Artificial intelligence is getting cheaper to use, but companies may still end up paying more for it. That is the central finding of a new report from Accenture Research, covered in New Delhi on 28 October 2026. The report expects the average price of AI tokens to fall by about 19 per cent over the next 24 months, while token consumption is expected to rise by about 78 per cent over the same period. In plain terms, each unit of AI work costs less, but companies are likely to do so much more of it that their total bill could still go up.
To understand why, it helps to know what a token is. In simple terms, a token is a small chunk of text or data that an AI system reads or produces, and companies are typically charged based on how many tokens they use. A short question might use only a handful of tokens, while a long report, a detailed analysis or a lengthy conversation can use thousands. As AI tools become cheaper and easier to use, employees naturally turn to them more often and for bigger tasks, which pushes total consumption up even as the price per token falls.
The report suggests this pattern is already taking shape. Only 35 per cent of organisations can precisely calculate the cost per business outcome for their highest-volume AI use case. That is a striking gap: many companies know roughly what they spend on AI, but far fewer can say exactly what each completed task, resolved query or finished report actually costs them. Without that clarity, it is difficult to judge whether AI is saving money or simply shifting costs elsewhere.
What would companies do if prices fell further? The report offers a revealing answer. If token prices fell by 25 per cent or more, only 5 per cent of organisations said they would keep the savings. Far more would spend them: 42 per cent would expand AI use to more employees, and 35 per cent would introduce applications previously considered too expensive. In other words, cheaper AI does not lead most companies to bank the difference. It leads them to do more with AI, to more people, in more parts of the business.
That expansion is not necessarily wasteful. The report finds that 60 per cent of executives plan to redirect newly released capacity towards higher-value or strategic work, while 51 per cent intend to invest it in growth initiatives. The logic is familiar from other technologies: when a tool becomes cheaper, the smart move is often to aim it at bigger problems rather than simply cut the budget. A customer service team might use AI to handle routine queries and free staff for complex cases. A research team might analyse far larger volumes of information than was previously affordable.
But the report also carries a warning, and it is one that finance teams will recognise. The true cost of AI is not the price per token. It is the total cost of AI-enabled work compared with the value that work generates. A company that doubles its AI usage because prices fell by a fifth has not saved money; it has spent more, and it needs to be able to show that the extra work was worth it. Measuring value, not just usage, is therefore becoming a core management task.
Accenture’s message, then, is not that AI is becoming unaffordable. It is that affordability and total spend are different things. The price of each token is heading down, while the volume of tokens is heading up much faster.
