Meanwhile, elsewhere...
https://futurism.com/artificial-intelligence/ … ble-burst-china
ElectroSoldier wrote on 2026-10-28, 17:30:The AI companies are telling investors its worth it because nvidia says the GPUs are worth money even if it goes wrong. nvidia is trusted because its selling lots of GPUs to the AI companies.
Might I buy a GPU after AI crash?
Maybe.
Would I pay even half what they're going for now?
Hahahahaa hahahahha hahahahahhaaaaah hahahahahah
*wheeze*
Hahahahaa hahahahha hahahahahhaaaaah hahahahahah
*wheeze*
Hahahahaa hahahahha hahahahahhaaaaah hahahahahah
*wheeze*
Oh wait, for serious?
NO!
Unicorn herding operations are proceeding, but all the totes of hens teeth and barrels of rocking horse poop give them plenty of hiding spots.
Big Pink wrote on 2026-06-02, 16:22:Can someone please tell me when 'compute' went from a verb to a noun? I hate it.
wrote:"[Copilot] now powers far more complex, agentic workflows that consume far more compute."
Rearrange parts of that sentence and the truth emerges: Copilot consumes far more power.
In information science and physics, computation and power are pretty much synonymous. So, it always has been? In this case though it's word play for marketing.
interesting is the circular value reassurances we are getting from them
sound vaguely familiar from a few years ago. Perhaps taxpayers can bail them out? We dont want these companies to go down BEFORE the end of the world.
BitWrangler wrote on 2026-10-28, 19:10:Might I buy a GPU after AI crash? […]
ElectroSoldier wrote on 2026-10-28, 17:30:The AI companies are telling investors its worth it because nvidia says the GPUs are worth money even if it goes wrong. nvidia is trusted because its selling lots of GPUs to the AI companies.
Might I buy a GPU after AI crash?
Maybe.
Would I pay even half what they're going for now?
Hahahahaa hahahahha hahahahahhaaaaah hahahahahah
*wheeze*
Hahahahaa hahahahha hahahahahhaaaaah hahahahahah
*wheeze*
Hahahahaa hahahahha hahahahahhaaaaah hahahahahah
*wheeze*Oh wait, for serious?
NO!
Given the form factor their GPUs are in most people wouldnt be able to use them even if they wanted one.
ElectroSoldier wrote on Yesterday, 02:11:Given the form factor their GPUs are in most people wouldnt be able to use them even if they wanted one.
Even if they were in the normal GPU form factor, the Nvidia A-series and B-series "AI" GPUs can only do OpenCL and CUDA. only the much older H-series cards can still do graphics rendering APIs.
You'd have to somehow reimplement Vulkan, DirectX and/or OpenGL on OpenCL.
Just a silly lil person in a very big world.
VanillaFairy wrote on Yesterday, 23:10:ElectroSoldier wrote on Yesterday, 02:11:Given the form factor their GPUs are in most people wouldnt be able to use them even if they wanted one.
Even if they were in the normal GPU form factor, the Nvidia A-series and B-series "AI" GPUs can only do OpenCL and CUDA. only the much older H-series cards can still do graphics rendering APIs.
You'd have to somehow reimplement Vulkan, DirectX and/or OpenGL on OpenCL.
Exactly.
I did touch on this earlier in the thread, in that when this bubble bursts it wont be like it was when the dead dot coms equipment hit the market and drove the prices down.
That was one of the things that lead the drop in RAM prices back then.
But this isnt the same... It is a bubble, but the hardware is so AI centric it wont end the same way. There wont be a massive glut of hardware to drive down prices.
ElectroSoldier wrote on Today, 00:37:Exactly. I did touch on this earlier in the thread, in that when this bubble bursts it wont be like it was when the dead dot com […]
VanillaFairy wrote on Yesterday, 23:10:ElectroSoldier wrote on Yesterday, 02:11:Given the form factor their GPUs are in most people wouldnt be able to use them even if they wanted one.
Even if they were in the normal GPU form factor, the Nvidia A-series and B-series "AI" GPUs can only do OpenCL and CUDA. only the much older H-series cards can still do graphics rendering APIs.
You'd have to somehow reimplement Vulkan, DirectX and/or OpenGL on OpenCL.Exactly.
I did touch on this earlier in the thread, in that when this bubble bursts it wont be like it was when the dead dot coms equipment hit the market and drove the prices down.
That was one of the things that lead the drop in RAM prices back then.
But this isnt the same... It is a bubble, but the hardware is so AI centric it wont end the same way. There wont be a massive glut of hardware to drive down prices.
Cant even recycle the HBM memory from the DCs and any DCs still using DDR4/DDR5 will have already cycled most of it out for HBM or REG ECC server memory which is pretty much useless for desktops.
Even the storage wont be much good and I doubt many DCs will attempt to recycle that either, most likely itll get destroyed for data protection.
So yeah the AI bubble bursting wont have much if any useful hardware to hand back to the consumer market.
AI is inevitable, there is nothing you can do to stop it, no Ironman snap to save you from the Basilisk.
Apparently there's 300 racks with 9000 nVidia A100s been listed on eBay for 60M but I haven't dug it out to gawp.
Unicorn herding operations are proceeding, but all the totes of hens teeth and barrels of rocking horse poop give them plenty of hiding spots.
BitWrangler wrote on Today, 04:03:Apparently there's 300 racks with 9000 nVidia A100s been listed on eBay for 60M but I haven't dug it out to gawp.
A100 that's Ampere right, which is RTX 3000 series power .. sits roughly between the RX6900XT and the 4070, its pretty much only good for AI and has no capability for graphics processing.
But .. if you wanted compute power in a home server a pair of these second hand from a server farm wouldn't be a terrible thing. (napkin math puts it at roughly 6.666 grand per card ...still too much for a second hand DC compute card)
30 cards per rack at 200,000 USD per rack or 6.666 per card ....pretty pricey.
AI is inevitable, there is nothing you can do to stop it, no Ironman snap to save you from the Basilisk.
