1: Fist of all, "GPU" vs "CPU" definition is a bit misleading. Nivida started as a graphics card company and that is why people even today refers this as GPU. But the more accurate definition should be "AI chip" as it really about the ability for large data set parallel processing, very little do to with "graphics".
From this perspective, the AI driven infrastructure is dominated by the need for GPU, rather than CPU. Yes, every data center server still need a CPU in it, but the $$ is going to GPU for its data processing capability.
2: This means the revenue driver, and more importantly, margin, are all going into GPU. A CPU may cost $8K to $10K in a datacenter server, but a H100 GPU could cost $80K-100K. I am grossing simplying this but you get the picture.
3: Today, nVidia has commanding share 90% of the GPU market. Every friend in the industry I speak to, everyone is rushing to buy nVidia GPU. 50%-70% of every hyperscaler's Capex today goes to nVidia.
4: So for AMD, it does not matter how good their CPU is, the problem is that they are not able to compete with nVidia on GPU. One apsect of this is performance and nVidia's relentless push under leadership of Jensen (nVidia is not Google or Boeing who has been ruined by bad leadership). While AMD has shown progress in its hardware capability and performance, a more difficult hurdle for them to overcome is CUDA. NVDA has "standarized" how to develop app on GPU via CUDA. AMD may have a comparable GPU from performance aspect, but company still would not buy AMD GPU due to CUDA.
5: This is not to say that AMD has no chance. The future of AI will shift from data center to edge, where smartphone, cars, robots, industry automation, every day smart home devices, will all have AI chips optimized for these edge use cases. It is possible that Nvidia is not a dominate player in those market and AMD could grab a significant pie. However, as of today, there is no clear sign that AMD is that winner. We actually do not know who that winner is. It could be AMD, it could be some other new chip company, or it could contine to be NVDA
Now back to investing, there is no reason to consider AMD if you do not yet have NVDA, because NVDA is the undisputed leader today and selling at a discount. But if you already have NVDA estbliashed, and want to allocate a small position in AMD just to try to luck, that is probably ok
