Enterprise input to AI is moving from encouraging use to controlling costs. Many large technology and retail enterprises have recently tightened the privileges or levels of access to the AI tool for their employees because of the repeated calls for models for smart-body products, which have led to a rapid increase in business bills.

According to media reports such as The Information, Bloomberg and the British Financial Times, AT&T has restricted some of its employees from accessing GitHub Copilot under the Microsoft flag; Meta is also tightening staff spending on the use of Anthropic and other AI services. Uber, Wal-Mart and Amazon have also adjusted their internal policies, focusing on lowering AI usage costs.

Enterprises start capping AI

Cost pressures have spread from individual teams to the corporate level. According to the report, AI ' s higher-intensity enterprises can spend $7,500 a month per employee. Even if the single call price of the model falls, the smart body tool will be called on several times in the course of a mission, eventually pushing the total bill up.

Uber is a more prominent case. In April 2026, the company had used the full-year AI programming budget and subsequently adjusted the monthly ceiling per employee, per tool, to $1,500. Wal-Mart also set a ceiling for internal AI smarts. In the case of the Amazon, the internal list of employee rankings at the level of AI was removed to reduce the additional consumption to flush the rankings.

Microsoft also found that some engineers spent $500 to $2,000 on Claude Code only a month. This is an indication of the cost of an enterprise AI, which is more directly related to the way in which it is used than the procurement model itself.

Differences focused on efficiency and budget

Not all companies have opted for synchronized tightening. Box CEO Aaron Levie stated that the company had never used a ranking to encourage its employees to consume token in large quantities and therefore had no similar deviations. The director of the project in Databricks stated that the company currently did not have an AI budget cap for the engineer.

This reflects differences of judgement within the enterprise. Some consider that restrictions on use help to bring costs back to the budget as soon as possible, while others believe that continued input is still cost-effective as long as employees can translate AI into higher output.

However, the core of the problem has not changed. Enterprises had previously been willing to expand AI spending, mainly on the basis of productivity growth expectations. A significant reduction in usage today could also weaken this return.

Increased demand for cost-control tools

Budget tightening is also changing the direction of the AI industrial chain. An increasing number of enterprises have begun to shift simple tasks from high-cost front-line models to cheaper or more open-source alternatives in order to reduce expenditure without significantly reducing usage.

In this context, the demand for model gateways, the use of surveillance and model route tools has increased. Microsoft and Databricks have introduced gateway products to help businesses monitor staff calls and set spending caps. A model router was also released this month by the AI Software Company, Inc., Inc., Inc., which invests in Weida, with the goal of automatically allocating low-complex tasks to lower-cost models.

Mr. Nadra, Chief Executive Officer of Microsoft, has also stated publicly in recent days that the AI model should operate as a fungible commodity, rather than concentrating value on a few model providers. This statement is largely consistent with the reduced claims of current business clients.

Microsoft's emphasis on manageable costs when pushing new products

In response to the client ' s compression of the budget, Microsoft this week disclosed the pricing of the new product Copilot Work. The product is based mainly on the Anthropic model and is used to automate more complex multi-step tasks in Office 365.

The fee model is “Authorized fee plus volume charge”. Users are required to purchase the 365 Copilot authorization from the $30 monthly fee and to make additional payments for the actual usage of Copilot Work. This is close to the billing system that Anthropic introduced to corporate clients earlier.

At the same time, Microsoft stressed that clients could set a ceiling on the use of Copilot Cowork and planned to support the switch in the product to OpenAI, Microsoft self-study or other models to reduce costs. It was also reported that Microsoft was testing options for replacing the Anthropic model with an open source model under some scenarios.

This suggests that the new competitive point in the enterprise software market is moving from a simple modeling capability to who can keep AI costs within an acceptable enterprise while ensuring results.