This is why I'm local first. Local models are getting pretty capable for most tasks. I just worry AI companies are colluding to make hardware more expensive and force everyone on the cloud, but then this stuff will happen and worse.
The effect of the current sudden burst of extreme market demand from entities with very deep pockets for (GPU, RAM, SSD storage, etc) seems to be indistinguishable from the point of view of the small end user to market collusion. Standing up something like a new DRAM fab is an incredibly expensive proposition that takes years to do.
I hope so. From what I read the difference is companies are buying HBM which is different that DDR, and the memory manufacturers only have so much output. Maybe that levels off. I also think the local models will be good enough as any other that are run on the cloud and you would only get an expensive subscription for the really demanding tasks like research or enterprise scale software.
Comments
This is why I'm local first. Local models are getting pretty capable for most tasks. I just worry AI companies are colluding to make hardware more expensive and force everyone on the cloud, but then this stuff will happen and worse.
The effect of the current sudden burst of extreme market demand from entities with very deep pockets for (GPU, RAM, SSD storage, etc) seems to be indistinguishable from the point of view of the small end user to market collusion. Standing up something like a new DRAM fab is an incredibly expensive proposition that takes years to do.
I hope so. From what I read the difference is companies are buying HBM which is different that DDR, and the memory manufacturers only have so much output. Maybe that levels off. I also think the local models will be good enough as any other that are run on the cloud and you would only get an expensive subscription for the really demanding tasks like research or enterprise scale software.