Chinese AI Models Are Rapidly Gaining Users in the United States

American companies and developers are increasingly turning to Chinese language models instead of paying for OpenAI or Anthropic services. Data on app downloads, processed token volumes, and statements from tech executives cited in an Associated Press dispatch show this is no longer isolated cases, but a lasting trend in one of the world's most important AI markets.

Moonshot AI had to temporarily suspend new subscription sign-ups for Kimi K3 because demand outstripped the company's capacity. This isn't an isolated spike of interest in something new - Sensor Tower, cited by Associated Press, shows a similar dynamic around subsequent Chinese launches in 2026.

Sharp Rise in Downloads

Kimi K3's July launch drew the most attention. Moonshot AI's app shot to the top of download charts, and its growth in the United States was proportionally higher than globally - suggesting American users were deliberately seeking an alternative, not simply following a global trend.

A similar pattern is visible with GLM-5.2 from Z.ai, launched in June 2026. According to Vercel, which tracks model deployments, it was the fastest adoption rate among all models monitored in 2026 - and not in China, but among Western companies building products on top of the API.

The Price Advantage

Price is the main driver behind this shift. Open-source models from China cost 60 to 90 percent less than leading American offerings, according to OpenRouter data. With token consumption rising as AI agents automatically send millions of queries, the pricing gap quickly turns into real savings of hundreds of thousands of dollars a year, as in the case of one American entrepreneur who switched to Alibaba's Qwen models.

It just feels faster - Raffi Krikorian, Chief Technology Officer at Mozilla, on switching to Kimi K3

Tensions and Suspicions

The rise of Chinese models isn't happening without political friction. The Trump administration has accused Moonshot AI of using "hidden" methods in building Kimi K3 - officials and some American companies claim Chinese developers are illegally distilling knowledge from rival models overseas. Beijing calls the accusations baseless. Washington has also restricted access to its own most powerful models, such as GPT-5.6 "Sol," to about 20 vetted organizations, and in June 2026 temporarily blocked Anthropic's models globally under export regulations, before restoring them in early July.

The price differences add up - Curt Meinhold, a tech manager from North Carolina, on the cost advantage of Chinese models

Despite the tensions, some companies, including Coinbase, are switching to Chinese models mainly for cost reasons rather than ideological ones - for IT departments, what matters is the API bill, not the technology's origin. Other companies, such as Nvidia, Perplexity, and Stanford University, are for now testing models like Qwen mainly for research and comparison, not production.

What This Means for Poland

For Polish companies and developers building products on top of language model APIs, a price difference of several dozen percent in token costs is hard to ignore, especially for agentic projects that generate enormous query traffic. At the same time, growing regulatory and export pressure between the US and China means choosing a model provider is no longer purely a technical decision - it increasingly carries the risk of sudden access restrictions, as shown by July's episode involving Anthropic's models.