Quote from: Apocalypse on October 28, 2025, 01:05:24

Quote from: GeorgeS on October 27, 2025, 18:37:48When is the question

IMO, it's already kinda begun. We saw how Jensen was frantically flying to Beijing when China banned their biggest companies from buying Nvidia gpus. Then the biggest AI companies are all switching to and making contracts for custom in-house ASIC solutions as an alternative to Nvidia hardware. Lastly, there's the drop in Nvidia GPUs in consumer market below MSRP which is pretty incredible considering that their stuff never drops in value. People keep saying it's due to lack of demand but was there any demand to begin with? DIY market is comparatively tiny to begin with and Nvidia has been overpriced for 6 years now. Feels like it's more of a result of AI bubble finally popping and no longer taking all the resources and buying all the gpus.

GPU prices have been high first due to crypto craze, then the AI craze. I know even my AMD card that I bought on a deal for 180 went up to 500-600 in price many years later. Nvidia did even better in the AI craze due to CUDA. But these days more and more libraries are supporting not just CUDA.

Though personally I don't see the AI craze lasting that much longer. At issue is they already data mined the entire public internet. They have run out of data to train the AI. It is so bad, they are even going backwards because as AI slop fills the internet, it poisons the training data.

Currently there is a big craze for making deals to access people's private data to train AI. But even that is bound to hit road blocks be it from privacy outrage, extra cost of these deals or again running out of data as training processing power increases, new data simply can't keep up to train.

Quote from: GeorgeS on October 27, 2025, 18:37:48When is the question

IMO, it's already kinda begun. We saw how Jensen was frantically flying to Beijing when China banned their biggest companies from buying Nvidia gpus. Then the biggest AI companies are all switching to and making contracts for custom in-house ASIC solutions as an alternative to Nvidia hardware. Lastly, there's the drop in Nvidia GPUs in consumer market below MSRP which is pretty incredible considering that their stuff never drops in value. People keep saying it's due to lack of demand but was there any demand to begin with? DIY market is comparatively tiny to begin with and Nvidia has been overpriced for 6 years now. Feels like it's more of a result of AI bubble finally popping and no longer taking all the resources and buying all the gpus.

Quote from: anan on October 27, 2025, 09:06:00Those companies are building out AI capacity for a projected demand that might not materialize. There are articles that China currently has an oversupply of AI capacity. Data-centers built specifically for AI workloads on government subsidies are underutilized. Some have resorted to selling off GPUs from datacenters in order to recoup costs.
Like, sure, nobody knows if this will all implode. But the signs sure as hell point to a not so healthy situation.
This is why DeepSeek was so disruptive - not that it was better, but because it did the same thing with less hardware. And other companies have sunk most of their investments into the 'necessary' hardware.

Implode? Sure. When is the question.

As Technology companies dump Trillions into "AI" investments without ANY idea what so ever how it might play out to their benefit, they are ALL more scared of being 'behind' if/when a way to actually make any money off of "AI" ever comes to light then the nearly countless funds they have tossed at it.

In the mean time the "gold rush" is in Nvidia's favor. :)

Those companies are building out AI capacity for a projected demand that might not materialize. There are articles that China currently has an oversupply of AI capacity. Data-centers built specifically for AI workloads on government subsidies are underutilized. Some have resorted to selling off GPUs from datacenters in order to recoup costs.
Like, sure, nobody knows if this will all implode. But the signs sure as hell point to a not so healthy situation.
This is why DeepSeek was so disruptive - not that it was better, but because it did the same thing with less hardware. And other companies have sunk most of their investments into the 'necessary' hardware.

Quote from: A on October 25, 2025, 16:27:53nobody has found a way to make money on AI.

You heard it here first folks, the biggest AI bro on NBC, admitting it.

;)

The whole AI market will burn down with only the hardware suppliers making any money.

Because so far, nobody has found a way to make money on AI. People are just buying up AI hardware but can't make enough return on it. Those trying to host the hardware are also paying high prices for the hardware but as dedicated processors for AI are made, all the old hardware becomes much less useful

The whole financial system and wall street is rigged and run by fraudsters.