AI has been a focus within the Trump administration, often framed as a bilateral competition between the US and China. And while U.S. companies like OpenAI, Google, and Anthropic may be developing some of the most advanced AI models in the world, their prices are among the most expensive. As the costs associated with using tokens and AI rise, some consumer companies are now turning to China’s cheaper open source model.
Take DoorDash, for example. According to a post on X Wednesday by co-founder and CTO Andy Fang, the DoorDash CLI is an experimental tool in limited beta that will allow users to order DoorDash through an AI agent or directly from their device. Fan said earlier this month that using a model from Chinese startup Moonshot AI would be “better quality” and “lower cost.”
DoorDash isn’t the first to turn to Chinese AI companies, or Moonshot for that matter. AI coding startup Cursor used Moonshot’s Kim to build its Composer 2 coding agent, while fellow startup Lindy reportedly ditched Anthropic’s tools entirely in favor of DeepSeek’s V4 model. F.T..
These companies are joining the likes of Airbnb and Siemens in experimenting with moving day-to-day operations to Chinese AI companies like Alibaba and DeepSeek to save on rising AI costs.
For Yasir Attaran, deputy director and data fellow at the Futures Lab at the Center for Strategic and International Studies, this change comes down to three factors: cost, functionality, and the availability of open source models.
“What we’re seeing now is that recent high-quality, high-performance models from American companies seem to be expensive compared to Chinese-made models,” Attaran said. luck. “For some people, especially in countries outside the U.S., the idea of an open source model is much more appealing because they don’t want to share corporate data.”
As excitement around open source AI grows, companies looking for more control over their data are adopting China’s open source model. Running these models locally gives companies more control over how sensitive information is handled and reduces the need to send proprietary data to external providers.
“Instead of just hosting the closer model, it’s better to host the local model, because then everything stays on that computer and doesn’t get passed on to any company,” says Attaran. “The open source model provides that peace of mind for people who want to keep their data.”
This approach comes with tradeoffs. “We need very sophisticated computers in-house, like the ones we paid $30,000 for GPUs, RAM, storage, etc.,” he said.
Cheap at the cost of safety
Others in the industry are more skeptical. Some startups are turning to cheaper Chinese AI models to cut costs, but experts warn they may be overlooking major security risks.
Snehal Antani, co-founder and CEO of Horizon3.ai, said in a statement. luck It says startups that adopt such models “risk serious data sovereignty violations by exposing their proprietary code and user data to foreign surveillance,” while also overlooking “significant vulnerabilities in model integrity and inference.”
Still, Attaran cautioned against viewing this trend as a wholesale shift to China’s AI models. He said companies are experimenting with alternative models for different tasks rather than completely replacing the U.S. model.
“A company could try to use one of these open source models for one task and Claude for another. That’s very likely,” he said.
Few companies publicly disclose their use of Chinese AI models, but they are widely available through platforms such as code-hosting site GitHub and AI model hub Hugging Face, where developers can upload, download, and run open-source models. Hugging Face’s March 16, 2026 study found that Chinese open source models accounted for 41% of downloads.
However, lowering costs does not eliminate risk. Even as companies experiment with cheaper models, questions remain about security, data control, and how these systems will perform in high-risk use cases.
For many companies, the decision may come down to cost and capacity rather than country of origin. As Attaran suggests, if a model is “cheap, fully capable” and can be run locally, companies are likely to use it, regardless of whether it’s made in the U.S. or China.
