Chinese AI models now dominate OpenRouter, a major routing platform, capturing over 60% of traffic in July.
American companies are switching to Chinese models for cost reasons, not coercion: DeepSeek costs one-twelfth what GPT-5.5 costs at comparable performance.
The article warns that most corporate AI strategies are stuck between expensive frontier models and cheap commodity alternatives, and that without open-weight competition from US labs, American firms risk losing the ecosystem to China.
What happened
Chinese-developed models claimed all five top positions on OpenRouter in July, with Xiaomi's MiMo V2.5 ranked first by token volume. Chinese models now carry more than 60% of the platform's traffic, which exceeds 20 trillion tokens a week. By mid-July, Chinese models accounted for 58% of tokens processed by American firms on the platform—a shift from roughly 70% US model traffic a year ago.
Why it matters
US companies are choosing Chinese AI workload-by-workload because of cost, not compulsion. DeepSeek V4 Flash costs $0.14 per million input tokens versus $5.00 for GPT-5.5; Chinese open models run 60% to 90% cheaper than leading American offerings. Meanwhile, Alibaba's Qwen family has passed one billion cumulative downloads and replaced Meta's Llama as the most-downloaded open model family—a shift that locks developers into an ecosystem. Most Fortune 500 AI strategies are stuck in a "death zone": neither frontier-class nor cheap enough to compete.
What to watch
The article argues American labs must release frontier-class open-weight models on a regular schedule to compete. Without credible US open-weight alternatives, the risk is that "when the next generation of global software is built, [it will be] built on" Chinese models. Companies deploying hybrid routing—frontier models for high-stakes work, efficient open models for high-volume tasks—are cutting inference costs 60% to 90% on the majority of workloads.
Ask the AI about this article →
The article identifies a structural shift in AI adoption driven by economics rather than capability. While American labs (OpenAI, Anthropic, Google) still hold the frontier on the hardest reasoning tasks—GPT 5.5, Claude Fable 5, and Gemini 3.x lead on demanding enterprise work—the deployment market has bifurcated. Chinese labs, constrained by export controls on large GPU clusters, engineered around scarcity with token efficiency and novel architectures from the start. The cost gap is now decisive: American firms face a choice between paying premium prices for frontier capabilities they often do not need, or adopting cheaper alternatives. The author notes that Anthropic captures roughly half of total OpenRouter spending despite holding only 12% of token share—the "premium lane." Everything else is being crushed into either the commodity floor (cheap, open models) or disappearing from the middle.
The ecosystem lock-in effect mirrors historical platform shifts: developers optimize for what they can download and build tooling around what they deploy, just as Linux won servers and Android won phones. Alibaba's Qwen family passing one billion cumulative downloads signals that Chinese models are becoming the default foundation for developer projects globally. The article argues this outcome was not accidental: state support lowered effective costs, Xiaomi cut API prices by up to 99% in May, and the labs prioritized inference efficiency from day one. For American strategy, the implication is stark—the race is no longer just about frontier capability but about who owns the distribution and ecosystem.
For example, today's edition would include:
AI-summarized, only the topics you pick — one digest a day via Email, Slack, or Discord.
Free · takes 30 seconds · unsubscribe anytimeWhat is AIToday? →
Ask AI anything about this article. Q&As are published on this page for other readers too.
AI system scaling has pushed interconnect requirements inside data centers from chips and boards up to racks…
Chinese large-model developer Z.ai says it can now support large-scale inference using roughly 100,000 domesti…
Analyst Ming-Chi Kuo says Nvidia has revived the Rubin CPX AI accelerator with a substantially redesigned arch…
Palantir Technologies stock has posted multi-year gains, including an 11x return over 3 years
Apple has escalated its legal battle against OpenAI, claiming in a new court filing that OpenAI is actively de…
Samsung Electronics has locked up as much as 70% of its memory production capacity under long-term supply agre…
