Google Targets AI Sticker Shock: What Gemini’s Pay-As-You-Go Pricing Means for Enterprises

AI agents can work for hours without taking a break. Unfortunately, they can also keep spending money the entire time.

Google introduced a pay-as-you-go edition of Gemini Enterprise on Aug. 26, giving businesses an alternative to fixed per-user subscriptions. The company is positioning the consumption model as a lower-commitment way to test AI agents while adding controls intended to prevent unexpectedly large cloud bills.

For enterprise buyers, the important question is whether paying for actual usage will reduce costs or simply replace predictable license fees with a bill that changes every month.

Google adds consumption pricing and spending limits

Under Google’s new billing options, organizations can combine conventional seat subscriptions with pay-as-you-go access.

The consumption edition has no upfront commitment or base subscription charge. Instead, companies pay for resources such as tokens, memory, compute, and storage at Google’s standard pay-as-you-go rates.

That is different from the API discount eWeek recently covered when Gemini 3.7 Flash launched. That announcement reduced the cost of using a particular model. The new program changes how businesses can pay for the broader Gemini Enterprise app and its agent workloads.

Google describes pay-as-you-go as an option for organizations with at least 20 seats. The edition is gradually rolling out and remains available to a limited group of customers.

The update also gives administrators several ways to constrain spending. Companies can set hard monthly caps at the project level, receive alerts when spending reaches 50%, 80%, and 100% of the budget, and choose whether workloads pause or continue at consumption rates after exhausting pooled quotas or reaching a spending limit.

Google is also pooling quotas across Gemini Enterprise, its Antigravity development environment, Android Studio, and custom agents. Unused capacity purchased for one group can therefore absorb heavier demand elsewhere in the organization.

What eWeek found: Pay-as-you-go savings narrow as usage grows

Pay-as-you-go pricing could make sense for pilots, seasonal projects, and teams with irregular demand. Its cost advantage may narrow as agents begin running larger, recurring workloads.

Google lists Gemini Enterprise Standard/Plus as starting at $30 per user per month. That puts the starting subscription cost at $750 per month for 25 seats, $3,000 per month for 100 seats, or $15,000 per month for 500 seats.

The pay-as-you-go edition carries no seat fee and includes the capabilities in Standard/Plus with one current exception: Gemini Notebook is not yet included, although Google says it is coming later. Organizations must contact Google for actual pricing and eligibility, so the figures below are directional rather than complete customer quotes.

The consumption alternative does not have one universal price because the bill depends on the models, token volume, memory, compute, storage, region, and workload. The following eWeek calculations use Gemini 3.7 Flash’s global-region Standard PayGo input and output token rates to illustrate how utilization can change the comparison. They show estimated model-token charges, not total Gemini Enterprise deployment costs.

Google charges 10% more for Gemini 3.7 Flash Standard PayGo usage outside the global region. Non-global rates are $0.825 per million input tokens and $4.125 per million output tokens through Dec. 31, 2026. Those rates rise to $1.65 and $8.25, respectively, on Jan. 1, 2027.

These figures show only input and output token charges. Google says pay-as-you-go customers may also be billed for resources such as memory, compute, and storage, so the calculations do not represent complete Gemini Enterprise bills. They also do not account for the quota allowances included with seat subscriptions.

Even with those limitations, the comparison illustrates the central trade-off. Consumption pricing may produce substantially lower model-token charges when usage is limited or intermittent. As recurring agent workloads consume more tokens and other metered resources, the difference between consumption and subscription pricing could shrink.

Companies evaluating the new option should measure actual resource consumption during a limited pilot rather than estimate demand based on employee headcount. eWeek’s AI pricing guide shows how sharply costs can vary across models and service tiers.

Google offers discounts for predictable workloads

For organizations with steady demand, Google introduced Flexible Savings Plans that reduce eligible token costs by 10% with a one-year commitment or 20% with a three-year commitment. Customers select a monthly spending amount, but the commitment cannot ordinarily be canceled or modified after purchase.

If eligible usage falls below the chosen amount, the customer still owes the full monthly commitment, and the unused portion does not roll over. Usage above the committed amount is billed at standard on-demand rates without the plan’s discount.

Deferred execution pricing is also coming for select workloads. Businesses will be able to mark nonurgent agent jobs for off-peak processing, potentially reducing inference costs by up to 50%.

These options come as Google continues to promote smaller models to reduce the cost of enterprise agents. Its earlier Gemini 3.6 launch similarly emphasized lower-cost models for high-volume business workloads.

The new pricing structure could lower the financial barrier to starting a Gemini Enterprise pilot by removing the seat fee for participating customers. It does not eliminate the risk of overspending once autonomous agents begin chaining together searches, tool calls, and model requests.

The companies most likely to benefit will be those that treat AI consumption like cloud infrastructure: measure it, cap it, and verify that each recurring workload generates enough value to justify its costs.

What this means for enterprise buyers

For IT and finance leaders, the choice should not begin with the number of employees who might use Gemini. It should begin with the volume and type of work those employees and agents are expected to perform.

Before committing, companies should run a controlled pilot and track token consumption, model selection, memory, compute, storage, and the frequency of automated workflows. They should also test spending caps to understand what happens when a project reaches its limit and determine which workloads can safely pause without disrupting business operations.

Ownership matters as well. Finance teams need visibility into how costs are distributed, while IT administrators should establish which models and agent workflows employees can use. Without those controls, a small number of automated processes could generate a disproportionate share of the bill.

Pay-as-you-go could be the better starting point for uncertain, seasonal, or intermittent demand. Seat subscriptions may provide more predictable budgeting for teams that use Gemini consistently and can take full advantage of their pooled quotas. Flexible Savings Plans add another option for organizations that can reliably forecast monthly consumption, although businesses still pay their committed amount when usage falls short.

The most useful comparison will therefore come from complete monthly costs, not token prices or employee counts alone. Buyers should examine every metered resource, included feature, quota allowance, and commitment term before deciding which pricing model offers the better value.

Also read: Explore the biggest Gemini and AI-agent announcements from Google I/O 2026 and what they reveal about Google’s broader strategy for enterprise AI.