More and more mainstream analysts are identifying the coming AI crash, which is a good indication that it will happen soon.
So, what happens to all the data centers? They are already built but probably very expensive to maintain. Will many of them just be abandoned? Bought up by cloud computing companies? Scammers? Crypto miners? Can they be parted out and sold off piecemeal?
Will they be put to some productive use, or just become massive e-waste sites left to the locals to deal with?
The operating costs of AI are prohibitively expensive even by the standards of US military budgets, and they are not the utility we were promised. Companies were told they could replace their entire workforce with AI and not only is that not true but now the subscription costs are going up.
It doesn’t matter how much consolidating you do the product still isn’t compelling and it’s still expensive as hell to operate.
Anthropic is expected to post a profit next quarter, it may prohibitively expensive to run but that doesn’t mean it can’t be profitable. But who knows, maybe they won’t.
I read somewhere that for every $1 they’re taking in in cash they’re spending $100 in operating costs. If they charge the amount of money they would need to charge in order to actually have net income they wouldn’t have any customers.
We’ll see if they post a profit or not but if they do then your source was probably an AI hallucination.
I’m not saying changes aren’t coming, if they can’t make a profit off it then they’ll need to adapt or die. So it will either get more expensive or more efficient, but I don’t believe the bubble will burst and there will suddenly be a fire sale on cheap data centres or their hardware.
The issue is really the use cases for AI have been way overblown. The AI companies sold it as an employee replacer. “It’ll cost pennies on the dollar, have better accuracy, and doesn’t need time off!” That’s not actually what it is though. Companies are coming to terms with AI being more expensive, less accurate, and dumb.
A piece of software I use at work is a great example and they barely oversold the capabilities. The software has a built in AI assistant which is pretty good at pointing you in the right direction to get the result you want. Then some genius realized it could also pull report data, so they removed some metric options from the reports and now I have to spend significantly longer prompting the AI to pull the metrics I want in a usable format. This is also far more computing power. So they decided to let you link other AI agents to their software so you could use another bigger model for “better results” (they don’t want to pay for all that AI themselves). Nobody bit, nobody linked to another AI because nobody wants to pay more. They’re quietly adding those metrics back to the reports now.