Looks like the styling got goofed up. Sorry about that, unless it's what you wanted. If this isn't what you were looking for, try force refreshing your page. You can do that by pressing Shift + F5, or holding Shift and clicking on the "reload" icon. (It's the weird circle arrow thing "⟳" just above this page, usually next to where it says https://blog.unitedheroes.net...)
First, I will acknowledge that I am generally skeptical about Large Language Model based AI. I have a number of concerns ranging from the fact that the "training" data set are often collected questionably, the training weights are almost never disclosed, and using an LLM for anything other than general language based taking seems ill-advised. Using an LLM to do things like write code seems to me like creating an algorythm by clipping stuff out of back copies of PC Magazine.
Second, I will note that there are absolute use cases for LLMs. They are a tool, and like any other tool, there are things they excel at and things that they do poorly. I view them a bit like I view tablesaws and lathes. Very capable devices when used with knowledge and care, significantly dangerous devices to be readily unleashed upon an unsuspecting populace.
My concerns about LLM/AI in its current form go beyond that. (I'll note that I am having dinner with my financial advisor later, so yeah, this might come up in conversation, and I figured I should share.)
Currently, there are ~5400 Datacenters in the US, and a total of 6,923MW of capacity (Data centers are denoted by electricty consumption because I guess you need to measure something). This means the average datacenter consumes about 1,282kW per hour. The average home in the US consumes 10MW per year, so we can say that a home consumes about 1.14kW per hour. This means that, on average, one datacenter is equivalent to roughly 1,000 homes in terms of power, and cumulatively they use the equivalent of 6,073,807 homes. Mind you, power is cheap and more renewable now, so it's less of a concern.
Data centers also use evaporative cooling systems, because everything gets hot. The amount of water used by a given data center varies greatly, and it's hard to find a reasonable breakdown, but if we presume that only 10% of the total number of datacenters are "Hyperscale" and use 550,000 gallons per day and the rest use 18,000 gallons, that means that every day, data centers consume 384,480,000 gallons of fresh water. Comparitively, the average person in the US consumes 82 gallons of water a day at home, so a family of 4 would consume 328 gallons. The average data center therefore consume about 1,006,492 homes worth of water, per day. I'll also note that there are significant portions of the US currently under drought conditions. Bonus points if you overlay the map of data centers with drought.
Since 2024, the number of datacenters being built just in the US has increased by 34%, with a vacancy rate of 1.9%. Locations like Atlanta have seen a several magnitude growth. It's no secret that one of the biggest drivers of data center expansion is to accomodate machines to do AI. Mind you, while technically, you could set up and train an AI model at home, you really need a lot of machines working in parallel in order to do this correctly. That means, you need to have a fairly big wallet in order to pay for this. That means that the companies able to do this either are already very large (Google, Microsoft, Meta, etc.) or have enough backing that they can compete (OpenAI, Twitter, etc.). There are some companies also competing by incorporating AI into their products, of course, but those are just customers of the providers. That's kind of important.
If you consider the S&P 500, the top seven companies are either building AI networks, or providing the chips for AI. Those seven companies comprise a market cap of about 21,000ドルB dollars. Which is a lot. Mind you, the total investment number for this is probably a lot higher because of the fact that a lot of the other companies also have pinned their dreams on AI.
So, a lot of big companies, ones that are driving our economy, have investments that are either in or reliant upon AI.
Which leads to the next question. What is the killer app? What is the one thing that you, right now, can not imagine getting through the day without that relies on that level of investment and expenditure? What app are you paying for every day because it's of the same value as your internet connection or your subscription to Netflix or Amazon?
A few might immediately specify something like ChatGPT, but let's look at that. ChatGPT isn't going to disclose how many paying daily active users they have, but they do offer some numbers that can help us figure that out. About 10% of the world users live in the US. They have 10 million paying subscribers so one can presume that 10% are also in the US (probably more), so let's presume that to be one million people pony up at least 20ドル a month for the service. Since ChatGPT is the biggest user, and because we can be sure that there's some overlap, let's presume that there's another 100,000 folk that use some other AI service that's not ChatGPT. That gives us a rough user base of about 1.1 million US users of LLMs. Impressive, before you realize that there are 320 million residents of the US. Basically, all this is being built up for most of the population of Dallas, TX.
So, yeah, my concerns...
We have a growing system that consumes about as much power as about twice the population of Los Angeles, about as much fresh water per year as Sydney Harbor, to satisfy the needs of the population of Dallas, Texas. This does not seem sustainable, pretty much by any rational measure.
I didn't even mention the fact that data center capacity backup generators and cooling towers are both in high demand and short supply, driving the prices up on those for any new construction, or tariffs kicking up the cost of, well, everything, or all the other problems.
Having this much money wrapped up into something with so little potential return reminds me of more than a few other bubbles I've experienced, like the Housing Bubble, and the S&L Crisis. I am a bit worried that this will not end well.
So, what to do? At this point, I'm not sure. I've heard some folk say that they were thinking about shifting some of their money into smaller cap stock indexes, but I'm not really sure that's much of a safegard if the whole thing is a house of cards to begin with. There's the usual group that will insist on putting their money into things like gold or other precious commodities, but again, not really sure that's a viable option unless you're going to have that gold buried in your backyard and have a trusted party you can use to convert it into something more transferrable. (John Wick may have Kuggerands to pay for stuff, but once one realizes just how much actual money is changing hands, you start to realize he seriously overpaid for a lot of things.)
I think we're going to be in for a wee bit of turbulance, again. Might want to make sure that your buckled in as well as you can be.
Hosted on Dreamhost.
See our Advertisement Policy .