As you stated, NVDA is trading at ~20x 2027E and AMD at ~45x 2027E. Could part of this NVDA discount be due to potential future margin compression for NVDA because of AMD? Similarly, could AMD be trading richer than NVDA because of expectation of continued margin expansion as it trends towards NVDA's margin profile?
This is a super popular narrative, but Tom Brown (Anthropic chief compute officer) was actually quoted saying that Anthropic is accelerating Claude with AMD MI455x GPUs used in both training and inference
This is true - all roads and rockets lead to Nvidia. I remember my Voodoo 3D and then my Nvidia GF2 MX after.
Unfortunately for AMD's current valuation, it is out of bounds due to other competitors that all go to the same suppliers. That's where Nvidia has bought up as much capacity as possible.
As you stated, NVDA is trading at ~20x 2027E and AMD at ~45x 2027E. Could part of this NVDA discount be due to potential future margin compression for NVDA because of AMD? Similarly, could AMD be trading richer than NVDA because of expectation of continued margin expansion as it trends towards NVDA's margin profile?
I’ll be writing more about that soon. Training is mostly NVIDIA’s at this point. Inference is the battleground
Any thoughts on Cerebras as a key player in the Inference space. Its latest deals with AWS, OpenAI and AMD in the Inference space look interesting.
They can play but won’t survive as a standalone. It’s a rich man’s game. Someone will absorb them
This is a super popular narrative, but Tom Brown (Anthropic chief compute officer) was actually quoted saying that Anthropic is accelerating Claude with AMD MI455x GPUs used in both training and inference
This is true - all roads and rockets lead to Nvidia. I remember my Voodoo 3D and then my Nvidia GF2 MX after.
Unfortunately for AMD's current valuation, it is out of bounds due to other competitors that all go to the same suppliers. That's where Nvidia has bought up as much capacity as possible.
Great work Ben!