✨ Begun, the US vs. China AI race has

How should America respond to DeepSeek?

OK, we've seen the initial, first-order impact of DeepSeek's whizzy new AI model: a stock market freak-out. The Chinese startup seemingly shows that you can build solid AI models without throwing endless chips at the problem — which has technology investors rethinking just how much companies need to spend on AI hardware.

A few headlines:

• Stocks Hit by AI Jitters as Traders Rush for Haven - Bloomberg

• Advances by China’s DeepSeek sow doubts about AI spending - FT

• Stocks Sink as Investors Worry About China’s A.I. Advances - NYT

To recap: DeepSeek's R1 achieves its impressive performance by eschewing traditional fine-tuning in favor of pure reinforcement learning (a training method where the AI learns through trial and error with rewards for good performance, thus skipping the compute-intensive process of analyzing massive datasets) rather than supervised training. This approach lets the model develop its own reasoning abilities, though some fine-tuning training is later added.

DeepSeek’s R1 has the entire tech world talking, with Marc Andreessen calling it “AI's Sputnik moment” — perhaps a sign that the assumed rules of the game might be changing. As one market strategist told the Financial Times, “It shows how vulnerable the AI trade still is, like every trade that is consensus and based on the assumption of an unassailable lead.”

Now this isn’t an investment newsletter for AI and chip stock day-traders. I’m not going to speculate on which technology companies stand to be the big winners and losers as each chapter of this Age of AI unfolds. But I do have three insights right now that I hope readers find helpful.