NVIDIA unveiled the Vera Rubin platform today: 3.5 times the training performance of Blackwell, 5 times the inference performance, inference cost at one-seventh of Blackwell. It's an extraordinary product roadmap. The stock is reacting accordingly.

What I want to add is the other side of this: on the same day Jensen Huang is presenting the most compelling AI hardware roadmap in history, the US government quietly closed the loophole that allowed NVIDIA and AMD to export AI chips to Chinese companies' overseas entities. These two events are happening simultaneously, and the market is only paying attention to one of them.

The overseas entity closure matters for a simple reason: it signals the direction of travel on export policy. Each restriction is followed by another. The China addressable market for NVIDIA's AI products keeps shrinking incrementally. Over a multi-year period, that erosion adds up. The Rubin platform will be subject to these same controls, which means the revenue growth trajectory depends entirely on what happens in the US, Europe, and the rest of Asia.

I am not saying sell NVDA. The AI infrastructure demand cycle outside China is real and large. What I am saying is that a full-position bet on NVDA at elevated multiples, with policy risk that keeps compounding in one direction, is a portfolio management question as much as a stock analysis question. The best companies can still be poor risk-adjusted buys at the wrong sizing.