$Advanced Micro Devices(AMD)$ AMD is unquestionably narrowing the performance gap – but it is not yet catching up in the strict economic sense, because Nvidia still owns the entire “stack premium” (CUDA software ecosystem, developer mindshare, and the early network effects) that buyers treat as strategic, not merely technical.
Competitive positioning
AMD has proven that it can compete at the silicon level. MI300 is finally gaining real traction, and hyperscalers are now buying AMD – not to diversify for fun, but because they must, to avoid vendor lock-in into Nvidia and to lower cost per token / cost per watt.
However:
Nvidia still controls the software meta-layer (CUDA, TensorRT, cuDNN)
Nvidia still has the best vertical integration to cash conversion
Nvidia gets design wins earlier in the AI cycle
So – AMD is doing very well relative to its historical base, but Nvidia remains the default standard.
Can AMD hit US$300 in 2025?
It is possible – but that would require two conditions to break AMD’s way:
1. hyperscaler adoption continues to compound quarter-on-quarter
2. the Street stops treating AMD as a “second derivative on NVDA’s cycle” and rewards AMD with a higher P/S band
Right now the market still prices AMD as “Nvidia-lite with slower margins”. To re-rate to US$300, the narrative must shift to:
> AMD is not the alternative to Nvidia – it is a co-line leader in a market now measured in trillions.
That psychological transition – not EPS alone – is what will push the stock into the 300 handle.
Biggest highlight to watch at Analyst Day
The most critical single datapoint:
the granularity of the AI TAM upgrade – and how much is attributed specifically to Data Centre inference vs training.
If Lisa Su explicitly signals that inference is expected to become the dominant profit pool and AMD intends to remain pricing-aggressive there – that would be the major unlock.
Because inference is where Nvidia is relatively more vulnerable.
Training is Nvidia’s fortress.
Inference is where AMD can actually take share faster.
Comments