It is so constrained that companies are literally buying smaller GPU cloud companies.

So, if by “hysterical,” you mean it is pretty funny and a reflection of the hype cycle, then yes. If you mean that I am being overwrought, I can assure you I am not. Even the hyperscalers do not have enough GPUs

AMD released ROCm (its competitor to CUDA) in 2016, nearly 9 nears after NVidia released CUDA. They relied on OpenCL and failed to invest in "GPGPU", and as a result were so far behind NVidia they couldn't keep up. As a result, for about a decade most scientific GPU code was written in CUDA.

Today, AMD support in PyTorch is minimal. Actually getting anything running is very difficult, and random crashes are common. This is in contrast to NVidia, which spends a lot of money to ensure a full compiler stack and compatibility with AI libraries.

Today, the AMD hardware itself is pretty capable and has a good price/performance ratio. However, actually taking advantage of that performance is difficult because of the poor quality of drivers and software.

They're not really playing ball, NVIDIA did the right thing in pushing software support early on. CUDA really has a good strong hold and AMD isn't doing much by way of pushing code support for their CU's.

It's going to take a big long investment, which people have been arguing about for the past 6 years, and AMD really isn't jumping up take the mantle. It's really a shame too because we need a strong competitor if we ever expect more realistic pricing for the average users/company.

Essentially all Nvidia gpus support cuda, even cheap ones. The amd rocm supports only like 10 very specific models.

AMD exposed an OpenCL API for GPGPU and hoped the open source community would do the rest. That approach failed.

ATI was acquired by AMD in 2006. :)