The rapid growth of artificial intelligence is creating a new challenge for businesses: working out how much AI-powered services should cost.
Consumers can access free versions of services such as ChatGPT, Claude and Gemini, while companies including Microsoft, Google and Anthropic have invested huge sums in developing large language models. Paid AI products offer additional capabilities, including coding, automation and business functions.
At the same time, technology companies are developing AI agents designed to perform specific tasks. These systems can use several AI agents together to analyse information, make decisions and take actions. However, the unpredictable amount of computing required by these systems makes pricing difficult.
Simon Gooch of identity management company Saviynt said companies cannot easily commit customers to fixed AI costs for several years because the economics of the technology are changing rapidly.
AI systems process user instructions and responses through units known as tokens. Prompts are converted into tokens before being processed by a model, while the resulting response is also generated as tokens.
The number of tokens required can vary depending on the request. Slight changes in instructions can produce different results, while different AI models may consume different amounts of tokens. Agentic systems can increase usage further because several AI agents may work together on a single task.
Although the cost of individual tokens has fallen sharply, overall consumption is rising quickly. Goldman Sachs expects monthly token use to increase 24 times between 2026 and 2030, reaching 120 quadrillion tokens per month as businesses increasingly adopt AI agents.
Companies can therefore face unexpected bills when employees use AI extensively. Will Venters, associate professor of Digital Innovation and Information Systems at the London School of Economics, said businesses are struggling to control spending because AI outputs and usage are difficult to predict.
Some smaller companies have reportedly relied on flat-fee personal accounts to keep costs under control. Oliver King-Smith, founder of smartR AI, said such arrangements may become harder to maintain as major AI providers seek stronger returns from their investments.
Businesses are also being encouraged to select AI models carefully and give systems more precise instructions. Rob Steele, chief financial officer at UK accounting software company iplicit, said detailed prompts can help prevent unnecessary usage.
The challenge becomes greater when AI is built into products used by thousands of customers. Companies may need additional tokens for software development, testing, security and safety controls. Adding more AI agents can also be much easier than hiring additional employees, potentially causing usage to grow rapidly.
Yet higher AI spending could also deliver greater value if the technology improves productivity and results.
Technology companies are still trying to determine how those costs should be passed to customers. Bill Peterson of Sumo Logic said possible approaches include higher overall prices, charging according to results or offering bundles based on specific incidents.
However, any pricing strategy could quickly change if major AI providers alter their own charges.
As AI agents become more widely used, businesses face the difficult task of balancing unpredictable technology costs with the value those systems deliver.
