DeepSeek-R1, Qwen erode US AI lead

This is less a model race than a distribution race. Once weights are downloadable, cheap, and easy to customize, the adoption flywheel can beat a closed API strategy surprisingly fast.

• –Open models win on cost, local deployment, and fine-tuning flexibility, which matters more than benchmark hype for production teams.
• –Chip restrictions slow hardware access, but they do not stop model files, forks, or community-driven improvement.
• –The embodied-AI warning is the sharpest part of the report: robotics and industrial automation reward whoever accumulates real-world deployment data first.
• –If U.S. labs want to counter this trend, they probably need more open weights and ecosystem gravity, not just better frontier demos.