Isn’t a lot of that money from circular cash flows where they give money to companies like openai to get them to buy GPUs via companies like oracle along with a giant IOU that they will never come close to being able to fulfill who then promise to buy GPUs for the data centers they can’t finish?
Yes, there is a large amount of circular money on the books.
But also they’ve induced a FOMO race to buy hardware where all the big tech companies are selling off chunks and raising money/debt as fast as possible to acquire more GPUs, lest they lose out on the AI race.
The cost to train a bigger model scales exponentially with increased performance (in theory, with current architecture). So you can only beat your competitors by buying a multiples mire GPU than them.
NVIDIA is making money, they’re also inducing demand by juicing the race to try and build that FOMO.
Unfortunately these companies are paying 20x more than consumers are for the same wafers, so for the allocation of RAM to become favourable to keep consumer lever RAM production running prices will need to increase at least that much.
Yes. Ultimately even hardware sellers in this scenario need to be paid by endpoint AI subscribers whose money will filter back along the investment circle.
There is loads of profit, if you are making GPUs or memory.
Isn’t a lot of that money from circular cash flows where they give money to companies like openai to get them to buy GPUs via companies like oracle along with a giant IOU that they will never come close to being able to fulfill who then promise to buy GPUs for the data centers they can’t finish?
Yes and no.
Yes, there is a large amount of circular money on the books.
But also they’ve induced a FOMO race to buy hardware where all the big tech companies are selling off chunks and raising money/debt as fast as possible to acquire more GPUs, lest they lose out on the AI race.
The cost to train a bigger model scales exponentially with increased performance (in theory, with current architecture). So you can only beat your competitors by buying a multiples mire GPU than them.
NVIDIA is making money, they’re also inducing demand by juicing the race to try and build that FOMO.
Unfortunately these companies are paying 20x more than consumers are for the same wafers, so for the allocation of RAM to become favourable to keep consumer lever RAM production running prices will need to increase at least that much.
Yes. Ultimately even hardware sellers in this scenario need to be paid by endpoint AI subscribers whose money will filter back along the investment circle.
Or by taxpayer bailout dollars.