The real AI race may no longer be at the frontier

The competitive focus in AI development is shifting away from cutting-edge frontier models toward open-source alternatives that are cheaper and customizable. Chinese open-weight models have become dominant on major developer platforms, accounting for a significant share of downloads and popular deployments, while frontier models from leading U.S. companies are increasingly used only for specialized tasks. This shift matters because companies are recognizing the costs and risks of relying on proprietary models controlled by a single provider, prompting them to build their own private models or adopt open alternatives instead. The debate reflects broader disagreements about safety, with some arguing that open models pose dangers that should be restricted, while others contend that concentrating powerful AI in few companies creates greater risks through asymmetry of power and reduced transparency.