Stu Clott works in operations management in San Diego and codes on the side. A one-hour coding session on Claude was running him around ten dollars. Switching to DeepSeek brought that cost down to less than fifty cents. “I laugh every time I go see the costs,” he told Rest of World. “The output quality, to be honest, I can’t tell the difference.”
Clott is not an outlier. Across the United States, a quiet but accelerating shift is underway among developers, startups, and even some larger technology companies, who are replacing American AI models with Chinese alternatives, not out of ideological preference, but out of straightforward economic calculation. The models from DeepSeek, Alibaba’s Qwen series, and MiniMax produce results that are indistinguishable from those of their US counterparts at a fraction of the price in many use cases.
The Economics Are Difficult to Ignore
The cost differential is not marginal. Flo Crivello, founder of Lindy, a San Francisco-based AI work assistant company, announced in June that his company had switched from Anthropic to DeepSeek, saving what he described as “millions of dollars.” His reasoning was blunt: “You don’t need God to write your email. If you can get those lower tiers of intelligence for a tenth of the price, it would be foolish not to do it.”
Ruben Garcia Jr. runs a development practice in Dallas, building websites and mobile apps for business clients. His monthly AI spend breaks down into two tiers: roughly five hundred dollars for Claude and ChatGPT, which he reserves for the most demanding planning and review tasks, and around two hundred dollars for a set of Chinese alternatives, such as MiniMax, Moonshot’s Kimi, and Xiaomi MiMo, a trio that collectively handles about ninety percent of his workload, from coding to voice recognition. “If the Chinese models come out and they are frontier and cheaper, I’m going that direction,” he said.
The data from infrastructure providers confirms the trend. On Vercel, DeepSeek’s slice of total token consumption grew from below 1% to 17% between April and May 2026, yet its revenue share held near 1%, a gap that illustrates both the scale of adoption and the pricing pressure it creates for US model providers. On OpenRouter, the platform that routes developer queries across dozens of models, the top four by usage are all Chinese: DeepSeek, Tencent, MiniMax, and Xiaomi.
Microsoft and the Enterprise Dimension
The shift is not confined to individual developers. Microsoft is exploring using DeepSeek or another open-source model as a lower-cost alternative to power Copilot Cowork, which is currently built on Anthropic and OpenAI. The exploration reflects a broader recognition among enterprise software companies that the cost of running AI at scale is becoming a significant line item, and that Chinese open-source models offer a way to manage that cost without sacrificing capability.
The political dimension of enterprise adoption is more fraught. Airbnb and Anysphere, the company behind the Cursor coding assistant, both faced congressional investigations after disclosing their use of the Chinese open models Qwen and Kimi, respectively. Airbnb CEO Brian Chesky was compelled to clarify publicly that no user data was being sent to the model developers. The episode illustrates the reputational and regulatory risk that large companies face when adopting Chinese AI, even when the technical case for doing so is strong.
Kyle Chan, a fellow at the Brookings Institution, has identified the medium-sized business segment as the most likely growth market for Chinese AI adoption: “AI adoption at large companies is fairly saturated. The growth market for Chinese companies would be medium-sized businesses that are starting to get into AI but are wary of the costs.” Large companies in regulated industries remain reluctant due to data security concerns, censorship risks, and geopolitical exposure.
The Infrastructure Behind the Price Gap
The cost advantage enjoyed by Chinese model providers is not accidental. It reflects a combination of lower salaries and infrastructure costs in China, a deliberate strategy of releasing open-weight models that reduce the marginal cost of deployment, and subsidized API pricing that prioritizes adoption over near-term revenue. DeepSeek, Xiaomi MiMo, and MiniMax are among the most cost-efficient models per token on the Artificial Analysis index.
Poe Zhao, a Beijing-based founder and author of the Hello China Tech newsletter, has framed the dynamic in terms of infrastructure capture: “Chinese models are becoming part of the global AI infrastructure layer. But usage is only the first step. The next test is whether Chinese model developers can convert that usage into revenue, enterprise trust, and durable distribution.”
That conversion challenge is real. Many American users access Chinese models through American cloud providers specifically to avoid their data transiting through Chinese infrastructure, a workaround that preserves the cost advantage while limiting the data exposure. Whether Chinese AI companies can build the enterprise trust required to move beyond that workaround is the central question for the next phase of their global expansion. EastFrontier has previously reported on how Chinese open-source models have been gaining ground in the Global South; the Rest of World report suggests the same dynamic is now playing out in the United States itself.