In recent weeks, Silicon Valley has been involved in an AI model released by a lab in China. The scheme, called GLM-5.2, is named “wonderful,” “very nice,” and “step change.” Billionaire venture capitalist Marc Andreessen published on X that “AI experts say GLM-5.2 is China’s first AI model that matches and often beats” top US public models. Guillermo Rauch, CEO of AI authoring platform Vercel, he said that he is “impressed, almost shocked” by GLM-5.2’s coding capabilities. Or as one AI startup at a San Francisco dinner party recently told me: “Don’t hide the GLM-5.2.”
In a sense, GLM-5.2 is China’s answer to Claude’s principleAnthropic agent assistant that has reinvented the growth of AI. This has been the year of AI agents—tools that aren’t just chatter but promise to do things for you, whether it’s writing a website or booking a vacation. Chinese AI models have been rapidly improving, but none have previously been proven capable or robust enough to serve as agents. Now GLM-5.2, developed by the Chinese company Z.ai, competes with other advanced versions of OpenAI and Anthropic and, by many measures, has jumped ahead of Google Gemini. And it is several times cheaper.
Despite Silicon Valley’s horror over GLM-5.2, a cheaper competitor couldn’t have come at a worse time for America’s frontier AI lab. After successfully convincing the American company to test their products, OpenAI, Anthropic, and Google are now struggling to prove that their tools are worth the money. These bots can be very expensive to run, adding up to thousands of dollars, per employee, per month. Uber reportedly spent its entire 2026 budget on Anthropic models in just a few months. Other Big Tech companies including Meta, Amazon, Tesla, and Adobe have also been reported squeeze down on employee AI usage. Citi once closed employee access to the more expensive models of OpenAI and Anthropic, according to to report from 404 Media (which Citi has contested).
Although political leaders and tech executives have framed the United States as the center of a contentious technological competition with China, Americans can still use Chinese models. While it’s too early to tell if GLM-5.2 can truly replace America’s top-tier AI agents, any cost-conscious company or developer may now have an alternative. The arrival of GLM-5.2 poses a business problem for Silicon Valley—and possibly a national security problem for the country as a whole.
In some ways, America’s AI industry has been here before. Most of the interesting early Chinese AI models haven’t started a large exodus of OpenAI and Anthropic customers, with one major exception: In January 2025, after DeepSeek launched a cheap AI model on par with America’s best, the use of Chinese AI models skyrocketed. Within two months, the share of global Internet traffic for Chinese AI models increased from about 3 percent to 13 percent, according to research from RAND. OpenAI, Anthropic, and Google quickly responded to DeepSeek with cheaper prototypes of their own, and then the rise of AI agents earlier this year seemed to put American labs on top.
Even before Z.ai unveiled its new model, there have been quiet signs of a shift to cheaper, Chinese models amid growing concerns about AI bills. In the first five months of 2026, adoption of DeepSeek among the 70,000 U.S. companies using Ramp, a financial operations platform, increased from 0.1 percent to 0.3 percent, Ara Kharazian, Ramp’s chief economist, told me. Kharazian added that 6 percent of Ramp’s AI customers use third-party platforms that provide access to many different AI products. On OpenRouter, one such platform, six popular AI models are Chinese; in less than a month, GLM-5.2 already ranks fifth. This data excludes all developers and companies that directly download Chinese AI models, which are usually open source, and install on their computers—that is, startups and academics who don’t have the budget to use a fintech service like Ramp. Chinese designs accounted for nearly half of open source AI downloads from February 2025 to 2026, according to data from the popular AI Hugging Face forum.
Meanwhile, OpenAI and Anthropic continue to release more, rather than less, expensive products. And soon, they will have more competition. Claude’s code and AI agency’s enthusiasm started seven months ago, which is also roughly behind the Chinese AI companies have been lagging in prototype development. Other Chinese labs, including DeepSeek and Moonshot AI, will surely release AI agents with similar capabilities and at a lower cost soon. Coinbase, the famous crypto company, claims is about to cut its AI use in half by not pursuing cheaper models including the GLM-5.2 and Kimi, another popular Chinese robot. “The worrying situation is that China has enough prototypes for a quarter of the price,” Kyle Siler-Evans, an AI researcher at RAND, told me. “I think that’s the possibility of the future we’re heading towards.”
Kharazian cautioned against exaggerating this narrative. While some tech companies are pulling back on their use of AI, most US companies are already spending very little on the technology: The average Ramp client spends just $11 per employee on AI. “It goes without saying that the rest of the market won’t go in that direction,” Kharazian said, but Anthropic and OpenAI will have “ample time to respond” to competitive pricing—as they did with DeepSeek.
Perhaps the biggest obstacle to the GLM-5.2’s uptake, at least in the US, will have little to do with model capability or price. They are there serious anxiety about Chinese companies using their AI to collect sensitive data and steal corporate secrets. The perceived risks of using Chinese technology can create a ripple effect among any American company’s customers—not to mention the uncertainty that comes from potential federal regulation. Consider that Chinese electric cars are by all accounts better and cheaper than their Western counterparts, but they cannot be bought in the US; one possibility is a future in which Americans are practically or legally barred from accessing the most cost-effective AI designs.
The larger implications of GLM-5.2 and the coming invasion of cheap Chinese AI agents may also be geographical rather than economic. Any soft power the US gains from its technology could wither as developers in countries with less strained relations with China turn to DeepSeek and GLM-5.2 in greater numbers (note how the EU, UK and Canada are importing Chinese EVs) Meanwhile, despite Big Tech spending more money on advanced AI chips to train their models—chips that are banned from being exported to China—the gap between US and Chinese models widens. it has not expanded; if anything, it may be slowing down. That means that any military, economic, cyber or other benefits that AI might provide to a nation may disappear. America’s lead in the AI race, for the first time since DeepSeek, is in serious danger of slipping.
