In order to get numbers like 5 million gallons one has to be looking at the very largest data centers, counting all water use as single use, even though water used for cooling is often reusable, and counting all the water used not by the center directly but used for power plants also as discussed earlier. Typical data center consumption is much lower. For example, see https://www.brookings.edu/articles/ai-data-centers-and-water/ which has one of the high-end estimates for what a typical data center consumes. As for the idea that there's a lot of data centers being built in the Southwest, more are being built or planned to be built in California or on the East Coast. Northern Virgina is the fastest growing region for data centers. See map here https://usdatamap.com/ (This isn't a perfect map. The situation is in flux. And admittedly, this map doesn't show size of them. My impression is that at least some of the ones being built in Arizona are very large so the map here isn't showing everything.)

Yes, building some of the largest data centers, making them all near one city, would take up a lot of water. However, that would be silly; the people building these are not idiots and aren't going to go shove all their centers in a region they know they then won't have enough water for all of them. Moreover, in many jurisdictions, one has historical water rights to contend with. In many jurisdictions for major water resources, historical users get priority over new users, so farmers and others would get priority before data centers if it came down to that. (Yes, this does mean that in parts of California, golf courses get priority over some other uses.)

This is not remotely why AI systems are being used. ChatGPT is being used daily by hundreds of millions of people https://explodingtopics.com/blog/chatgpt-users. Right now, ChatGPT is the 5th most visited website in the world by some independent metrics. https://en.wikipedia.org/wiki/List_of_most-visited_websites. These systems are not being used just because some rich people want to not have to pay people or bother with civility. The regular, common people are using them. Understanding where this is going, the impacts it will have, both positive and negative, requires understanding the actual usage, not what one imagines it to be.

The water use for AI seems to be greatly exaggerated. Estimating water use complicated. Different data centers use different amounts of water. Also, systems need more water use for cooling when the weather is hot, so centers may use more water in summer. A data center will use more water when the center is at close to maximum usage, so data centers will use less water if they are handling queries when few users are using the system. Complicating things even further, some people are counting not just data cooling water but also counting the indirect water use from the needed electricity production (fossil fuel and nuclear plants use a fair bit of water for their steam turbines). There's a good article here discussing the difficulties in making water estimates https://theconversation.com/ai-has-a-hidden-water-cost-heres-how-to-calculate-yours-263252 However, all things considered. they estimate that all things considered it takes about 39 milliliters of water per a typical query. Now, for comparison, a high efficiency shower uses about 1.5 gallons of water a minute, which is about 95 ml of water a second. So making a query to an LLM AI system costs less than a second of water. If this estimate is off even by a factor of 3, this is equivalent to taking 1 second longer on a shower. The water use is just not hat high. The total water use is also not very high. If for example you use estimates for how much water is used by golf courses in the US https://www.usga.org/content/dam/usga/pdf/Water%20Resource%20Center/how-much-water-does-golf-use.pdf, the largest estimates of AI use put the water use as about a tenth of the water use by golf courses, and golf course water estimates put it at most about 1% of total US water use. So even if one is concerned, just getting rid of some of the gigantic water hungry golf courses in California and Arizona (seriously who the heck puts a golf course in Arizona) would largely offset this. Now, it is true that as data centers grow, more water will likely get used. But as we switch to more wind and solar power, the indirect water use will go down, and data center builders are working hard on reducing water use since it is such a hotbutton issue.

There are a lot of legitimate concerns about AI. Water use should not be high on the list.

Going to have to agree to disagree. I still have a MacPro5,1 from 2012 that I regularly use. All four drive trays are in use, both optical drive bays, and I have two PCI addon cards for added functionality. The expansion capabilities of the MacPro5,1 were absolutely useful and justified.

Apple's Mac Pro, and before that the Power Mac, used to be a reasonably affordable machine for the capabilities it offered. The trash can was silly, but still affordable.

The 2019 return to tower form also came with an insane price increase. The base price was double that of previous generations. That killed the Mac Pro.

It's about time they finally had the funeral.

These issues are all very minor. Submarines are cramped and people can remain incommunicado for months. Messages to Earth is essentially just means one will be relying on email equivalent. And vacuum packed food can keep for years. Drinking water that is reclaimed from pee isn't fun, but isn't a big deal.

