The AI ‘death zone’ is here and most corporate AI strategies are standing in it
Chinese models just swept the top five spots on the world's largest AI marketplace. American labs still own the frontier. Don't get caught in between.
In July, Chinese-developed models took all five top positions on OpenRouter, a platform closely monitored by the AI industry. Xiaomi's MiMo V2.5 ranked first by token volume, followed by models from DeepSeek, MiniMax, Alibaba's Qwen family and Moonshot's Kimi. Chinese models now carry over 60% of the platform's traffic, which exceeds 20 trillion tokens per week.
US models carry about 30% of OpenRouter's traffic, down from roughly 70% a year ago. Chinese models also accounted for 58% of tokens processed by American firms on the platform in mid-July.
The race between American and Chinese AI labs has split into two contests: capability and distribution. American labs lead in capability with GPT 5.5, Claude Fable 5, and Gemini 3.x, excelling at reasoning, long-horizon agents, and demanding enterprise work. However, they are losing in distribution. Chinese open models run 60% to 90% cheaper than leading American offerings, with DeepSeek V4-Pro priced at roughly one-twelfth the cost of GPT-5.5 for comparable benchmark performance.
The death zone is the area between the frontier models and the cheapest open models. It's filled with models and corporate AI strategies that are neither clearly the best nor clearly the cheapest. Most Fortune 500 AI strategies are currently standing in the middle, purchasing frontier API contracts and routing all workloads through them. This approach is increasingly costly and inefficient, as Chinese open models offer superior performance at a fraction of the price.
To navigate this landscape, executives should adopt hybrid routing, prioritize efficiency, differentiate above the model layer, and avoid remaining in the middle. Hybrid routing involves routing the most critical workloads to frontier models while using efficient open models for high-volume, cost-sensitive tasks. Treating efficiency as a first-class weapon includes optimizing inference, quantization, speculative decoding, and model hardware co-design.
Differentiating above the model layer involves leveraging proprietary data, application layer fine tuning, domain-specific customization, and rigorous evaluation to create a competitive moat. Finally, executives should avoid remaining in the middle and choose a direction this year, either by differentiating with real capabilities or competing on cost and openness.
Written by urgent.news from Fortune's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.