"Easy to use but expensive" vs. "Easy to use but not expensive"
With AI costs remaining high and US enterprise clients unable to burn cash, they have begun to turn their attention to Chinese large models.
According to the latest report from the American enterprise management platform Ramp, DeepSeek topped the "Software Trends List." The ranking mainly tracks companies' initial purchases from software vendors, making DeepSeek one of the fastest-growing software on the platform.
Ara Kharazian, Chief Economist at Ramp Economics Lab, said this may be the clearest signal, indicating that (U.S.) companies are seeking low-cost alternatives to OpenAI and Anthropic, and some companies are willing to adopt lower-priced Chinese large models. According to platform data, these enterprise customers pay DeepSeek directly and use the DeepSeek model, rather than deploying the open-source DeepSeek model on their own internal servers.
Ramp currently serves about 70,000 enterprise clients, a 40% increase from 50,000 at the beginning of the year. CEO Eric Glyman stated that a significant portion of the new customers come from rapidly expanding AI startups that need to manage increasing model calls, computing power procurement, and related expenditures.
Looking back at early 2025, after the DeepSeek R1 burst onto the scene, global attention skyrocketed. Its app not only topped the free app chart in Apple's China app store but also surpassed ChatGPT to top the US chart.
Apple's App Store rankings mainly focus on consumer data, and Ramp data shows that DeepSeek also experienced a brief boom in the B2B market at the time. According to Ramp's tracking at the time, DeepSeek's adoption rate among US companies once rose to 0.3%, but quickly fell back to 0.1%, and remained at this level as of April this year. For comparison, Anthropic and OpenAI have U.S. enterprise adoption rates of 34.4% and 32.3%, respectively. Ramp has not yet disclosed the latest relevant data for May.
"Easy to use but expensive" vs. "Easy to use but not expensive"
Now, DeepSeek is heating up again among US enterprise clients, directly because "US AI is too expensive." Kharazian pointed out that companies are increasingly prioritizing costs in AI spending and will increasingly explore open-source models, or model products priced lower than those of OpenAI, Anthropic, and Google.
Not long ago, an AI consultant reported that a client company forgot to set a cap on Claude licenses for employees, spending $500 million on it in just one month. Uber revealed that in just the first four months of 2026, the company has exhausted its entire "token budget"; Salesforce stated that the company will pay Anthropic about $300 million this year.
Faced with such high AI spending, even tech giants "can't burn money anymore." Amazon has halted its internal AI usage leaderboard to prevent employees from deliberately performing unnecessary tasks to boost rankings, which would increase token consumption. Additionally, Microsoft plans to cancel Claude Code subscriptions for employees in several key product divisions, planning to phase out the code by the end of June.
Consulting firm Bain recently surveyed 951 companies worldwide with annual revenues exceeding $100 million and released a report stating that after corporate AI spending exceeds $1 trillion, the actual cost savings from AI are generally far below expectations; Moreover, 44% of large companies are using "unpaid savings from the previous round of AI to fund the next round of AI investments—Bain describes this as a revolving gamble with structural vulnerabilities."
From fully embracing AI at all costs to focusing on returns and scrutinizing ROI, the trend has shifted among AI enterprise users. Compared to the "easy but expensive" American AI large models, the "easy to use but affordable" Chinese large models are increasingly favored worldwide.
Notably, in May, DeepSeek announced a permanent 75% price reduction for the DeepSeek V4 Pro model's application programming interface (API), with input fees as low as 0.025 yuan per million characters and output as low as 6 yuan, setting a new global price low for mainstream large models. MiniMax, which has repeatedly topped OpenRouter call volume charts, had its founder Yan Junjie previously revealed that MiniMax is committed to AI accessibility and has set a goal earlier: a complex agent can run for 1 hour at a cost of 1 USD.
Through underlying architecture and technological innovation, combined with energy advantages, cost-effectiveness is becoming the core competitive advantage for domestic large models attracting global users.
