Fast-Growing AI Costs: Artificial intelligence has become one of the fastest-growing business expenses in the U.S. Many companies face a challenge balancing the necessity of AI with its high cost.
Flo Crivello, founder of the San Francisco startup Lindy.ai, had initially favored Anthropic’s advanced AI models. However, financial analysis revealed that Anthropic was the company’s largest expense, surpassing payroll for over two dozen employees and even rent.
Crivello decided to switch 100% of Lindy’s traffic to the Chinese AI model DeepSeek-V4, which proved to be ten times cheaper and saved the company millions.
Switching Trends: Many firms are transitioning to more affordable Chinese AI models. U.S. companies like Anthropic, OpenAI, and Google lead in AI development, but Chinese models have gained prominence in open-source options, which are freely available for download and adaptation.
According to Crivello, many in the AI industry are contemplating a switch to Chinese models due to rising expenses.
Uber CEO Dara Khosrowshahi highlighted budget issues on a podcast, explaining that the company rapidly exceeded its AI budget.
Barely a year ago, Airbnb CEO Brian Chesky found Alibaba’s Qwen model beneficial due to its affordability and performance.
Notably, numerous companies remain discreet about using Chinese models due to political implications, yet these are readily available on platforms such as Hugging Face and GitHub.
Comparison and Market Impact: According to Eugene Cheah of Featherless, which provides access to about 30,000 AI models, Chinese models offer an effective, economical alternative. He likened the situation to driving choices: luxury versus practical scale.
Users often select open-source models for routine tasks, finding them dependable. OpenRouter has reported increased usage of China’s DeepSeek and other models from MiniMax, Xiaomi, and Tencent.
Cost Analysis: Companies pay for AI models by purchasing tokens, representing AI work units. Victor Su-Ortiz from MiniMax emphasized competitive cost per token as a key factor in adopting such models.
For specific tasks requiring high-volume AI work, many Chinese models provide impressive performance at lower costs.
Challenges and Adaptation: However, not all companies find Chinese models suitable. Jon Gordner of Comment.io highlighted the need for high-quality AI performance, mentioning Anthropic and OpenAI models’ reliability, partly enhanced by user subsidies.
Gordner foresees a time when evaluating Chinese models may make sense, as current discounts from American AI giants are unlikely to remain indefinitely.
Market Responses: Ara Kharazian from Ramp suggested that American companies might adapt by offering similar competitive open-source models to counteract Chinese market advantages.
Nevertheless, Gordner remains skeptical, anticipating potential U.S. AI price increases due to profitability pressures as leading American AI firms move towards public offerings.
Anthropic financially supports NPR.