Analysis
A KPMG survey of 2,145 senior leaders across 20 countries found that nearly 29% of corporate executives struggle to understand and manage operating costs as they scale enterprise AI deployments, The Register reported July 3, 2026, as major AI vendors including Anthropic, OpenAI and GitHub shift portions of their services away from flat-rate subscriptions toward usage-based billing models.
The survey found the cost-comprehension problem is severe enough to actively change deployment behavior: nearly half of the leaders surveyed said their organizations have postponed AI deployments specifically when expenses exceeded the projected value of the initiative, and a full third cited limited cost comprehension itself as a direct obstacle to deploying AI agents at all -- suggesting the problem isn't merely budgetary surprise after the fact, but genuine difficulty forecasting costs well enough to make deployment decisions with confidence.
Organizations are adapting in measurable ways: KPMG found a 7-percentage-point quarter-over-quarter increase in companies favoring lower-cost, high-fidelity models over premium frontier options, a concrete behavioral shift toward cost discipline rather than defaulting to whichever model has the best raw capability. That shift mirrors dynamics covered elsewhere in this issue, including Together AI's neocloud pitch built explicitly around dramatic cost savings versus closed-model pricing.
The governance dimension compounds the problem: KPMG noted that while most organizations report having some AI governance structures in place, relatively few describe those practices as fully embedded into daily operations -- meaning many companies lack the operational tooling and processes to actually monitor and control usage-based AI costs in real time, even where formal policies exist on paper.
The underlying shift from flat-rate to usage-based billing is happening across the industry simultaneously rather than at a single vendor, meaning enterprises can't simply solve the problem by switching providers -- Anthropic, OpenAI and GitHub moving in the same direction means usage-based, harder-to-predict billing is becoming the industry default rather than an isolated vendor choice.
For enterprise buyers and finance teams, the survey is a clear signal to build dedicated AI cost-governance and monitoring capability now, given that nearly half of peer organizations are already pausing deployments over cost unpredictability rather than proceeding and adjusting later. For AI vendors and infrastructure providers, the data is a real opportunity: tools that make usage-based AI costs more predictable, forecastable or controllable address a documented, widespread pain point rather than a hypothetical one.
What to watch: whether major AI vendors introduce better cost-forecasting or spend-capping tools in response to this documented enterprise pain point, whether the shift toward lower-cost models continues accelerating, and whether AI governance practices catch up to formal policy in the way KPMG's research suggests they currently haven't.
