Nearly a Third of C-Suite Executives Cannot Understand Their AI Bills, KPMG Survey Finds

Twenty-nine percent of corporate leaders cannot understand their AI operating costs as they scale enterprise deployments, according to a KPMG survey of 2,145 senior leaders across 20 countries published this week. The shift from flat-rate subscriptions to usage-based billing at Anthropic, OpenAI, and GitHub has left finance teams struggling to forecast, monitor, and manage AI spending.

The Numbers

A third of senior corporate leaders identified limited understanding of AI costs and economics as a direct challenge to deploying AI agents. Nearly half of the organizations surveyed have rephased their AI deployments after costs outweighed expected value. Lower-cost, high-fidelity models are now the fastest-growing influence on AI strategy, up seven percentage points from Q1 2026.

“As usage-based pricing models become more common, many organizations are still building the capabilities required to forecast, monitor, and manage AI spending effectively,” KPMG said in the report.

The cost confusion matters for agent deployments specifically because agents compound spending in ways traditional software does not. An autonomous agent that runs continuously, calling APIs, invoking tools, and processing context windows generates variable costs that scale unpredictably with usage patterns rather than seat counts.

Vendors Respond with Engineering, Not Transparency

Rather than simplifying pricing, the two largest cloud providers are investing billions to embed engineers directly with enterprise customers. AWS announced a $1 billion investment in a Forward Deployed Engineering organization designed to help customers build and deploy agentic AI solutions, compressing deployment timelines from months to days. Microsoft followed with $2.5 billion for a new entity called Microsoft Frontier Company, providing 6,000 engineers to work directly with customers on AI deployments.

The combined $3.5 billion in forward-deployed engineering commitments addresses one half of the problem: enterprises need help building. But neither program directly tackles the cost visibility gap that KPMG identified. Companies can now get expert help deploying agents faster while remaining unable to predict what those agents will cost to run.

Governance Gaps Compound the Problem

KPMG also flagged that AI governance remains immature across most organizations. The report found that while most organizations report having some governance mechanisms in place, few describe those practices as “fully embedded” in daily operations.

“Organizations need clear rules for when employees can intervene, who owns AI-related costs, how AI outputs are reviewed and what happens when systems fail,” the report stated.

The governance question intersects directly with cost control. Without clear ownership of AI-related spending at the operational level, usage-based costs can proliferate across departments with no single team accountable for the aggregate bill.

The Credibility Footnote

KPMG’s own track record on AI reporting adds an uncomfortable layer of context. In June 2026, GPTZero published a forensic review of a prior KPMG report on agentic AI, finding that only five of its 45 citations pointed accurately to the cited source. KPMG removed the report and acknowledged the failure. The new cost survey’s underlying data, drawn from 2,145 respondents across 20 countries, appears methodologically distinct from that earlier report, but the episode demonstrates exactly the kind of AI governance failure the firm is now warning others about.

The Cost Visibility Gap for Agent Builders

The KPMG data quantifies a problem that anyone running autonomous agents already knows intuitively: usage-based pricing makes AI costs a function of agent behavior, not headcount. For teams deploying OpenClaw or similar agent platforms, the 29 percent figure likely understates the challenge, since always-on agents with persistent context windows and tool-calling loops generate spending patterns that are harder to model than batch API calls. The enterprises rephasing their deployments are not losing faith in AI. They are discovering that deploying agents without cost observability is like running a fleet of taxis without meters.