American AI Is Locked Down and Proprietary: It's Losing
American AI is locked down and proprietary. It's losing
I argue that America's closed, proprietary AI strategy is failing while China's open-weights approach is winning. By releasing models openly, Chinese companies turn compute disadvantages into distribution advantages, allowing global adoption. In contrast, US firms chase short-term profits with locked-down systems, risking the broader US economy as the performance gap closes and startups increasingly rely on Chinese models.
"Locked-down business practices for a technology with no real moat but significant potential ecosystem benefits is an obviously losing strategy; permissively releasing it with an open, collaborative approach is obviously a winning one."
HN discussion
- Open weights models require significant engineering effort for inference, system prompts, and quantization, making managed proprietary services like Elasticsearch often more efficient for production despite the lack of raw model ownership.
- Chinese government strategy involves prioritizing critical minerals and infrastructure to create a competitive market environment, contrasting with US protectionism that restricts foreign competition like BYD while subsidizing domestic incumbents.
- The shift to cloud-based AI and device enrollment models represents a regression to a 1960s mainframe timesharing revenue model, where users rent perpetual access to computing power rather than owning local hardware capabilities.
- Despite the ideological appeal of open source, company spending data indicates that frontier proprietary models dominate enterprise adoption, with open weights models largely used by organizations facing deeper financial constraints.
- China's 'open-weights' strategy is viewed by some as a geopolitical tactic to undermine US dominance, potentially leading to a future where Chinese models control global AI infrastructure if US firms fail to adapt their incentive structures.