The Coinbase CEO suggested experimenting with cheaper open-weight AI models to keep AI spending in check as token consumption rises.
This proposal has led to concerns about the security and geopolitical risks of routing enterprise workloads through Chinese-origin systems.
Why are companies using Chinese AI models?
US export controls have made it difficult for Chinese companies to access US AI chips, but that has not stopped them from building competitive models and selling them at much lower prices.
For example, Zhipu’s GLM 5.2 costs $1.40 per million input codes and $4.40 per million output codes Compared to Anthropic’s Opus 4.8 Priced at $5 and $25 for the same size.
GLM 5.2 scored 62.1 on SWE-bench Pro, a major programming benchmark, beating OpenAI’s GPT-5.5 at 58.6. One AI researcher said that GLM 5.2 is “at least as good as Opus 4.8 and GPT 5.5.”
Another called it “the first open model that can truly compete with closed source systems.”
Does Coinbase use Chinese AI models?
Coinbase says CEO Brian Armstrong The best way to control rising AI costs is to use cheaper open-weight models, including systems from China such as GLM 5.2.
Instead of spending more and more on AI, companies need “better defaults, routing and caching,” Armstrong said. His suggestion of using Chinese models, even if they are cheaper, raised concerns about security and political risks.
In addition to its convenient pricing, GLM 5.2 uses an MIT license, which means companies can download, modify and run it on their own servers, removing any risks of sending sensitive company data to an external API.
Spending on AI has become a real problem, causing companies to retreat from using the technology in operations.
Cryptopolitan I mentioned recently Uber has exhausted its entire 2026 AI coding budget by April, and is now capping engineers at $1,500 per tool each month. Meta sent a note Warning about the “exponential increase” in the use of artificial intelligence and began setting spending controls. Amazon eliminated an internal leaderboard that ranked employees by AI consumption because people were manipulating it and driving up costs.
A KPMG survey found that only 26% of companies have full visibility into their AI costs, while 22% only discover spending after receiving the invoice. Goldman Sachs predicts that consumption of AI tokens could increase 24-fold by 2030, reaching 120 quadrillion tokens per month.
IDC predicts that 70% of leading organizations relying on AI will use multiple models by 2028 rather than relying on a single provider.
What makes Chinese AI models so risky?
Z.ai’s cloud API, which allows developers and companies to use its AI models (including GLM 5.2), falls under China’s National Intelligence Law. This raises real concerns for any company dealing with sensitive information.
US lawmakers opened a formal investigation in May into cybersecurity risks posed by Chinese-origin artificial intelligence models in critical infrastructure.
There are also concerns that models trained under different legal systems could exhibit undetected behaviors. Additionally, one AI builder tested GLM 5.2 against GPT-5.5 in a debugging task and found that it was “not even close” to the OpenAI model’s ability to detect issues, despite reports that Chinese models outperform their more expensive counterparts.
Anthropic revealed in an open letter to the Senate Banking Committee that Alibaba Qwen operators ran 28.8 million Claude exchanges through about 25,000 fake accounts between April and June. They called it the largest known campaign to steal model abilities.
Self-hosting Open Weights eliminates the risk of routing API data, as companies running the model on their own servers do not send the data to China. But concern about the models themselves remains.
