
"Or to put it another way, a semiconductor, car or airplane factory wouldn’t really count as “infrastructure”, would they?"
No, but flying on an airplane 50+ years ago was only for the rich. Everyone I know now has flown somewhere, and it's very obtainable. I remember cell phones in the 80s being > $1 minute and a sign of being wealthy.
Capitalism can and does bring the costs of these things down.
Data centers will definitely benefit the public, as long as the government doesn't decide on creating their own monopoly.
If you are for green energy and the electrification of primary energy this giant wave of unlimited private investment is the best gift you could have ever dreamed for. The fact we are doing our best to squander it is amazing to me.
We have a generational chance to build entire industrial supply chains and rapidly catch up from multiple generations of lack of investment in our electric grid but instead we are going to resist it every single step of the way.
It just might have to do with where electrical grid interconnect happens to currently be with spare capacity. These corridors also happen to be where it's going to be easiest to put up even more long distance high voltage transmission.
Datacenter siting has went from "where the fiber lines run" to "where we can get power" rather rapidly. I have friends who's entire job is basically trying to find locations where there is still transmission capacity available, and if not immediately, where it can be built out rapidly over existing rights-of-way to avoid as much of the NIMBY as possible. Projects that were taking a decade or more are now being planned for completion in 1-3 years. Delays went from permitting, lawsuits, funding, etc. and are now seemingly gated by component availability. These manufacturers are ramping up as rapidly as they seemingly can do so.
Transmission lines are by far the hardest part of the puzzle to solve. And yes, datacenters are about the only thing that has accelerated their buildout in my lifetime. I've watched projects that were paused for close to a decade to bring wind energy from the southern part of my home state to the metro load center area all of a sudden get jump-started again due to there actually being immediate capital available to build them and make it worth dealing with the NIMBY folks.
Datacenter siting is obviously more complex than this, but yeah - if you can get 80% of your transmission line from a power generating locating closer to your urban load centers you would be an idiot not to jump on the opportunity regardless of the reasons why.
Wide scale infrastructure build-outs are always messy and wasteful. Every single one in history has been. This is how shit gets built - not 20 years of careful planning and consideration.
Some of us have been paying attention to the grid far longer than this latest AI bubble craze. This was always coming for us, demand simply got brought forward half a decade or more. Eventually you run out of the investments your grandparents made and have to actually build stuff again. The US simply has collectively forgotten what that looks like.
There will be plenty of stranded misallocated capital. That's just how these things go. But there will be plenty of reusable infrastructure left in it's wake like always. If the US is ever going to reindustrialize at all, then this is about the easiest and least impactful means to begin doing so.
Yes, they want to put these things near existing power infrastructure (ie cities), but per NIMBYs they cannot. To the extent they can use existing power infrastructure, they are using existing power infrastructure and then only adding what they need. In lieu of a significant bust, this is far more likely to be a negative externality than a positive one.
Of course people realize this, so they do not want to build data centers nearby, ergo data centers are increasingly likely to be sited "in bumfuck nowhere," even up to and including literal outerspace.
Here, the positive externalities are basically zilch regardless of bust or not.
I get the 30,000 foot view that you are describing, but I am saying specifically in this scenario, for this infrastructure, in the specific existing political/economic/land use context we are talking about, I don't see the argument.
Past infrastructure booms have built infrastructure in and around cities. You have just described why that is increasingly not the case, and even if it were the case, it's only a positive externality if there's a giant bust, which feeds back into people wanting to keep these things away from cities!
Please identify a few of the specific infrastructure booms that you think are analogous.
Dams: nope.
Telecom: nope.
Broadband: nope.
Electrification: nope.
Fiber: nope.
None of these look like "build infrastructure in middle of nowhere where if the project succeeds, it consumes everything it created, and if it fails, there's nothing very useful for others to consume that infrastructure in those locations."