Artificial intelligence is getting smarter, but making money from it is proving to be a shockingly expensive exercise. According to new findings from Gartner, the cost of generating a positive return from AI has skyrocketed 18-fold in just two years.
The Alarming 18x Cost Increase
The central finding from technology research firm Gartner is a wake-up call for any organisation investing in artificial intelligence. While more than half of companies are reporting some positive value from their AI initiatives, the investment required to achieve those gains is climbing dramatically. The data, presented at the Gartner IT Symposium/Xpo 2026 in Kochi, India, reveals a growing disconnect between the corporate world's enthusiasm for AI and its commercial viability. According to Gartner senior director analyst Neha Kumar, rising costs for infrastructure, model consumption, integration, and governance are significantly eroding the financial benefits. This means it's not enough for AI to just work; it must deliver enough value to justify a rapidly ballooning price tag.
Meet the AI Agents
A key driver of this cost explosion is the shift from basic AI chatbots to a new, more powerful class of technology: AI agents. Unlike a chatbot that answers a single prompt and stops, an AI agent is a more autonomous system. It can understand a goal, create a multistep plan, access other software and data, and take action without a human approving every single step. Think of it as a digital employee that can not only provide information but also execute tasks like filing a purchase order, updating a customer record, or managing complex workflows. This autonomy is what makes them so promising, but also what makes them so much more expensive and risky to operate.
The Hidden Costs of Autonomy
Why are these agents so much more costly? Gartner identifies what it calls the 'Inference Paradox': even as the per-unit cost of AI processing (tokens) falls, the total cost of running complex AI systems is escalating. An AI agent performing a multistep task can use 5 to 30 times more resources than a simple chatbot answering a question. One single user request can trigger a chain reaction of dozens or even hundreds of calls to the underlying large language model. Gartner forecasts that inference costs—the cost of running the AI model to make a prediction or decision—for each agent-driven workflow will increase by more than five times by 2028. This is because organisations are reinvesting efficiency gains into more powerful, and thus more expensive, frontier models to stay competitive.
A Widening Governance Gap
The speed of adoption is far outpacing the ability of organisations to control these new systems. Gartner's surveys reveal a massive gap between intent and readiness. While nearly three-quarters of organisations plan to deploy agentic AI within the next year, only a small fraction have the necessary governance capabilities in place. In the Asia-Pacific region, including India, only 14% of CIOs report having agentic AI governance. This lack of oversight is not a theoretical problem. One survey showed 95% of organisations had already faced at least one AI agent-related incident in the past year, from cascading system failures to security breaches and budget overruns. Without proper guardrails, these autonomous agents can expose a company to significant financial, operational, and reputational damage.
The Path to Control
Gartner's warning comes with a clear call to action: CIOs must shift their focus from rapid experimentation to establishing robust governance frameworks. This involves creating what Gartner calls 'good drag'—controls that reduce harmful risk while allowing valuable initiatives to proceed safely. Recommendations include establishing clear accountability, implementing automated safeguards and spending caps, and ensuring human oversight for high-impact actions. The goal is not to slow down AI adoption indiscriminately, but to ensure that the risks do not grow faster than the value AI creates. Success with AI will no longer be about simply having the technology, but about having the maturity to control it effectively.
















