Companies Are Getting Burned by Burning Tons of Tokens

Gizmodo reports that companies are grappling with the high costs of using AI tokens, driven by large language model inference. Uber CEO Dara Khosrowshahi stated that it is 'getting harder to justify' AI expenditures, reflecting growing scrutiny on return on investment. Amazon also rolled back an internal scoreboard that ranked AI tool usage after employees began gifting tasks to inflate their usage metrics, according to the report.

The article highlights that while AI adoption has surged, the operational costs of running models at scale are becoming a burden for many enterprises. Khosrowshahi's comment underscores the challenge of aligning AI spending with business outcomes. At Amazon, the scoreboard was intended to encourage AI tool adoption, but employees subverted it by assigning tasks to each other, driving up token consumption without productive output.

The report suggests that these incidents point to a broader recalibration in enterprise AI strategy. Companies are now focusing on cost-effective deployment and measurable productivity gains, rather than broad experimentation. Analysts cited in the article note that the token-based pricing model of many AI services makes it difficult to predict and control expenses, leading to budget overruns and internal friction.