> I really fail to see what people see even in Nvidia/Cuda, it won't take long until others catch up, and they'll likely move back to margins closer to the industry, which are extremely low
I thought that would happen by now too, in 2016, so I started slowly selling much of my NVDA stake after a few years. This was wrong. It turns out they have created an ecosystem effect where everyone makes their papers and code work in CUDA and everything else is best effort. AMD hasn’t even really been seriously trying from my point of view, but once they do they’re still trying to bail water with a leaky bucket - new models come out every day where the researchers have run and tested on CUDA only.
I thought that would happen by now too, in 2016, so I started slowly selling much of my NVDA stake after a few years. This was wrong. It turns out they have created an ecosystem effect where everyone makes their papers and code work in CUDA and everything else is best effort. AMD hasn’t even really been seriously trying from my point of view, but once they do they’re still trying to bail water with a leaky bucket - new models come out every day where the researchers have run and tested on CUDA only.