For much of the past two years, artificial intelligence has been sold as an almost limitless productivity engine. Company executives spoke about faster coding, automated customer service, improved research capabilities and new ways to reduce labour costs. Investors rewarded anything connected to AI with soaring valuations, while businesses rushed to integrate chatbots, coding assistants and autonomous software agents into daily work.
What received far less attention was a simple question that many finance departments are now beginning to ask: what happens when the bill arrives?
That question is becoming increasingly difficult to ignore as companies across the technology industry take a closer look at the cost of running large language models. While AI adoption continues to grow, reports from corporate users suggest that spending on these systems is rising much faster than many organisations initially expected. The result is a growing debate about whether the economics of AI can support the expectations that have been built around the industry.
The discussion has gained fresh attention following reports that enterprise software company Workato saw its AI costs rise dramatically after changes to the way it was charged for access to Anthropic’s models. According to accounts from company executives, a move from a flat subscription arrangement to usage-based pricing caused expenses to jump sharply within a short period of time.
The episode has become a widely discussed example of a broader issue facing businesses that rely heavily on AI services. During the early stages of adoption, many customers gained access through pricing structures that made experimentation relatively inexpensive. As usage increased and companies became dependent on these systems, billing models increasingly portrayed the actual computing resources consumed by each request.
That distinction matters because large language models require enormous computing power. Every query, response, image generation request or software task consumes processing resources in data centres packed with advanced chips. While the user may see a simple conversation box, the costs involved in producing those responses can be substantial when multiplied across millions of daily interactions.
Companies Begin Looking More Closely at AI Spending
The growing focus on costs is not limited to one company or one AI provider. Several large corporations have reportedly begun placing limits on internal AI usage after discovering that spending was rising faster than expected.
Technology companies that were among the earliest advocates of workplace AI are now paying closer attention to employee usage patterns. Reports suggest that some businesses have introduced spending caps, approval processes or tighter monitoring of AI subscriptions after discovering that employees were using multiple services simultaneously or running expensive automated tasks without clear business benefits.
The issue portrays a reality that many organisations are only now beginning to confront. Generative AI can improve productivity in certain tasks, but measuring the financial return remains difficult. While software engineers may complete projects more quickly and customer support teams may handle larger workloads, translating those gains into direct financial savings is often less straightforward than promotional presentations suggest.
Survey data from consulting firms and industry researchers has pointed to a growing gap between expectations and outcomes. Many companies report positive experiences with AI tools, yet a sizeable share say the financial benefits have been smaller than originally forecast.
That does not necessarily mean AI has failed. Rather, it highlights the difference between technical capability and economic value. A system may perform impressively while still being expensive to operate at scale.
The challenge becomes even more complicated when businesses begin deploying AI agents that perform tasks continuously. Unlike a traditional software licence, where costs remain relatively predictable, usage-based pricing means expenses can increase rapidly as activity grows.
Finance executives who once focused primarily on cloud computing bills are now finding AI expenditure becoming a major line item. In some cases, internal reports suggest that individual departments are generating monthly AI costs that rival those of entire software teams only a few years ago.
A Difficult Moment for AI Companies Seeking Public Investors
The timing of this debate is particularly important because several leading AI firms are preparing for a period in which public market scrutiny may become far more intense.
OpenAI and Anthropic have both been linked to discussions surrounding future stock market listings, with valuations that would rank among the largest technology offerings ever attempted. Such valuations depend heavily on assumptions about future revenue growth and customer demand.
Yet investors examining these businesses will inevitably focus on a different question as well: can AI providers generate sustainable profits while keeping customers satisfied?
The issue is complicated by fierce competition throughout the industry. If one provider raises prices aggressively, customers may consider alternatives. If companies reduce prices to maintain market share, profitability becomes more difficult to achieve.
Recent market developments portray that tension clearly. New AI models from Chinese developers have entered the market with substantially lower operating costs than some leading American rivals. Benchmark studies comparing identical workloads have shown wide differences in pricing between competing models.
Those comparisons do not necessarily indicate that lower-cost systems are superior. Quality, reliability, safety controls and customer support all matter. However, they do highlight the increasing importance of economics in a market that has often been dominated by discussions of model capabilities.
For AI providers, this creates a difficult balancing act. Building advanced models requires enormous spending on data centres, specialised chips, electricity and research. Customers, meanwhile, are becoming more sensitive to costs as they move from experimentation to routine use.
Public market investors have seen similar stories before. During earlier technology booms, companies often prioritised rapid growth over profitability. That approach can work for a time, particularly when capital is plentiful and investors are willing to support expansion. Eventually, however, attention tends to return to margins, cash flow and long-term economics.
None of this suggests that demand for AI is disappearing. Businesses continue to invest heavily in the technology, and many organisations view AI tools as an increasingly important part of their workflows. What appears to be changing is the nature of the conversation.
The focus is moving from what AI can do to what AI costs.
