In a recent NPR story, John Ruwitch reports on how the rising cost of AI is pushing some U.S. technology companies to switch from top-tier American models to cheaper Chinese ones, saving millions of dollars. For Flo Crivello, whose San Francisco startup Lindy.ai builds AI assistants, the math was stark: his biggest expense wasn’t payroll or rent. “By far, our No. 1 expense was Anthropic,” he told Ruwitch. “Like, more than payroll.”
Below are key highlights from the article:
- The scale of the shift. Ruwitch reports that AI has become one of the fastest-growing costs for U.S. businesses, necessary but expensive. Last month, Lindy migrated 100% of its traffic to the Chinese model DeepSeek-V4. “It was just 10x cheaper,” Crivello said, calling it “a very, very simple business decision.” American companies like Anthropic, OpenAI, and Google still lead on capability, with experts putting Chinese models six to 12 months behind, but China now dominates the open-source scene. Ballooning costs aren’t only a startup problem: Uber’s CEO said the company burned through a year’s AI budget in a single quarter, and Airbnb, Perplexity, and Nvidia have all used Chinese models like Alibaba’s Qwen.
- Like a Ferrari or a Honda. Ruwitch notes many firms stay quiet about using Chinese models due to political sensitivities, though the models are widely available through hubs like Hugging Face and GitHub. Eugene Cheah, CEO of the model platform Featherless, frames the tradeoff simply: “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.” As the capability gap shrinks, he adds, many teams are “perfectly fine being N-1.” Victor Su-Ortiz of the Chinese firm MiniMax points to cost-per-token as the driver, noting repetitive, high-volume tasks like routine coding can run on a cheaper model “at like only one-tenth the cost.” To ease data concerns, many companies access these models through U.S.-based hosting providers that keep user data in the country. On OpenRouter, use of DeepSeek has roughly doubled since January, to nearly 20%.
- Saving a few dollars isn’t worth it for everyone. Ruwitch reports the shift isn’t universal. Jon Gordner, CEO of the newly founded Comment.io, told him cheap models aren’t worth it if his team has to spend weeks fixing their mistakes. For now, he also benefits from American companies subsidizing users with discounted tokens to win customers, though he doubts that lasts. Ara Kharazian, lead economist at Ramp, expects U.S. firms to respond competitively: “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,” he said, adding that he assumes American providers will answer the challenge.
- What happens next is unsettled. Ruwitch reports differing predictions. Kharazian thinks U.S. companies will keep prices in check or release strong open-source models to compete. Gordner suspects the opposite, that American providers may have to charge more as pressure to show profitability grows, with both Anthropic and OpenAI having filed confidential paperwork toward eventual IPOs.
For now, the decision facing companies boils down to a simple question: how good does the AI need to be for the job? In many cases, the Honda is enough to get them where they need to go. But as Gordner cautioned, today’s era of inexpensive, subsidized frontier AI is unlikely to last forever: “At some point, the music’s going to stop.”
Acknowledgement: This summary was generated using AI writing tools as part of my experimental learning process. For more information, check out the original article listed below.