I’ll agree that there are 3 bubbles AI is riding to burst, but not at all that there will be a silver lining to its bursting.
As it stands, people just arent really mass adopting AI, no matter how hard they’re pushing. We’re not in any sort of economy where the small use cases can hold it up after the crazy valuation it’s undergone. People cheating on homework and making slop content for instagram and tiktok arent enough to keep the engines running even after the bubble pops.
This makes the claim that the infrastructure will find cheap use dubious because of large loans taken out to build the data centers will more likely find them shuttered. These arent water or power sources for municipal use, it’s business. Especially if Ed Zitron’s reporting is accurate, there’s been a “paying myself to furnish gpus” scheme happening with some of these data centers, so big money’s been sunk already that will need to be recovered in profit or operating costs. The broadband infrastructure upgrade was something that the government told telecom had to be done. There’s no necessity to these data centers. They will be closed when the loses seem inevitable.
The advice at the end is basically “try running or exercise when youre depressed” but for AI business. I doubt any of the existing ai companies will pivot like this to ride out the coming storm.
I’m fully expecting a dotCom bubble style disaster
Atom Bomb vs Coughing Baby type shit
I’m interested in what will happen to the AI infrastructure they’ve built. Some of it can just be repurposed for data storage or server space or processing power, but some of it is specialized in a way that makes it only really useful for running AI models. If that use-case shrinks there are going to be a lot of stranded assets that they built out in preparation for a new paradigm that never came.
The gpus to run coda or whatever the coding language is called are already starting to yield comparatively low returns for their price compared to the previous gen of hardware, at least to reporting I’ve heard.
Be ready for a fire sale
You mean CUDA. And it would be rather nice to be able to build a proper gaming PC again wouldn’t it.
cursed with only just enough knowledge to get it wrong
yeah, it would, even given i would be out $450ish cause i’ll have to upgrade my psu to get a XX60 card to be mid tier again
The issue is that datacenter GPUs for ML cost 30 grand, are bigger than most PCs, draw insane amounts of power, have no fans, and there isn’t software support for Vulkan or anything
Someone will get them to run counter strike
maybe steam could launch a Stadia-like service to sell more skins?
cloud gaming requires you to live within 50mi of a data centre for input latency not to ruin everything.
cursed with only just enough knowledge to get it wrong
And people think AI isn’t just like us
It’s also worth noting that any computing infrastructure has a fairly short lifecycle and becomes obsolete in a matter of years. I guess we’ll likely see this hardware being sold off on the cheap as companies go bankrupt and their assets get cleared out.
🫧
Can’t wait for thousands of useless refurbished NVidia cards to suddenly show up for dirt cheap.
Keeping an eye out for discounted rack equipment, my homelab could use an upgrade!Looking forward to running full sized DeepSeek at home. :)
