Chinese AI Models Dominate Usage Rankings, Pressuring U.S. Labs

The competitive landscape for artificial intelligence is shifting beneath Silicon Valley's feet. Chinese AI models now dominate the top usage rankings on OpenRouter, a major marketplace where developers access competing AI systems, signaling a fundamental change in how businesses approach AI deployment.

All five of the most-used models by weekly token volume come from Chinese companies—Tencent, Xiaomi, DeepSeek, MiniMax, and Z.ai—according to details first reported by Axios. Each system is released as "open-weight," meaning organizations can download, modify, and operate them on their own infrastructure without ongoing licensing fees.

Why it matters

This usage pattern reveals that raw model capability matters less than cost and control for most enterprise AI work. If the majority of business tasks can be handled by inexpensive, customizable systems, the economic foundation supporting multibillion-dollar valuations of frontier labs becomes questionable—with implications extending to stock market stability and U.S. economic growth projections.

The economics driving adoption

Most corporate AI applications don't require cutting-edge intelligence. Routine coding assistance, document summarization, data extraction, and customer service can run effectively on less expensive systems. One AI investor told Axios that open-source models will eventually handle 95% of enterprise queries, leaving only the most complex 5% for premium providers like OpenAI or Anthropic.

Kong CEO Augusto Marietti reported that open-weight adoption has surged over the past quarter because flagship models cost too much. Mozilla CTO Raffi Krikorian compared deploying frontier AI for everyday tasks to "driving a Ferrari to Whole Foods"—cheaper models deliver adequate speed and capability at up to 50 times lower cost.

Closing capability gaps

The cost advantage would matter less if Chinese systems lagged significantly in performance. But that gap is narrowing rapidly. In May, Anthropic CEO Dario Amodei assessed that China remained six to 12 months behind the United States in the most dangerous cyber capabilities. Ten weeks later, Moonshot released Kimi K3, a model that matches Anthropic's Fable and OpenAI's GPT-5.6 on key benchmarks.

American responses

Some U.S. companies are pivoting toward open approaches. Thinking Machines, founded by former OpenAI CTO Mira Murati, launched this week with an open-weight model designed for deep customization. Nvidia is expanding its Nemotron family of open models, calculating that customizable AI will increase demand for its chips and software. SpaceXAI open-sourced Grok Build, the software powering its coding agent, extending openness beyond models to tooling.

The valuation question

OpenAI and Anthropic are preparing for public offerings premised on frontier AI remaining scarce and commanding premium prices. If businesses obtain most needed capabilities from systems they control at fraction-of-the-cost, those valuations face pressure. The stakes extend beyond individual companies—AI spending currently carries an outsized share of U.S. growth, and equity markets have concentrated exposure to a small group of AI boom beneficiaries.

"They're clearly terrified," Krikorian said of U.S. labs confronting the rapid rise of Chinese competitors.

These details were first reported by Axios.