This is a much more accurate statement than the hate on the TPM. As the article describes, it is the GPU that has its own separate memory space that it can show on the screen without the CPU being involved at all.
I expect next generation workarounds will involve virtual GPUs.
If that worked it'd have been done over a decade ago.
The remote server is handshaking cryptographically with the GPU itself, which identifies itself using certificates and keys tied at the factory. You can't emulate such a GPU unless you find a way to steal the keys.
I expect next generation workarounds will involve virtual GPUs.