大型科技公司在商业化人工智能服务的过程中遭遇定价难题。由于token消耗量存在显著的不可预测性，企业难以为客户制定稳定的订阅或按使用量付费模式[1]。Goldman Sachs预测，从2026年至2030年，token消耗量将增长24倍，达到每月120万亿枚[1]。
这种成本的波动性已经在实践中显现。微软曾被迫限制工程师使用第三方编码工具[1]，而打车服务Uber在今年数月内就耗尽了全年的AI编码token预算[1]。身份管理公司Saviynt负责人Simon Gooch指出："试图将某人锁定在12个月、2年或3年的成本模型中没有意义，因为我们不知道"[1]。伦敦经济学院教授Will Venters认为，根本问题在于"人们发现很难管理这些成本...这是非确定性输出，所以是非确定性价值"[1]。尽管各行业仍在探索如何向客户转嫁AI成本，但Sumo Logic产品营销高级总监Bill Peterson坦言："没人真正搞清楚这个问题"[1]。
Major technology companies are grappling with how to price artificial intelligence services as consumption patterns remain difficult to forecast. Goldman Sachs projects that token consumption will surge 24-fold between 2026 and 2030, potentially reaching 120 trillion tokens monthly.[1] This explosive growth is compounding uncertainty around cost structures that enterprises must pass on to their customers.
The unpredictability of token usage has made traditional pricing models impractical. Simon Gooch, from identity management firm Saviynt, notes that "it makes no sense to try to lock someone into a 12-month, 2-year, or 3-year cost model when we don't know" how consumption will evolve.[1] Real-world examples underscore the challenge: Microsoft has restricted engineers' access to third-party coding tools, while Uber exhausted its entire annual AI coding token budget within months of the year.[1] Will Venters, a professor at the London School of Economics, explains that "people find it difficult to manage these costs... it is non-deterministic output, so it is non-deterministic value."[1] Bill Peterson, senior director of product marketing at Sumo Logic, sums up the industry's confusion: "Nobody has really figured this out."[1]