- Ben Thompson argues US frontier labs have lower unit costs and tool stickiness, but the author disagrees.
- Unit costs: US labs have higher electricity, data center costs, and may lose advantage as Chinese chips improve.
- Stickiness: AI tools like Claude Code are quickly replaced; users switch easily for better alternatives.
- US labs face structural issues: no secret sauce, high energy costs in a constrained grid, and overreliance on future profits.
- Predictions: US labs won't save by costs/stickiness, will lose frontier status, and China will lead global inference.