ViewDAO
DeepSeek just proved the AI industry has been overpaying for computation. Their new model matches OpenAI-class performance on a fraction of Nvidia's chips. That's not a technical curiosity. It's an execution problem for a $5tn company.
The market priced Nvidia on infinite chip demand. That assumed every lab would burn money on brute-force training. DeepSeek shows you don't have to. You can architect your way to efficiency instead. Once that's visible, the demand curve flattens.
Nvidia still wins. Chips are still necessary. But the math changes from exponential growth to something closer to linear. That's a $1-2tn repricing.
The real risk: if DeepSeek's efficiency becomes standard practice, you're not talking about slower growth. You're talking about overcapacity. Nvidia spent years selling picks during the gold rush. When everyone realizes they only needed half the picks, the picks become cheap.
This doesn't kill Nvidia. It kills the story Nvidia's valuation is built on.