随著全球人工智慧供应商从固定收费模式转向按使用量计费，以及更为复杂且消耗资源的「AI代理」（AI agents）日益普及，企业使用AI的成本急剧上升，许多公司（如Uber）甚至在2026年年初就耗尽了全年的AI预算，使得控制AI成本成为企业面临的重大挑战。

为了解决预算超支的问题，各大企业纷纷采取应对措施，包括限制员工在特定时间内可使用的Token数量（如Atlassian）、开发智慧路由工具以针对不同任务选择最合适且最具性价比的模型（如Notion和Hostinger），以及转向使用可本地部署且安全性更高的开源AI模型来降低高达70%的成本。

尽管企业开始加强预算管理与使用纪律，高盛仍预测到2030年全球Token消耗量将成长24倍，这主要由AI代理的广泛应用所驱动；因此，企业在保留创新与实验空间的同时，如何将AI的使用与实际工作成果紧密结合，将是未来发展的关键。

As global AI providers transition from flat-fee structures to usage-based pricing and highly complex, resource-intensive "AI agents" grow in popularity, the cost of utilizing AI for businesses has surged dramatically, with companies like Uber exhausting their entire 2026 AI budgets early in the year, turning AI cost management into a major challenge.

To address budget overruns, companies are implementing various cost-control measures, such as capping the number of tokens employees can use (e.g., Atlassian), building routing tools to select the most suitable and cost-effective model for specific tasks (e.g., Notion and Hostinger), and adopting open-source AI models that can be hosted locally to reduce costs by up to 70%.

Despite growing controls on spending and the emphasis on discipline, Goldman Sachs predicts a 24-fold increase in global token consumption by 2030, driven by the adoption of AI agents; consequently, the key for businesses moving forward will be balancing room for experimentation with tying AI usage directly to work outcomes.