What happened
AI operational costs are escalating rapidly, prompting companies to re-evaluate their adoption strategies. This follows an initial period of "subsidised intelligence" where investors absorbed costs, but now major AI providers like OpenAI and Anthropic are preparing for IPOs in late 2026, shifting focus to profitability. The surge in expenses stems from rising token prices, the computational demands of AI agents that "burn through dozens of times more tokens than a simple chat message," and persistent shortages in computing chips and data centres. Meta's CTO Andrew Bosworth and Uber's COO have both questioned the productivity returns on current AI spending.
Why it matters
Rising AI operational costs will force procurement teams and platform engineers to re-evaluate model selection and deployment strategies. The shift from subsidised pricing to profitability for frontier model providers means budget holders must now account for escalating token consumption, particularly with agentic workflows. This pressure drives a move towards smaller, specialised, or open-source models to reduce expenditure, impacting vendor lock-in and infrastructure choices. Uber's COO recently stated that the company's significant AI spending has not translated into proportional productivity gains, indicating other enterprises will face similar scrutiny regarding AI investment returns.
