https://www.theregister.com/2024/08/09/amd_zluda_take_down/
We guess the House of Zen wants all you HIP kids to ROCm out with its own runtimes instead
Nvidia may be using an EULA to try and make people not use a translation layer, but if the EULA doesn't apply or the consequences of breaking it don't prevent you continuing then what Nvidia wants means diddly.
I don't use CUDA or Nvidia so I don't know but Google release Android Studio and have an EULA saying you can't do bla bla bla. But Android Studio is open source so if I don't use their binary and compile it myself then (as far as I know) their EULA doesn't apply (only the open source license used before they added an EULA on top of it for distribution).
An EULA is an End User License Agreement. It has no legal authority over a customer who does not even use an nvidia product, let alone a company.
Perhaps not even when you use an Nvidia product like if I buy Nvidia hardware but don't use their software (i.e. use open source drivers instead). I don't know enough about CUDA to say if you're not using Nvidia software (normally, the topic discusses a reverse-engineered one which doesn't infringe on Nvidia's copyright of their software).