The headline says it plainly: “as AI Bills Balloon.” Google launched a pay-as-you-go edition of Gemini Enterprise on August 26, 2026, and the name they chose for the problem is exactly what finance teams have been saying behind closed doors for the past year.

The new pricing option lets businesses pay for actual token, compute, memory, and storage consumption rather than committing to a fixed per-seat subscription. It comes with no upfront commitment or base subscription fee, which lowers the barrier for companies that want to pilot AI agents across departments without locking into a large monthly bill before they know what they are getting.

The more consequential addition is the spending cap feature. Businesses can now set hard monthly budget limits at the project level, receive automated email alerts when spending hits 50 percent, 80 percent, and 100 percent of the ceiling, and have agent API calls paused automatically once the cap is reached. The pause affects only that project’s agent calls; other workloads continue running normally.

Google also introduced Flexible Savings Plans that apply discounts of 10 percent for a one-year commitment and 20 percent for a three-year commitment, with no minimum or maximum spend required. An off-peak feature is coming that will offer discounts of up to 50 percent for tasks customers are willing to defer to lower-demand windows.

The two pricing models are designed to coexist. Per-seat subscriptions are not going away, and Google positions pay-as-you-go as something companies can blend with their existing seat licenses rather than replace them.

What This Means for Business

The sticker shock problem is real, and Google acknowledging it in the headline of its own announcement tells you something. Enterprise AI spending went from a line item to a board conversation faster than most companies anticipated. The shift from fixed seats to consumption-based pricing reduces the risk of overcommitting, but it introduces a different challenge: variable costs that are harder to budget for and harder to explain to a CFO.

The spending cap behavior is worth paying attention to if you are running AI agents in production. When an agent hits the budget ceiling, its API calls pause. That sounds reasonable in a billing context, but it means a workflow that was running fine in the morning can stop mid-task in the afternoon. Any agent-powered process needs to be designed with graceful stops in mind, which is not something most teams think through during a pilot.

The broader signal here is that enterprise AI has crossed from the “let’s try it” phase into “now we have to manage it.” Usage controls, audit trails, billing visibility, and budget guardrails are not nice-to-haves; they are the infrastructure that makes AI sustainable at scale.

For businesses evaluating their AI commercial model right now, the choice between per-seat and pay-as-you-go comes down to how predictable your usage is. Consistent, high-volume workloads favor seats. Variable, project-by-project use favors consumption pricing. Most companies at mid-scale will end up running both, which adds a new complexity layer to AI governance.

This is exactly the kind of infrastructure decision that Omni Advisory exists to help businesses navigate. Getting the commercial structure right early prevents you from re-platforming everything six months later when the bills come in higher than modeled.

If you are running or planning AI agents at scale, book a discovery call with Sam McKay to pressure-test your AI cost model before you hit a spending ceiling in production.