AI agents hide infrastructure costs, Google Cloud survey finds

Google Cloud conducted a survey of 1,402 IT leaders focused on infrastructure needs for AI agents in operational environments. The results painted a picture of organizations rushing to adopt AI agents without fully understanding the infrastructure and cost consequences.

The headline finding was stark: 83% of respondents said their organizations need to upgrade infrastructure to support agentic AI. Yet the survey also uncovered a critical blind spot—62% of leaders acknowledged that hidden costs are present in their AI deployments, with the biggest culprits being data egress (the cost of moving data out of cloud systems), legacy storage systems, and new hardware requirements. Even more troubling, 81% of surveyed leaders said they are failing to track these costs.

When asked where these costs are hiding, the research revealed a pattern of fragmented planning. Organizations are typically forecasting AI agent requirements in isolation, without integrating them into overall infrastructure strategy. As a result, costs accumulate in places IT teams are not monitoring—particularly in the background operations of AI systems running on shared cloud infrastructure.

Google Cloud stressed that managing these costs requires visibility into where compute is actually running. The company recommended that organizations move workloads to the edge—deploying AI processing on-device or on local servers—to shift computing away from expensive cloud resources and reduce variable costs. The survey findings were published as part of Google Cloud's report titled "2026 State of infrastructure in the agentic AI era," which ranked among the top infrastructure trends expected in the year ahead.

The survey data points to a fundamental gap in how organizations plan for AI agent adoption. While 83% acknowledge the need for infrastructure upgrades, the parallel finding—that 62% of leaders recognize hidden costs but 81% fail to track them—suggests a disconnect between awareness and action. Google Cloud's analysis indicates that IT teams are compartmentalizing their AI agent planning, treating agents as isolated workloads rather than integrating them into broader infrastructure forecasting. This approach allows costs to accumulate invisibly, particularly through data egress charges, which can escalate rapidly in cloud environments.

The report frames cost management as a core infrastructure problem, not merely an accounting gap. By recommending edge deployment and workload migration away from centralized cloud resources, Google Cloud is implicitly acknowledging that the current cloud-first model for agent computing is expensive. The finding that organizations need upgrades yet lack cost visibility suggests that many will face bill surprises only after deployments are underway—a pattern that could reshape how businesses budget for AI infrastructure in 2026.