The Old World of Predictable Spending
For decades, IT budgeting has been a relatively straightforward affair. When a company decided to roll out new software—be it a CRM, a design suite, or office productivity tools—the cost structure was clear. Businesses typically paid a flat, per-seat
license fee, either as a one-time purchase or a recurring annual subscription. This Software-as-a-Service (SaaS) model allowed Chief Financial Officers (CFOs) to forecast expenses with a high degree of accuracy. If you added 100 new employees, you knew exactly what the corresponding increase in software costs would be. This predictability provided comfort and control, forming the foundation of technology budget planning for a generation.
Why AI Breaks the Mold
Generative AI operates on a fundamentally different economic model. Instead of a fixed fee per user, most AI platforms use consumption-based pricing. This can include costs per API call, per 'token' (the chunks of text processed), or based on the amount of computational power used. This shift from predictable IT spending to variable, usage-based costs is at the heart of the current challenge. In a recent Deloitte survey, nearly half (46%) of CFOs cited cost uncertainty and a lack of transparency as their biggest internal concern regarding AI adoption. The more your team uses the AI, the more you pay—and usage can be difficult to forecast. A simple query costs less than a complex one, and an AI agent designed to perform a multi-step task can generate numerous calls to a model, multiplying costs in ways that are hard to see in advance.
The Hidden Cost Iceberg
The visible, token-based fees are only the tip of the iceberg. Several other factors contribute to the spiraling and unpredictable nature of AI expenses. High-performance GPU infrastructure, essential for running AI workloads, is expensive to provision and operate. Many teams over-allocate these resources to guarantee availability, leading to significant waste when they sit idle. Furthermore, there are hidden data taxes; moving large amounts of data between cloud services can incur steep egress fees. Other often-overlooked expenses include the costs of data preparation and cleaning, ongoing model maintenance, and the need for specialised talent to manage these complex systems. One IBM report found that 70% of executives cite generative AI as the primary driver behind an expected 89% rise in computing costs between 2023 and 2025.
Taming the AI Budget
Despite the challenges, businesses are not powerless. The key, according to experts, is shifting from passive budget approval to active financial governance. Deloitte emphasizes that data readiness and strong governance are critical elements that need to catch up with the technology itself. This involves implementing a robust AI cost management strategy. Companies are now turning to new tools and practices, collectively known as 'AI FinOps', to gain control. This includes setting clear budgets and limits for teams, projects, and even individual users. Real-time alerts can warn managers when spending approaches a threshold, preventing budget blowouts before they happen. Another powerful strategy is intelligent model routing—using high-cost, state-of-the-art models only for tasks that require them, while routing simpler queries to cheaper, more efficient models. By establishing this granular visibility and control, organisations can turn unpredictable AI spend into a manageable part of their operations.
