As companies' spending on artificial intelligence surges like an unbridled wild horse, many Western giants are also starting to ask themselves a simple question: Since Chinese AI models are much cheaper, why spend a lot of money on purchasing "leading" US AI models?
According to a report by the Financial Times on July 13th, large corporations such as DoorDash, the largest food delivery giant in the United States, Airbnb, a global travel home rental platform, and Siemens, a German technology giant, are turning to Chinese artificial intelligence models. What attracts them is not only the lower costs but also the 'open-weight' model of these models, which allows companies to customize the models according to their specific needs.
According to data from the OpenRouter platform, Chinese leading models such as DeepSeek and Z.ai have surpassed American counterparts like Anthropic's Claude and OpenAI's ChatGPT. OpenRouter is a platform that provides one-stop access to various AI models and tracks their usage. Cost reduction seems to have overpowered all geopolitical competitions.
China’s leading models have surpassed similar products from the United States, FT.
"AI platform Featherless' CEO Eugene Cheah told FT that China models are “the elephant in the room.” "Companies have started to realize, 'Hey, we don't need the best model, we can use faster and cheaper ones.'"
Reports say that American AI models have long been considered the most advanced, but this perception is changing. The GLM-5.2 model released by Chinese startup Z.ai last month caused a sensation in the Western tech community. Several important figures in Silicon Valley praised its capabilities, stating that it is comparable to or almost comparable to American systems, while the cost of using it is much lower.
Chinese models are both high-quality and affordable.
Affordable and high-quality Chinese AI is coming at the right time, as many companies are being discouraged by high costs. Andy Fang, co-founder of DoorDash, said on the X platform last week that the food delivery group is now assigning "lower-level tasks" to the Kimi K2.6 model developed by Chinese startup Moonshot AI, while retaining Anthropic's Fable only for "the most difficult tasks." He stated that the new combination is more cost-effective than using only Anthropic's advanced models in the US.
Siemens of Germany also stated that it hopes its AI models to be "flexible". The range of models it uses includes tools from Chinese companies like DeepSeek and Z.ai, as well as models from advanced laboratories in the United States, NVIDIA, and the French AI company Mistral.
Some companies have gone even further, completely switching to the Chinese model. The startup based in San Francisco, Lindy, has switched from Anthropic’s AI tools to DeepSeek’s V4 model. Founder Flo Crivello praised this change as "transformative" on the X platform last month, stating that it saved the company millions of dollars and improved performance in "many core use cases."
By having China's early-stage startup companies Moonshot AI delegate "lower-level tasks" to their models, the company saved a considerable amount of funds.
George City University's Center for Security and Emerging Technologies researcher Sam Bresnick said, "Companies are motivated to shift some of their workloads to cheaper models. Since Chinese models can basically be used for many of the workloads you need, why pay a premium for Anthropic or OpenAI?"
But cost is not the only consideration: many Chinese models have an ‘open weight’ setting, which means their parameters or values are completely visible to users. This allows organizations to shape the models according to their specific needs—from a cybersecurity perspective, it also enables better control and understanding of how the models handle sensitive company data.
Airbnb stated that it uses a "limited number of Chinese source models," and operates these models only through approved American service providers, thereby protecting its data and operations.
While proprietary models like OpenAI's ChatGPT and Anthropic's Claude are primarily accessed through their creators' systems or third-party platforms.
For foreign companies that are disappointed with American leadership, the choice becomes even easier. Especially after the Trump administration suspended the Anthropic’s Mythos model from being used overseas, European trust in America as a leader in AI further decreased.
Bullhound Capital’s founder, Per Roman, said, “Two years ago, the main concern was China. Now, Europe has a bigger concern: the United States. This is shocking.”
