How Competitive will AMD be against Nvidia in AI Accelerator Market

I've been delving into the specifics of AMD's New MI 300 platform and it appears to stack up favorably against the H100, often surpassing it in several aspects. However, there's an ongoing discussion about ROCm lagging behind CUDA, though AMD is vigorously working to bridge this software gap. Despite not being a semiconductor novice, I find myself drawn to AMD's chiplet-based design over Nvidia's monolithic approach. As we move toward smaller die sizes with High NA EUV, expertise in chiplet-based design will prove crucial, yielding numerous advantages. Additionally, AMD's ability to integrate different process nodes within the same package adds to cost efficiency.

This discussion is pertinent due to the striking disparity in revenue estimates between AMD and Nvidia. The projections suggest Nvidia might achieve over six times the Accelerator revenue in 2027. It's worth noting that the actual figure could be even higher, considering that Data Center revenue encompasses enterprise CPU sales as well. Does anyone have any thoughts? $Advanced Micro Devices (AMD.US)$ $NVIDIA (NVDA.US)$