DoorDash, Airbnb, Siemens shift to cheaper Chinese AI

The number in Futurism's writeup that stops you is this: at the businesses most committed to AI, per-employee spending now runs around $7,500 a month, according to the Ramp AI Index, and one organization reportedly blew through $500 million in a single month on Claude usage fees. That is the pressure sitting behind everything else in the piece.

The response, per the Financial Times reporting Futurism cites, is that DoorDash, Airbnb and Siemens are turning to Chinese AI tools instead of the American frontier models. Data from OpenRouter, the multi-model routing platform, reportedly shows Chinese models from DeepSeek and Z.ai have overtaken Anthropic's Claude and OpenAI's ChatGPT in usage. Eugene Cheah of Featherless AI frames the shift plainly in the article: enterprises are starting to realize they don't need the best model, they can use the faster, cheaper ones.

Some of the concrete moves are named. DoorDash cofounder Andy Fang said on X, per Futurism, that the company is saving money by pushing lower-level work to a model from the Chinese startup Moonshot AI. San Francisco startup Lindy has completely ditched Anthropic's tools in favor of DeepSeek's latest V4 models. The pull is not just price. Chinese models are typically released as open-weight, which lets enterprises mold them to their own workflows.

The honest caveat is that this is a demand-side story reported from a single mainstream feature, and OpenRouter usage is not the same as production revenue at Anthropic or OpenAI. Futurism doesn't disclose how much of the switch is real production traffic versus experimentation, or whether firms are self-hosting the open-weight Chinese models or piping data to Chinese-hosted APIs. It also notes that the Trump administration recently suspended overseas access to an Anthropic model, which adds a geopolitical tail no CFO can price cleanly.

What a leader should take from this: assume your own AI budget is about to be audited on the same logic. The upside for platform teams is real, cheaper open-weight models are a genuine lever to tier workloads by model class and to renegotiate with the US frontier labs, but only if the evals and the data-governance work get done first.