AI's 'invoice shock' hits the C-suite as usage-based billing replaces the free lunch

The enterprise love affair with generative AI is hitting a very hard, very expensive wall. A new KPMG survey of over 2,000 senior executives across 20 countries reveals a growing panic in the C-suite. The source of the dread? A brutal shift from flat-fee subscriptions to usage-based, per-token billing that’s making AI operating costs wildly unpredictable. Nearly a third of execs—29 percent—now admit they are struggling to understand the costs as their AI deployments scale.

This isn’t a minor budgeting hiccup. The sticker shock is so severe that nearly half of the companies surveyed are already moving to “re-phase” their AI strategies. That’s consultant-speak for slamming the brakes. As The Register’s Lindsay Clark explained, businesses are now scrambling to swap out expensive, high-fidelity models for a mix of cheaper alternatives, rather than blindly maxing out on premium AI. The game has changed: vendors like Anthropic, OpenAI, and GitHub hooked enterprises with all-you-can-eat samples, and are now, as the industry grapples with its own lack of profits, turning the screws.

The dependency is the killer. With developers increasingly reliant on AI copilots—to the point of losing core coding skills—the bargaining power is shifting. AI labs are in a precarious position. Charge too aggressively and you’ll push customers toward open-source Chinese models or other off-ramps. Charge too little, and the already unprofitable AI bubble has no path to survival. The labs are trying to thread a needle, but the KPMG data suggests the thread is already fraying. The question isn’t just whether enterprises can afford the AI bill that just landed; it’s whether the entire usage-based business model can survive the blowback from the very leaders who signed the checks.