As artificial intelligence disrupts the traditional per-seat pricing model, dozens of enterprise software vendors are gradually moving away from charging customers a single fixed subscription fee per user.

Data tracked by Kyle Poyar, a former investment partner at OpenView, covering 500 software companies shows that by the end of 2025, 79 companies had begun charging customers extra based on actual AI usage, including HubSpot, Adobe, and Salesforce. This figure more than doubled compared to 2024.

The background for this pricing shift is that after enterprise customers subscribe to AI features for a fixed fee, their usage has surged dramatically, continuously driving up the operational costs for software vendors.

Simultaneously, as companies increasingly rely on AI agents to replace employees in operating systems, the number of subscription accounts required by businesses may decrease in the future, also forcing vendors to adjust their profit models.

This year, companies like ServiceNow and ManpowerGroup have already implemented and expanded usage-based pricing, applying tiered charges for intelligent tools based on the volume of data processed by AI.

Another industry giant, Atlassian, announced it will soon introduce a consumption-based billing model for its AI features. Its AI tools can support office scenarios such as file retrieval, copywriting, and automatic meeting summarization.

Previously, most software vendors bundled AI features within high-end enterprise subscription plans without charging separately.

It's not just traditional enterprise software companies undergoing transformation. In recent months, Anthropic and OpenAI have also comprehensively adjusted their pricing, fully implementing AI usage-based billing for large enterprise customers.

The direct reason for Anthropic's price adjustment is that fixed-plan users heavily increased their use of high-frequency functions like code generation, causing vendor costs to soar, or users quickly hitting their plan's usage limits.

However, pay-per-use models can easily lead to uncontrolled corporate spending.

Adrian Balfour, Founder and Chairman of AI consulting firm Envorso, stated, "The vast majority of my clients dislike this model. AI costs can skyrocket in a very short time."

Even large enterprises with mature technology systems, like Uber, are struggling to adapt to the new AI pricing, especially concerning code development tools.

Uber's Chief Technology Officer, Pravin Nepali Naga, admitted that just a few months into 2026, the company's annual AI budget had already been completely exhausted.

Google: Sticking Primarily to Fixed Subscriptions Some companies still prioritize traditional bundled subscriptions.

Snack giant Mars Group, early this year, procured the full suite of Google AI tools for all 62,000 employees. Its Chief AI and Technology Officer, Shubham Merish, stated that the core reason for choosing Google was the predictable cost of a fixed annual fee per employee.

Merish added that Google only charges overage fees after an employee's usage exceeds the limit, and this additional cost constitutes a very low portion of the overall partnership. The labeled price for this cooperation exceeds $20 million annually, with an actual corporate discount applied.

"Under our contract model, Google bears most of the risk associated with token usage. The vast majority of employee AI usage is included within the fixed monthly fee."

Industry Evolution: From Usage-Based to Pay-for-Performance More vendors are moving beyond simple usage-based billing to explore new rules.

Adobe announced last week that it plans to adopt a task-completion-based pricing model for some AI tools, charging for actual delivered outcomes. HubSpot has already implemented a similar model this month, stating plainly that the entire industry will eventually shift entirely to "charging only for effective output."

Chris Hogan, Executive Vice President of Operations at HubSpot, pointed out, "Currently, businesses are blindly betting on AI, often paying for potential value rather than actual output. This model harms customer interests and is not conducive to the healthy development of the industry."

However, pay-for-performance poses challenges to software company profits: different customers require varying amounts of AI resources to achieve the same business outcomes, significantly increasing the difficulty of profit forecasting.

Furthermore, various AI products must face homogenized competition from similar offerings by competitors like Anthropic and OpenAI.

Salesforce CEO Marc Benioff stated in an interview that the company will adopt a hybrid pricing strategy, including a pay-for-performance model, to flexibly adapt to different customer needs.

"There is no one-size-fits-all solution for serving large, complex enterprises."

Salesforce launched its usage-based AI product, Agentforce, in late 2024 and adjusted its measurement approach in February of this year, moving away from solely counting token consumption to tracking the number of actual tasks completed.

In the future, it may use a self-developed metric, "Intelligent Work Units," as the core basis for charging.

Currently, the pay-for-performance model remains controversial within the industry.

Rick Opal, an executive at IT consulting firm BDO Digital, believes, "The maturity of enterprise users' AI applications is still insufficient. There is a lack of unified measurement standards, making it impossible to determine if the pricing for a given AI output is fair and reasonable."

In contrast, Anthropic's current policy charges enterprise edition Claude a base fixed subscription fee, on top of which full usage-based billing is added, further increasing the cost for enterprises through this dual-charge structure.