There is such a long term history of failure here that somebody has to make a very strong case that the next time is going to be different and I've never seen anyone at Intel try that or even recognize that history of failure.
I was also going to mention some dev boards NVIDIA had come out in the last few years that were affordable (I think) but they are all out of stock now and the ones you can get are $2000+ now.
Nowadays only Chinese seem to be able to give it away to grow market share. $1 Espressif is a good example, Govin $7 GW1N-1 fpga devboards. Raspberry <$1 RP2040 is one western exception I could come up with.
It also claims to do this across radically different architectures like FPGA and all I can say to that is "I find that very hard to believe"
One common instance is, blender's Cycles renderer. Every time there is a new NVIDIA GPU. It needs to be recompiled to support it. Sometimes that also requires a new CUDA version to be able to do it. CUDA versions over the years have deprecated different operations and what not.
Although for completeness I'll note that Intel's GPU architecture is documented: https://01.org/linuxgraphics/documentation/hardware-specific...
I could see Apple and AMD, working with TSMCs latest node, stepping up to the challenge.
The fact that a similar API can be used for training on servers, inference from laptops to phones, is an appealing proposition
Best part: most devices have decent vulkan drivers. Unlike openCL.