The push toward production and its hidden costs
58% of heads of IT infrastructure and operations consider it a priority to turn experimental artificial intelligence projects into large-scale operational use over the next twelve months. The figure, reported by Gartner, comes from a survey conducted between April and May 2026 among 272 executives and accompanies the report on the five AI trends for infrastructure and operations in 2027, presented on October 26. According to analyst Jonathan Forest, the transformation under way is among the most significant ever faced by this corporate function.
The analysis urges companies to concentrate investments on a few strategic initiatives rather than intervening in all applications simultaneously. The expected benefit is twofold: less manual work and faster incident response. But the same adoption introduces new spending categories and greater complexity, requiring closer coordination among operations, security and finance leadership.
Autonomous agents and the question of control
The first trend is the rise of agentic AI, namely software capable of pursuing assigned goals with limited human intervention. In infrastructure, these agents can classify incidents, analyze anomalies, verify system changes and apply corrective procedures. Gartner forecasts that by 2030, 50% of organizations will use AI agents with minimal human involvement in IT operations, versus almost no adoption in 2026: this is a forecast, not a market figure already achieved.
The organizational model would change accordingly, moving from direct supervision of decisions to control over systems that act autonomously. This requires new safeguards: defining permissions, logging operations, and procedures for stopping and restoring previous configurations, because an agent that makes a mistake on critical infrastructure can rapidly amplify the effects of an incident. On October 20, 2026, Gartner included AI agent identity and autonomous vulnerability remediation among the ten strategic technology trends for 2027.
Unpredictable spending and hybrid architectures
The second trend concerns the actual cost of agentic AI. Language models are billed per token, and an autonomous agent can make numerous queries and repeat operations, making spending less predictable than with traditional licenses. Gartner estimates that through 2027, companies that use agents with variable rates without consumption limits will exceed initial cost expectations by more than 50% compared with those adopting subscriptions at predictable prices. Hence the principle proposed by the analyst: no agent in production without tools capable of measuring its consumption, performance and costs. The issue is confirmed by the FinOps Foundation's State of FinOps 2026, according to which 98% of participating organizations manage AI spending, versus 31% two years earlier.
The third trend concerns the distribution of computing resources. Companies mainly use already-developed models via inference, and Gartner forecasts that by 2029, 80% of large enterprises will manage AI applications in production across at least three infrastructure environments, including public cloud, neocloud, enterprise data centers and edge. Hybrid architecture makes it possible to place sensitive workloads in controlled environments and those with demand spikes on external capacity, but it entails higher costs for integration, data transfer, monitoring and security.
Record investments and financial risks
The forecasts should be read alongside the scale of investments. On October 21, 2026, Gartner estimated that worldwide IT spending will reach $7.352 trillion in 2027, growing 15% from the $6.391 trillion forecast for 2026; data center systems would rise from $879 billion to $1.366 trillion. In research dated October 14, 2026, however, the same analyst estimated that more than $3.5 trillion in AI infrastructure investments could be exposed to impairment by 2030 if suppliers maintained their current pace of capital expenditure, because revenue generated by AI could prove insufficient. The available material stops at this point.
