My conspiracy spidy senses are tingling.
The current model of AI known as LLMs have a direct relationship between how well they work, and how much computational power you can put behind them. More computers = more smarter.
Remember how every major tech company since forever has been sitting on just stupid amounts of cash? All of a sudden, they all had a thing that if they spent more money on, it would get more better. And all of them are betting the one that spends the most, is going to win capitalism.
Meanwhile, China has been sitting back waiting to undercut all of them. Because at the end of the day, the AI war is a production war, and China already won that title years ago.
FOMO. In addition to the below, AI is a huge bubble and the only way to win is to have the biggest and the most datacenters. So companies are in a rush to grow the fastest.
When the bubble pops, each hopes to remain on top, dominating a new industry and reaping untold wealth. However most of those companies will lose. Most of those datacenters will shut down. It will be “dead malls” all over again. It will be society impacted the losses of business excess again
Meanwhile a huge part of the problems they cause is we are way behind in infrastructure spending. We already weren’t able to get sufficient power everywhere needed. We already couldn’t easily support new power generation or large power users. Now we’re suddenly tripling projected power growth?
And infrastructure spending from the 2022 bill was mostly cancelled in the current political chaos. And we’re still not talking about being so far behind on infrastructure spending. We don’t seem to care that infrastructure in good repair would be much more resilient to sudden needs like this, much more able to support innovation
Obviously thanks to AI. But I will share some information I got on a call with Microsoft the other day, they have had to pause their roll out of the data lake feature because they offered it to everyone and too many people clicked the button to turn it on and now they have actually run out of resources in a lot of their data centers. So just dumb shit all around.
data centers are highly concentrated super computer clusters and no need to have conspiracy spidy senses, they outright said that it is to replace human labor and therefore the cost of building and running them is to replace the money spent on wages.
it wont work tho bcs llms simply cant work human jobs. doesnt stop the oligarchy from believing they can, and destroying our planet in the pursuit of achieving it.
Whatever they might be able to replace humans for, they can’t do it anywhere as cheaply as a human. So, the results would need to be very valuable. Outside of scientific research, which would still need to be verified and refined by humans, I don’t see a sustainable use for these things. They’re way, way too costly to use as a toy for the masses, that will soon be shut down.
This reply is not pulling any punches, the official line is something about freeing humanity from the drudgery of labour. But you are right, totalitarian grip on a dwindling demoralised population by the few who are too arrogant to see they are on the list too is probably closer to the mark.
Funny how up until now they always talked about the dignity of work, and now that they think they can replace us, labor is drudgery.
We need datacenters to enable all the Internet stuff. If you buy stuff online, binge TV shows, or rot your brain on social media, your phone is communicating with some back-end server in “the cloud”. And the cloud is all in datacenters now, because they are all connected to the Internet, and each other. So if you want to host an Internet service where everyone can get to it, even if it’s just a virtual server on a cloud platform, that virtual server lives in a datacenter.
The reason it is all hitting the news now is that AI hardware takes much more power than a standard Linux server. So new hyperscalar /AI datacenters require a lot more power than older ones. And in areas without sane regulations, datacenters can go up much more quickly than new power plants can go up. So, if local governments and regulators aren’t looking that closely, new datacenters can quickly drive demand that the local utility can’t service without buying more power on the open market. And unless the datacenter is required by local regulations to pay the the higher price for that extra load, the cost is divided across all consumers.
Also, this increased power consumption inside each rack comes with additional cooling requirements. Old datacenters can make do with air cooling, but these new ones all use liquid cooling. And the cheapest way to do that is evaporative cooling, which “uses up” all the water we’ve been reading about. (There are closed-loop methods that don’t “use up” water, but they are more expensive so the datacenters don’t use it).
So, really, it’s all a scaling problem. One datacenter doesn’t cause that much of a problem. But they grow in clumps, like a fungus. Dozens of datacenters in a small area can cause environmental havoc.
Capex demands constant explo-oita/ra-tion. It’s really just massive scale “maybe this will work? Anyone else got any ideas? No?”
AI task needs Lots of compute power. Ivestors think that AI will take over more and more task and that it is a great Invest.
