AI Costs out of control: how to plan your token budget

The Moment the AI Bill Arrived on the Table

For years, the tech sector has been racing. The dominant logic was that of tokenmaxxing : maximize the use of language models, experiment, scale. However, as a recent analysis by TechCrunch , the industry is currently grappling with out-of-control AI infrastructure costs. The phrase that summarizes the change is eloquent: «The whole conversation shifted from tokenmaxxing and ‘go fast’ to ‘we need guardrails, how do we control this?’»

Therefore, 2026 marks a concrete discontinuity. It's no longer about exploring the possibilities of artificial intelligence. It's about understanding the real cost of integrating it into business processes. This applies to large corporations, but even more so to Italian SMEs.

The numbers that matter: tokens, costs, and real consumption

The concept of tokens is central to this analysis. A token roughly corresponds to three or four characters of text. Every call to a language model—whether it's GPT-4o, Claude 3.5, or Gemini—consumes tokens for both input and output. Consequently, the cost grows proportionally to the complexity and frequency of requests.

According to aggregated industry data, a medium-sized company integrating AI into three or four workflows can spend between €2,000 and €15,000 per month on API calls alone. Furthermore, these costs tend to grow over time as use cases multiply. Without a monitoring system, spending becomes opaque and difficult to justify during budget reviews.

Research of McKinsey confirm that AI cost management has become one of the main concerns for CTOs in 2026. Specifically, the difficulty is not just technical: it's organizational. Who approves the expenses? Who monitors consumption? Who decides when a model is too expensive compared to the value generated?

Strategic Reading: Why SMEs Are More Exposed Than Large Companies

Large enterprises have dedicated MLOps teams. They have engineers who optimize prompts, reduce context length, and choose the most efficient model for each task. In contrast, Italian SMEs often adopt AI solutions in a fragmented way, relying on SaaS tools that hide the underlying costs.

This creates a specific risk. In fact, an SME using an AI writing tool, a chatbot for customer service, and a data analysis system might not realize they are paying for tokens three times over. Therefore, the fragmentation of tools translates into a fragmentation of costs, which is difficult to aggregate and even harder to optimize.

In addition to this, the AI model market is rapidly evolving. Newer, more capable models tend to cost more per token. Therefore, automatically updating integrations to the latest available version can mean significant, often unplanned, cost increases.

We at SHM Studio we observe this pattern with increasing frequency. SMEs that turn to our AI services often lack a clear mapping of their consumption. The first step in our work is always to build this visibility.

Problem Architecture: Where Costs Are Hidden

To understand where to intervene, it's useful to map the main sources of token consumption in a typical company. There are three main categories of spending.

• Inefficient prompt engineering: prompts that are too long, redundant contexts, instructions repeated with every call. Every extra character has a cost.
• Oversized model for the task: Using GPT-4o to classify simple emails is like using a scalpel to cut bread. Smaller, specialized models cost less and often perform better on specific tasks.
• Lack of caching: Many implementations repeat identical or almost identical calls without leveraging caching mechanisms. This multiplies costs without any additional benefit.

Similarly, the absence of spending limits configured at the API account level is a concrete operational risk. Gartner has identified the cost governance of AI as one of the top ten technological priorities for the 2026-2027 biennium, as read in their technology forecasts .

Operational implications for Italian SMEs

Translating this analysis into concrete actions requires a structured approach. Below are the priority intervention areas for an SME that wants to bring its AI spending under control.

First of all, mapping. It's necessary to inventory all touchpoints with AI APIs: SaaS tools, custom integrations, automations. Without this visibility, any optimization is blind.

Next, segmentation by value. Not all AI tasks have the same strategic value. Therefore, it's useful to classify use cases based on the return generated and the cost incurred. This allows for rational budget allocation.

Then, choosing the right model. Today, there are open-source models and mid-range commercial models that offer excellent performance on specific tasks at a fraction of the cost of flagship models. Choosing a model is an economic as well as a technical decision.

Finally, continuous monitoring. AI costs aren't static. They grow with adoption, change with model updates, and vary with usage volumes. A monitoring dashboard — even a simple one — is an essential tool.

Our activities of Digital marketing and of SEO increasingly integrate AI components. Similarly, our clients are also exploring these integrations. For this reason, we have developed an approach that balances innovation and cost control.

What Nobody Tells You: The Hidden Cost of Iterative Prompting

There is an aspect that rarely emerges in discussions about AI costs. It is the iterative prompt : the practice of progressively refining a prompt through dozens of test calls. In a team of three or four people working on an AI integration, this process can consume hundreds of thousands of tokens even before the system goes into production.

Despite this, few companies track these development costs separately from operational costs. The result is a systematic underestimation of the total cost of ownership of an AI integration. Thus, a company's true AI budget is often double what is perceived.

This doesn't mean AI isn't worth the investment. On the contrary, it means that the ROI needs to be calculated more rigorously. Our web services and our activities SEO copywriting demonstrate daily that AI, used well, generates measurable value. However,