Last month, Marty Kausas, CEO of customer-support software company Pylon, described an AI budgeting problem that many enterprises are likely to confront. Pylon’s annual Anthropic bill, he wrote, was on track to rise from approximately $400,000 to $1.4 million when the company crossed 150 seats. The projected increase was not the result of a sudden surge in usage. It reflected a change in pricing structure: beyond 150 seats, Pylon would move from a plan that included usage to an enterprise plan under which tokens were billed separately at standard API rates.
Kausas’s conclusion was blunt: after years of encouraging employees to use more AI, Pylon had begun introducing spending limits and requiring approval for additional consumption. Pylon’s experience reflects one part of a broader shift. Many companies began adopting AI under unusually favorable conditions: bundled usage, introductory pricing, generous enterprise discounts and relatively limited deployment. As those arrangements expire and experimentation gives way to production, the full cost of enterprise AI is becoming more visible.
At the same time, the underlying technology is becoming cheaper. The cost of inference has fallen substantially, competition among model providers remains intense and companies have more opportunities to route work to smaller models.
The result is that AI can become cheaper to use in isolation while becoming considerably more expensive to operate across an organization. As models improve, companies give AI more work, deploy AI to more employees, and incorporate it into more products. Those increases in volume and complexity can easily outpace the savings from initial lower prices.
This is not unique to AI. Computing became cheaper, and companies consumed vastly more of it. Storage prices declined, and organizations retained more data. Bandwidth became less expensive, and video came to dominate internet traffic. Efficiency often expands the market. AI is likely to follow the same pattern, but with an important distinction: much of its consumption will be difficult for executives to observe.
Traditional enterprise software is generally priced around a visible unit: a seat, transaction or customer account. Agentic systems can generate costs autonomously and unevenly. Two employees with the same license may consume radically different amounts of compute. A system may become more capable while also using longer contexts, more reasoning steps and more calls to external tools.
That creates a new management problem. Enterprises can no longer assume that purchasing access to AI is equivalent to controlling its cost.
For the first phase of enterprise AI, this was easy to ignore. Most organizations were running experiments and distributing a limited number of licenses. This told us whether people would use AI, but did not answer the question that now matters: whether that use creates sufficient economic value.
As AI budgets move from experimental to material, enterprises will need to replace adoption metrics with unit economics. For a customer-support system, the relevant measures might include cost per successfully resolved case, resolution time and escalation rate. For my company Smartling, an AI-powered translation company, it might include the cost of producing content at an agreed quality level, the time required to enter a new market, or the amount of content a company can economically make available in each language.
These measures are harder to calculate than just utilization. They require companies to define the outcome before deploying a technology, establish a baseline, and account for both the cost of the technology and the human work that remains around it.
I have spent the past several years helping build Smartling around AI. The central challenge was not introducing AI into the organization or convincing people to use it, it was determining where AI materially changed the economics of the work.
A cheaper translation, for example, has limited value if it requires enough human correction to eliminate the savings. The appropriate metric is not how many words passed through an AI system, it is the cost and business value of producing usable multilingual content.
The same principle applies across the enterprise. Every AI deployment should begin with an economic hypothesis: which cost will decline, which constraint will be removed or which source of revenue will increase? It should then be evaluated against that hypothesis rather than against the amount of AI consumed.
The next phase of enterprise AI will therefore not be defined by which companies achieve the highest adoption. It will be defined by which companies understand the economics of what they have adopted.
Organizations that can connect AI spending to revenue, margin, capacity or strategic advantage will continue to invest, even as their absolute spending rises.
AI does not need to become inexpensive to justify its place in the enterprise. It needs to become economically legible. The companies that make it so will not necessarily spend the least on AI but they will know what each additional dollar is buying.
The opinions expressed in Fortune.com commentary pieces are solely the views of their authors and do not necessarily reflect the opinions and beliefs of Fortune.
The post AI will get cheaper. Enterprise AI bills probably won’t appeared first on Fortune.
