What Gartner's Finding Really Means
Announced at its recent IT Symposium in Kochi, Gartner's finding highlights a growing gap between AI enthusiasm and its commercial reality. It doesn't mean a single AI project will cost 18 times its initial budget. Rather, it signifies that the overall
spending by organizations to turn AI experiments into measurable business outcomes has skyrocketed. Many businesses are experiencing a 'rude awakening' when they move from contained, low-cost pilot programs to full-scale production. While a majority of leaders report seeing some positive results from AI, the rapidly climbing cost to achieve them is making it harder to justify the investment, with Gartner predicting that half of all generative AI projects will overrun their budgets by 2028.
Beyond the Pilot: Where Real Costs Hide
The core issue is a fundamental misunderstanding of an AI solution’s Total Cost of Ownership (TCO). Companies budget for the exciting development phase but fail to account for the far more expensive operational lifecycle. The initial training of an AI model is just the beginning. The real expense lies in 'inference'—the day-to-day running of the model to perform tasks. Gartner estimates that inference will account for at least 70% of a model’s lifetime costs. These hidden expenses cluster in a few key areas: the data pipelines that feed the AI, the infrastructure it runs on, the constant maintenance it requires, and the specialized talent needed to manage it all.
The Meter Is Always Running
A major driver of these escalating costs is the industry's shift to consumption-based pricing for AI models. Instead of a fixed subscription, many companies now pay per 'token', a unit of data processed by the AI. While the cost per token is falling, the complexity and capability of AI are growing even faster. Newer 'agentic' AI systems, which can perform multi-step tasks autonomously, consume five to thirty times more tokens than a simple chatbot for the same outcome, causing costs to spiral unexpectedly. Furthermore, AI models are not 'set and forget' assets. They require constant monitoring and maintenance to prevent 'model drift', where performance degrades over time, adding a significant and recurring operational expense that many budgets overlook.
The Human Factor and Governance Gaps
Technology is only part of the equation. The cost of people and process is another frequently underestimated factor. Demand for specialized AI talent has driven compensation for these roles to three or four times that of an average worker. However, the skills themselves have a short shelf life, creating a risk of overpaying for capabilities that quickly become outdated. Beyond salaries, there are costs associated with workforce transformation, change management, and the necessity of human oversight to manage AI outputs and prevent errors. Without strong governance that dictates how, when, and which AI tools are used, organizations risk 'tool sprawl' and uncontrolled spending that can easily erase any productivity gains the technology was meant to deliver.
From Cost to Value: A Smarter AI Strategy
Avoiding the 18-fold cost trap isn't about abandoning AI; it's about adopting it more intelligently. Gartner’s primary advice is for organizations to become much more selective about the projects they pursue. Instead of chasing the latest trend, leaders should focus on initiatives with a clear and measurable business case. The key is to demonstrate a return on investment before committing to a full-scale rollout. This requires a strategic shift from treating AI as a short-term IT project to managing it as a long-term product with a full lifecycle. Indeed, a recent CEO survey found that 86% of chiefs consider redesigning their operating models a priority for generating real value from AI, signalling a move towards deeper, more strategic integration rather than superficial tech upgrades.
















