Unpacking the 'AI Inflation' Finding
The 18-fold spending increase is a stark figure that captures the immense pressure on businesses to adopt AI. According to Gartner analyst Daryl Plummer, this surge has created a period of 'AI inflation', where the technology is driving up costs across
the entire IT landscape. This isn't just about buying new software; it's a ripple effect. Global IT spending is now forecast to hit nearly $7.4 trillion by 2027, with AI as a primary driver. Yet, this spending spree comes with a paradox. A recent Gartner CIO survey found that while 58% of chief information officers are pressured to cut costs using AI, 51% expect AI to actually increase their total technology costs. This highlights a fundamental disconnect between the executive mandate to innovate and the on-the-ground reality of implementation.
The Total Cost of Ownership Iceberg
The initial cost of an AI model or a software license is merely the tip of the iceberg. The real expenses—the Total Cost of Ownership (TCO)—lie beneath the surface. Gartner predicts that by 2028, the costs of running an AI model (inference) will account for at least 70% of its total lifetime cost. This is a profound shift from traditional software. Every time an employee uses an AI tool, it incurs a cost. Beyond that, hidden expenses accumulate rapidly. These include the high salaries of scarce AI talent, the cost of preparing and cleaning vast datasets, integrating AI with legacy systems, and ensuring security and governance. Some reports suggest that preparing enterprise data can be nearly as expensive as the model itself. Many firms find that production-ready AI systems are orders of magnitude more expensive than initial pilots, a rude awakening for those who budgeted based on experimentation.
The Elusive Hunt for ROI
Despite worldwide AI spending projected to reach $2.7 trillion in 2026, proving a direct financial return remains a major hurdle. A study by MIT found that a staggering 95% of GenAI pilots produce no measurable impact on profit and loss, even with billions invested. The problem is twofold. First, many of AI’s benefits are 'soft' and difficult to quantify, such as improved employee experience or future strategic advantages. Second, companies struggle to even track the true costs. One report found that reported AI spending can be 12 times lower than the actual AI-driven consumption of cloud resources, as costs get buried in general infrastructure bills. This makes the classic ROI formula—gains minus costs, divided by costs—nearly impossible to calculate accurately. As a result, Gartner predicts that around 30% of GenAI projects will be abandoned by the end of 2025 due to escalating costs and a failure to demonstrate value.
A Call for Strategic Discipline
The takeaway isn't to abandon AI, but to move from hype-driven experimentation to disciplined, strategic implementation. The era of writing blank cheques for AI is over. Experts advise a more grounded approach: start with a clear business problem, not with the technology. It's crucial to tie every AI project to a measurable outcome, whether it's cost savings, revenue generation, or productivity gains. Instead of building massive, custom models from scratch—which can cost up to $20 million—businesses are encouraged to use smaller, more efficient models trained on high-quality, domain-specific data. This requires strong governance to control usage, track costs per interaction, and ensure that projects demonstrate a positive return before being scaled across the organisation. The focus is shifting from 'can we build it?' to 'should we build it, and does it pay off?'.
















