Cypriot publication: Companies’ AI spending rises despite falling token prices
Cypriot publication Cyprus Mail reported that PwC is urging companies to monitor artificial intelligence spending more closely. The firm believes that spending discipline can become a competitive advantage, as AI models are increasingly becoming standardized and accessible to a wide range of companies.
According to PwC’s assessment, token prices are declining, but businesses’ overall AI spending is rising. Cheaper models encourage companies to use AI in more processes, which can complicate workflows and increase expenses.
Where unnecessary costs arise
PwC noted that companies often use more tokens than necessary without having systems to identify waste. Costs can accumulate during planning, tool use, information retrieval, model reasoning, process orchestration, the operation of safeguards, logging, and result verification. Initial budgets may also exclude indirect infrastructure costs.
Additional expenses can be generated by AI agents that create plans, pass tasks to other agents, search for information, and repeat processes if the result does not meet requirements. At the same time, according to PwC, the price of one million tokens can range from a few cents to $50, depending on the model and its tier.
More current news is available on the UA.News Telegram channel Telegram.
Cost control in AI systems
PwC advises selecting the model tier according to the specific task rather than always choosing the cheapest option. A weaker model, in the firm’s assessment, can lead to additional work, incorrect decisions, or compliance problems.
Companies are encouraged to assess the cost and value of AI projects before development begins, modify systems to reduce waste, link spending to business outcomes, and reinvest savings in new AI projects. Possible measures include reducing unnecessary context, combining several tasks into fewer requests, spending limits, and automatically routing work to the cheapest model capable of performing it.
PwC recommends embedding such mechanisms directly into AI systems: setting mandatory budget constraints, routing rules, workflow thresholds, and audit logs. The firm also advises determining the cost of each AI process per business outcome, involving the chief financial officer in controlling these expenses, and providing technical support to responsible finance staff.
PwC cited the example of a global technology company that reduced the cost of a single run by 65–80% through this approach. This allowed it to run AI three to five times more often on the same budget. Average execution time also fell from 12 to four hours, while a full-cycle audit showed that the quality of results was maintained.
