Companies Are Choosing Chinese AI Models to Cut Costs

Companies are overwhelmingly choosing Chinese AI models (e.g. DeepSeek, Moonshot AI Kimi, Qwen, and Z.ai) because they are 60-90% cheaper than US alternatives, are open, and help reduce growing business costs and dependency on OpenAI or Anthropic.

The main reasons why businesses are choosing Chinese solutions are:

• Huge price difference: US service providers are raising prices, while Chinese labs (often state-backed) offer extremely cheap “token” prices or open-source models. For example, DeepSeek V4 costs about $0.14 per million tokens, compared to US alternatives, where the same volume is a few to a dozen dollars.

• Open-weight: Many Chinese AI models allow companies full visibility into their parameters. This allows the model to be tailored to specific business needs and ensures greater security when handling sensitive company data on-site.

• Technological breakthrough: The latest models from China, according to tests conducted by the Artificial Analysis platform, show results that are very close to the most advanced US systems. This allows companies to effectively transfer everyday and simpler tasks (such as document processing) to Chinese models.

Business practices in the market:

Companies such as DoorDash, Siemens and Airbnb have been using Chinese-developed tools for some time. For example, DoorDash reported that the Moonshot AI models help save a lot of money. Meanwhile, US startup Lindy completely transferred some of its processes from Anthropic to DeepSeek models, hoping to save millions.

Risks and challenges:

Despite its huge popularity, concerns are rising in Western countries about data security, influence from the Chinese government and sanctions imposed by the US. Some US government agencies and lawmakers are even investigating the extent of Chinese AI models being used in the US tech sector.

Even the Lithuanian press, which is mostly engaged in propaganda, acknowledges the benefits China provides:

 

“Companies from Silicon Valley to Europe are increasingly choosing Chinese-made artificial intelligence (AI) models, seeking to reduce the cost of using the technology and their dependence on the most advanced developers in the US.

 

Among the companies that have started using Chinese-made AI tools are DoorDash, Siemens and Airbnb. They are attracted by cheaper, increasingly powerful models that are, in some cases, easier to use on their own infrastructure, the Financial Times reports.

 

According to OpenRouter, a platform that tracks statistics on the amount of text, code and data processed by large language models, AI models from Chinese companies such as DeepSeek and Z.ai have rapidly overtaken their US competitors in terms of consumption of text units (tokens) this year.

 

The shift is largely driven by cost, as companies seek to rein in the rapidly rising costs of AI. But in Europe, the process has also taken on a clear geopolitical dimension after the Trump administration imposed export restrictions on Anthropic’s Mythos and Fable models last month, forcing companies to reassess the risks of relying on U.S. technology.

 

“The Chinese models are the elephant in the room,” said Eugene Cheah, CEO and co-founder of Featherless AI. “Companies are starting to realize that we don’t necessarily need the best model—we can use faster and cheaper ones.”

 

The shift has also been accelerated by the fact that U.S. AI companies, including Anthropic and OpenAI, have shifted some of their services for enterprise customers from flat-rate subscriptions to pay-as-you-go models, which has significantly increased the cost of using their models.

 

Another advantage is that many of China’s most advanced AI models are so-called open-weight models, with parameters publicly available. This allows them to be deployed on company-managed servers and customized to specific needs, while preserving business ideas and trade secrets.”