Flo Crivello's San Francisco startup Lindy.ai migrated 100% of its traffic to the Chinese AI model DeepSeek-V4 last month after discovering its Anthropic bill exceeded payroll for more than two dozen employees. The switch saved the company millions of dollars and reflects a broader shift among U.S. businesses grappling with artificial intelligence as their fastest-growing expense.
"By far, our No. 1 expense was Anthropic," Crivello said. "Like, more than payroll." More than rent. More than anything else. "It was just 10x cheaper," he said of DeepSeek-V4. "So it was a very, very simple business decision."
The cost pressure driving the shift
AI costs are not just a startup problem. Uber CEO Dara Khosrowshahi said on the Invest Like the Best podcast last month that the company "blew through our AI budget in a quarter, you know, for the whole year, essentially. And it is forcing us to adjust." Bloomberg reported that Airbnb CEO Brian Chesky said his company relied on Alibaba's Qwen model last year, describing it as "good," "fast and cheap." Perplexity and Nvidia have also used Qwen.
Crivello said every founder he knows working in AI is either thinking about switching to Chinese models or has already done so. Many companies keep quiet about it due to political sensitivities, but the models are widely available on platforms like Hugging Face, GitHub, and through inference providers based outside China.
Chinese models dominate open-source
U.S. companies like Anthropic, OpenAI, and Google lead in building the most capable AI models. Experts estimate Chinese models trail by six to twelve months on raw capability. But China has carved out a commanding position in open-source models, which are free to download and adapt. "The open-source scene right now is absolutely dominated by the Chinese. It's not even close," Crivello said.
OpenRouter, a platform where startups access a range of AI models, reported that use of Deepseek has climbed from roughly 9% to nearly 20% since January. Usage of models from Chinese firms MiniMax, Xiaomi, and Tencent has also risen. San Francisco-based Featherless, which offers access to some 30,000 models, has seen similar demand. Founder and CEO Eugene Cheah said Chinese models are popular even when they are not best in class.
"It's like the difference between driving a Ferrari and a Honda. You can have the best luxury car, or you can just have a Honda at scale that works," Cheah said. "Actually, a lot of open-source AI groups are perfectly fine being N-1, N being where the frontier is. Because as the gap keeps shrinking, at some point the question is: Does it actually matter?"
When cheaper models are good enough
For many companies, the Honda of AI works fine. Victor Su-Ortiz, who does global product marketing at Shanghai-based MiniMax, attended a recent AI engineers conference in San Francisco and said the calculus comes down to cost per token. Companies pay for AI models by the token - a unit of AI work - and repetitive, high-volume tasks do not need a frontier model.
"A lot of repetitive tasks can be done with a model that's just as performant but has much lower cost per token" compared with leading AI models, Su-Ortiz said. He described a shift from "tokenmaxxing" - using as much AI as possible - to routing different types of work to different models. For deep reasoning or research, the top-tier models may perform better. "But if you're routing for a coding task that is repetitive, high volume … that's where one of our models, especially MiniMax M3, will perform exceptionally well at like only one-tenth the cost."
Some users download and self-host open-source Chinese models. Others use paid AI-hosting companies like Featherless and OpenRouter so that user data stays in the United States. Either way, the price difference is reshaping procurement decisions across the industry, a dynamic that falls squarely under AI for Executives & Strategy.
Where U.S. models still hold the edge
Not every company is ready to switch. Jon Gordner, CEO and co-founder of Comment.io, which is building a product he described as Google Docs for coders and AI agents, said his weeks-old startup needs the best available models right now. "We need to make as good software as we can as fast as possible. And for us, saving a few dollars on a cheaper model isn't worth it if we have to spend two or three more weeks fixing its mistakes," he said.
Gordner said Anthropic and OpenAI are currently subsidizing users to hook customers, offering tokens at a steep discount through monthly subscriptions. But he expects that to change. "Then for us, it's going to make a lot more sense to start evaluating Chinese models and open-source models," he said.
The competitive response from U.S. companies
Ara Kharazian, lead economist at Ramp, a company that helps businesses track and control spending, said he believes U.S. AI firms will adapt - either by keeping prices in check or releasing high-quality open-source models of their own. "The rise of these Chinese models is indicative of the fact that businesses want something that is today not being offered by the American model companies," Kharazian said. "The only reason why I'm bearish about the Chinese models is because I assume that the American model companies will respond competitively."
Gordner is less certain. Both Anthropic and OpenAI have filed confidential paperwork with the U.S. government to begin the process toward initial public offerings. As pressure to demonstrate profitability rises, he thinks the major U.S. AI companies may have to start charging more. "At some point," Gordner said, "the music's going to stop."
Why this matters for IT, development, and finance professionals
Cost per token is becoming a line-item concern for any team building on top of large language models. IT and development leads should evaluate whether repetitive, high-volume tasks - code generation, email triage, calendar management - can be routed to cheaper models without sacrificing output quality. Finance professionals, meanwhile, need to track AI spending as a distinct budget category. When a startup's model inference bill outstrips payroll, the margin math demands scrutiny. The gap between frontier and adequate is shrinking, and the companies that treat model selection as a procurement decision rather than a technical one will have a cost advantage.
Your membership also unlocks:
