DoorDash, startups turn to cheaper Chinese AI models as U.S. rivals get pricier

DoorDash, the food delivery giant, is preparing to launch DoorDash CLI, an experimental tool in limited beta that will let users order food through an AI agent or directly from the terminal. According to DoorDash co-founder and CTO Andy Fang, the tool uses a model from Chinese startup Moonshot AI, which Fang described as offering "better quality" at a "cheaper cost." This move exemplifies a broader trend among consumer-facing companies exploring lower-cost alternatives to expensive U.S. AI providers.

DoorDash is far from alone. Cursor, an AI coding startup, used Moonshot's Kimi model to build its Composer 2 coding agent. Lindy, another startup, has reportedly abandoned Anthropic's tools entirely in favor of DeepSeek's V4 models. Larger companies are also experimenting: Airbnb and Siemens are testing Chinese AI providers including Alibaba and DeepSeek for daily operations. These moves come as token and AI usage costs continue to rise across the industry, pressuring companies to seek cost-effective alternatives to pricier U.S. models from OpenAI, Google, and Anthropic.

Yasir Atalan, deputy director and data fellow in the Futures Lab at the Center for Strategic and International Studies, identifies three core drivers: cost, capability, and the availability of open-source models. Atalan told Fortune that "the recent high-quality, high-performance models by U.S. companies seem expensive compared to Chinese models." The appeal of open-source models is particularly strong for countries and companies outside the U.S. that want to avoid sharing enterprise data with external providers. Running these models locally offers companies greater control over sensitive information and eliminates the need to transmit proprietary data to third-party services. As Atalan explained, "It's better for you to host a local model instead of just a closer model because that means everything will stay in that computer and will not go to any company. Open-source models give that sort of relief to those people who want to keep their data."

However, this approach carries significant costs. Running local models requires substantial hardware investment—Atalan noted that companies may need to spend "$30,000 for GPUs, RAM, storage, etc." Moreover, security experts warn of serious risks. Snehal Antani, co-founder and CEO of Horizon3.ai, cautioned that startups adopting Chinese models "risk severe data sovereignty violations by exposing proprietary code and user data to foreign surveillance" while also overlooking "critical vulnerabilities in model integrity and reasoning." Despite these concerns, Atalan suggests companies are not wholesale switching from U.S. models but rather experimenting—using one open-source model for certain tasks and Claude (Anthropic's model) for others.

The trend's scale is evident in the data: a March 16, 2026 study from Hugging Face found that Chinese open-source models accounted for 41% of downloads on the platform. These models are widely available through GitHub and Hugging Face, where developers can upload, download, and run open-source models. Atalan's bottom line: if a model is "cheap and capable enough" and can be run locally, businesses are likely to adopt it regardless of country of origin—suggesting that cost and capability may ultimately trump geopolitical considerations.

The shift toward cheaper Chinese AI models reflects a fundamental tension in the AI market: while U.S. companies have developed some of the world's most advanced models, their rising costs are creating an opening for alternatives. As token and usage fees increase, cost-conscious businesses are exploring options that offer acceptable performance at lower prices. The appeal is sharpened by the availability of open-source models that can run locally, addressing a genuine concern among companies that want to avoid sending proprietary data to external providers.

Yasir Atalan, deputy director of the Futures Lab at the Center for Strategic and International Studies, identifies three drivers: cost, capability, and open-source availability. The data supports the trend's real scale—a March 16, 2026 study from Hugging Face found that Chinese open-source models accounted for 41% of downloads. However, Atalan cautions against viewing this as a wholesale migration; companies are more likely experimenting by using different models for different tasks. This suggests a pragmatic division of labor: cheap, capable Chinese models for some workloads; established U.S. providers like Anthropic for others.

Yet the cost advantage comes with a significant tradeoff. Security experts raise serious concerns about data sovereignty and the risk of foreign surveillance when adopting Chinese models. The decision ultimately hinges on whether a model is "cheap and capable enough" to run locally, which means cost and performance may outweigh geopolitical concerns for many businesses—at least in the near term.

No comments yet. Be the first to share your thoughts!