In recent months, a growing number of companies have confronted the unexpected financial realities of incorporating artificial intelligence into their operations. Andrew McDonald, chief operating officer of Uber, recounted being informed in April that the company had expended 95 percent of its annual AI budget within just four months, with little to no measurable impact on its profitability. This experience echoes a broader trend across industries, where AI investments are increasingly viewed as costly with uncertain returns.

The pricing structure for AI services, which charges based on the number of “tokens”—units of text processed by AI models—has sharply escalated as more advanced but resource-intensive models have come to market. While executives initially encouraged widespread AI adoption to boost productivity, many now recognize that the value generated by AI has not yet matched its high operational costs. Jeetu Patel, president of technology firm Cisco, highlighted this imbalance, noting that token costs currently exceed the value they produce, raising concerns that companies might reduce usage if the economics fail to improve.

This realization coincides with a pivotal moment for the AI sector, particularly in Silicon Valley. Leading AI developers including OpenAI and Anthropic recently filed for initial public offerings valuing their businesses at over $1 trillion, while SpaceX has prominently integrated AI into its long-term strategy. However, despite rapid revenue growth and expanding user bases, these companies continue to operate at substantial losses due to sizable investments in research, model development, and data infrastructure. Analysts caution that their unprecedented valuations are not yet supported by traditional financial metrics, and that their growth trajectories face uncertainty if corporate customers scale back AI expenditures.

To manage rising costs, many organizations are implementing what industry insiders call “routing layers” — systems designed to direct AI queries to different models based on complexity and expense. Routine, low-level tasks are increasingly handled by less costly, open-source models, while only the most demanding requests access the premium, expensive platforms. Chamath Palihapitiya, a tech investor and founder of the software incubator 8090, which has developed a routing platform, cited a comparison to emphasize cost discrepancies: processing two billion tokens monthly with a Chinese open-source model like DeepSeek-Ri would cost approximately $2,740, whereas the same volume routed through ChatGPT-5.5 Pro could escalate to $105,000.

Palihapitiya also pointed to a widespread lack of oversight within companies, cautioning that many executives remain unaware that their teams often default to costly AI models without proper governance or controls, leading to unchecked spending. As businesses reassess their AI strategies in the face of mounting expenses and ambiguous returns, the industry may see a shift towards more measured and cost-effective deployment of these technologies.