Chinese open-weight models are winning with a corporate calculator, not a geopolitical slogan

Futurism, citing the Financial Times, says US and European companies are increasingly turning to cheaper Chinese AI models. It names DoorDash, Airbnb and Siemens, points to interest in open-weight models and cites OpenRouter data suggesting models from DeepSeek and Z.ai have overtaken US alternatives in some usage.

Companies are moving cheaper tasks to Chinese models

The article cites DoorDash cofounder Andy Fang saying the company saves money by sending lower-level work to a model from Moonshot AI. Futurism also says startup Lindy has fully ditched Anthropic tools in favor of DeepSeek V4.

The driver is the bill. Futurism mentions an extreme case of one organization spending $500 million in a month on Claude usage fees, plus Ramp AI Index research saying the heaviest business users spend about $7,500 per employee per month on AI.

Multi-model routing is shifting from optimization to necessity

This is not an ideological turn toward China. It is an operational response to agentic costs. When a company runs multiple agents in parallel, even a small token price difference compounds across thousands of runs.

Open-weight models also promise more room for customization and control. That does not automatically make deployment safer, but for teams that want to tune models around their own processes, it creates a different bargaining position than a closed API.

A cheaper model can still be expensive in the wrong failure mode

Cost cannot override risk. Moving work to cheaper models makes sense for lower-risk tasks, summaries, classification and helper steps. Sensitive data, compliance and customer-impacting decisions still depend on governance, logging and the ability to shut a model off.

It is also worth separating open-weight from fully open development. Visible weights are not the same as complete training data, a reproducible process or clear accountability for model behavior.

The winner will route work by risk, not brand loyalty

The useful signal is less national rivalry and more enterprise routing: which tasks go to a frontier model, which go to a cheap open-weight model and where a human stays in the approval loop. If this becomes standard, US models will have to justify premium pricing with specific accuracy or safety advantages.

Lilith's verdict

Corporate AI no longer drinks from one golden tap. The CFO is standing over the pipes with a wrench, sending expensive models only where the cheaper stream cannot hold.