Signed by Nvidia, Microsoft, Meta, IBM, and over 20 other companies, Jensen Huang’s letter makes a single argument: don’t restrict open source models. The timing isn’t subtle. Just a few days earlier, the Trump administration had reportedly revived a push to ban Chinese models like Kimi and DeepSeek from U.S. enterprise use. Huang’s response was to compare the moment to the 1980s, when policymakers nearly strangled open-source software in the crib before it became the backbone of every industry on earth.
I found the framing from Jensen and the tech giants to conflict with that of regulatory bodies. Jensen frames it as an innovation question; our government frames it as a security one. Both are correct. Chinese open-weight models now account for nearly half of enterprise API tokens routed through platforms like OpenRouter (which there’s been talk of Stripe acquiring for $10B, a separate topic on its own). They’ve been downloaded, self-hosted, and fine-tuned by thousands of companies. But the enforcement problem is obvious: open weights, once downloaded, live on private servers. Restricting them starts to look less like trade policy and more like content regulation.
The gap between U.S. and Chinese models has narrowed dramatically. What was once measured in years is now speculated to be measured in months, and despite China’s chip industry still lagging 5 to 10 years behind at the individual silicon level, the implications have never been greater: controls on chips are a delaying tactic, with Chinese models becoming increasingly faster and stronger. I wonder, too, if this same idea will resonate with ex-U.S. geographies, and if they will adopt Chinese models over U.S. ones as well. The countries that learn how to govern openness without killing it will lead what comes next.