Mechanical problems are a big issue. That's why for example even today submarines carry some replacement parts, and why big surface ships historically had machine shops. Some things will need to be carried. But others may need to be ready to be made on site. That's why there's now a 3D printer on the ISS, to get used to doing exactly this. The gravity issue isn't a large one: on a 90 to 120 day mission to Mars, the level of bodily deteriorating isn't that big. We don't have a lot of data for the exact "how bad is it" but I'd strongly suspect that 3 days in microgravity and then 2 years on the Moon is going to be much worse than 90 days in microgravity and then 2 years on Mars.

A 90 or 120 day trip just isn't that big an issue. People have spent far longer on the ISS and managed fine.

People don't go insane and murder each other nearly as much as they do in movies. Again, look at submarines. The degree to which there are psychological issues is drastically exaggerated.

One of the lessons from the ISS and Mir is how incredibly difficult and expensive in-orbit construction is. Now, we've learned from it but its still a big issue. If you want to send 12 people then the most pessimistic estimates give you around 250 tons of stuff for a 2 year mission. That's only slightly higher than the Starship design, which is capable of being built on Earth, and masses about 5000 tons. Serious on Earth-designs for rockets have been bigger than that. For example Sea Dragon was going to mass 18,000 tons. And you can also do things to reduce the difficulty by launching multiple rockets, say one with the humans and the stuff they need for the first few months, and then another just full of cargo.

AI systems are useful to many people. Most users aren't getting it as an investment engine itself. They are using them to do things like assist with programming, look over drafts of documents, do specific searches and other things. Don't mistake valuable to you personally as the sole form of relevant value.

A Mars settlement has the long-term ability of self-sustaining. And that same benefit means that it is much easier than the Moon to make it while not fully self-sustaining, involve a lot less maintenance and resource. One can likely more easily grow food, obtain water, and do many other things. The higher gravity also means that the health issues for humans in a low gravity environment are likely going to be lower.

Ok.. This is a reasonable list in part. The gravity well point is a pretty reasonable one; if one does want to build large structures in space, then having the much lower lunar gravity to start with is reasonable, and mining there doesn't come with the same environmental damage it does on Earth. The telescope point also makes sense, especially for radio telescopes which would be in the lunar shadow. I'm not at all convinced by your last point; it has a pre-existing fixed orbit which severely limits when observations and communications can occur, and that would require much larger systems for Earth observation since the moon is far away, and would add really not good latency for communications due to the moon being about a light second away from the Earth. But your point about resource extraction/processing and telescopes is enough to make a strong case that there's a reasonable goal set here. So, yeah, my statement there was wrong.

Seems like a suboptimal use from a military standpoint. Sure, it works in Moon is a Harsh Mistress, but in practice, the travel time from the moon to Earth means it will be a minimum even with very high acceleration of about 6 hours to hit a target on Earth. Much more effective to have satellite based weapons in closer orbits.

Trip time is a valid concern and ability to get home quickly in an emergency is also valid. Another issue with distance is the ability to send emergency supplies is very limited. Radiation issues are not really that bad though (and that seems like more of a problem). The radiation level is high https://pmc.ncbi.nlm.nih.gov/articles/PMC11814067/ https://ntrs.nasa.gov/api/citations/20240009831/downloads/NAS%20BPS%20Simonsen%20v4%20strives.pdf but manageable. A 6 month (comparably long) trip to Mars adds around 300 mSv which is slightly over half a year of time on the ISS. For comparison, if one looks at some of the most intensely naturally radioactive places in the world that people live, like Ramsar, Iran, one gets around 150-250 mSv.

If there's no goal of making an eventually long-term self-sustaining colony, it isn't clear what goals there are here that are a reasonable use of resources on this scale.

