"...AMD has to innovate here with their next-gen, or they'll fall behind too much and NVIDIA will win ray tracing."
Yeah...obviously...the problem is that Nvidia has a multi-year head start and a much, much, much larger R&D budget. Nvidia's 2021 R&D budget is $5.27 Billion while AMD's is only $2 billion. Furthermore, Nvidia is able to spend that $5+ billion basically entirely in the area of graphics....perhaps a fraction goes to ARM CPU development, but the vast majority goes to graphics. AMD on the other hand has to split that $2 billion between x86 and graphics and because x86 makes up the majority of AMD's revenue and has a larger T.A.M. than graphics, we can safely assume that more than half of AMD's R&D budget goes to x86. So in the end, Nvidia has probably $4+ billion for GPU R&D while AMD has less than $1 billion for the same application.
Considering Nvidia is easily spending four times as much on GPU research and development as AMD, I think it's extremely impressive that AMD was able to basically Match Nvidia in raster on RDNA2 and make such a strong showing with Ray tracing and FSR 2.0. That said, AMD has been able to accomplish more with less than Nvidia or Intel, but I wouldn't expect them, especially considering Nvidia's head start, to match Nvidia on raw raytracing performance with RDNA3, though they should close the gap.