Does AMD really steal NVDA’s cake? I’ve been looking at the data for a week

I hadn’t bought AMD before, but I’ve been keeping an eye on it. Recently, AMD has been making a lot of moves. In early May, it released the MI350P PCIe GPU accelerator card, targeting enterprise AI inference. This product takes a plug-and-play approach. Its goal is to prevent enterprises from having to replace their data center architecture—deploy it directly using existing PCIe servers.

That idea makes sense. Many small and midsize businesses don’t have the resources to rebuild a dedicated AI data center. If the MI350P can solve that pain point, it can indeed take a big bite. AMD’s product roadmap also says it plans to release the MI400 series in 2026 and its first rack-level solution, Helios—moving very fast.

But my reason for not buying AMD hasn’t changed. NVIDIA isn’t just hardware—it’s a combination of hardware, software, and an ecosystem. The CUDA ecosystem has been used for so many years, and developer lock-in is simply too strong. For AMD to catch up, it may need better hardware, which could take five or even ten years. As a small retail investor, I can’t afford that kind of time cost.

Liquidity for AMD on Gate is also good, with a very small bid-ask spread. If I change my mind one day, I can buy anytime. But for now, I’ll keep holding NVDA, using it to capture the benefits of the growth in the entire AI infrastructure.

#分享美股交易赢英伟达股票