Not really relevant. That's a canonical example of a bubble. It is incidentally, an exaggerated bubble in many respects. See https://www.smithsonianmag.com/history/there-never-was-real-tulip-fever-180964915/ https://link.springer.com/article/10.1007/s11127-006-9074-4 and https://www.bbc.com/news/business-51311368 (although I find some of the arguments in the second of those not completely persuasive.) But the real key is this isn't relevant: tulips weren't a tech. The point is that tech bubbles can be very different than other bubbles because the underlying good can still turn out (and often does turn out) to have real value and continue to expand. Tulips are like gold or other nice shiny things, (or bitcoin which is like gold or tulips but doesn't even have the advantage of having something to look at) not really the same sort of bubble.

Previous tech bubbles have ratched up prices and caused all sorts of issues. The rail boom outcompeted other transport types, increased the cost of coal and iron( vital resources), and completely took away a lot of sources of wood. In parts of the American West, rails took up needed limited water for the steam engines, probably in a way far more serious than that of data centers (which does take up water but is largely exaggerated). The rail boom also prompted tens of thousands of Native Americans to be kicked off their land to make way for rail infrastructure. In the American South, freed slaves were functionally reenslaved and forced to build rail lines. A lot of this is discussed in for example, Wolmar's excellent book "The Great Railroad Revolution." The internet bubble of the late 1990s was in comparison pretty mild in terms of its negative knock-on effects.

This is pretty similar to the "dot com" bubble of the late 1990s, or the railroad boom of the late 1860s early 1870s. Both bubbles had people investing with minimal justification. Businesses which were "Normal thing X but on the internet" got insane amounts of money. And both bubbles popped in ways which were very harmful to the economy. However, at the same time the underlying techs stuck around and became far more common than they were at the height of the bubble. The 1873 collapse took the US about a decade to recover from, but by multiple metrics, such as number of train engines being used, number of passenger miles traveled per a year, tons of freight moved, rail continued to grow with just a small set of blips. Similarly, the internet now is far more extensive and used than even many starry-eyed optimists or hype-obsessed would have predicted in 1998. A tech can be part of an irrational bubble and still be a game-changing technology.

Mars is a much better destination for long-term habitation. The moon's gravity is much lower and doesn't have the benefit of even a thin atmosphere. The moon also has very little water and has little in the way of most of the CHNOPS humans need to survive. Mars has all those resources. https://pmc.ncbi.nlm.nih.gov/articles/PMC8227854/ A self-sustaining colony on Mars may be possible but is very, very difficult. A self-sustaining colony on the Moon is essentially impossible. Also, actually getting to Mars is not that much harder than getting to the moon. Once aerobraking is allowed, many launch profiles use only a bit more Delta-V to get from Earth to the Mars than going from Earth to the Moon. https://www.sciencedirect.com/science/article/pii/S0273117725002406 https://arc.aiaa.org/doi/pdf/10.2514/6.2023-72854.

There's also sometimes an argument made that the moon would be a good learning place for eventual Mars missions. But the environments are drastically different with not even the same major risk factors in many respects, so only learn a little bit from a base on the moon. The best way to learn how to live on Mars is to go on Mars.

Finally, there's a decent chance that we'll learn really neat stuff on Mars, including possibly finding extinct life or even current life. There's no likelihood at all that that's going to happen on the Moon.

What appears to be happening in part here is that Trump has a particular interest in going back to the moon because he imagines things the United States did in the past, especially things from when he was a child, as the greatest things ever.

All of that said, this is better than the idea for the in lunar orbit "Gateway" which was an absolute disaster of an idea where it would take extra Delta-V to get too and then more to go then leave and land on the moon. It . Gateway was based on that they had to do something that could justify the SLS even if it made no sense aside form that. So there's that at least.

Because there are both. We need to do a better job tracking people down like this and making it clear that this is not acceptable behavior, since functioning societies rely in a large part on trust of people not being assholes. But there's also a bug in the car here since human bad behavior is probably never going to end: in particular, there should be a manual driving override option so a human can get out of a seriously dangerous scenario. Unfortunately, elsewhere on the internet, there are already some anti-self driving car people who are seeing this as functionally a win for them.

The Guardian is hardly "deep red" but if you want, a basic Google search will find you many other references with the same essential details. This is not a situation where the basic facts are remotely in dispute.

Instead of sarcasm, insults, and strawmen, it might help to read what people write, and when they give a link click through it. Do you want to go and read the link about Trump's FCC threatening to remove broadcast liscences?