Enterprises are shifting their focus from encouraging use to controlling costs when investing in AI. Several large technology and retail companies have recently tightened employee access to or limits on AI tools because smart agent products repeatedly call upon models, causing corporate bills to rise rapidly.

According to reports from The Information, Bloomberg, and the Financial Times, AT&T has restricted some employees' access to Microsoft's GitHub Copilot; Meta is also tightening spending on Anthropic and other AI services for its employees. Uber, Walmart, and Amazon have also adjusted their internal policies, all focusing on reducing the cost of using AI.

Enterprises are starting to set limits on AI.

Cost pressures have spread from individual teams to the company level. The report mentions that companies with high AI usage can spend up to $7,500 per employee per month on related expenses. Even if the price per model call decreases, the fact that AI tools call the model multiple times consecutively when performing tasks ultimately pushes up the total bill.

Uber is a prime example. The company had already used up its entire annual AI programming budget by April 2026, subsequently raising the monthly cap for each employee and each tool to $1,500. Walmart also set usage caps for its internal AI agents. Amazon, on the other hand, eliminated its internal rankings of employees based on AI usage to reduce the extra costs associated with trying to improve their rankings.

Microsoft also found that some engineers spent between $500 and $2,000 per month just on Claude Code. This illustrates that the cost of AI for enterprises is no longer just about purchasing the model itself, but is directly related to how it is used.

The disagreement centers on efficiency versus budget.

Not all companies have chosen to tighten their policies simultaneously. Box CEO Aaron Levie stated that the company has never used leaderboards to encourage employees to consume large amounts of tokens, therefore no such bias has occurred. Databricks' engineering lead said that the company currently does not set a cap on AI budgets for its engineers.

This reflects differing judgments within companies. Some companies believe that limiting usage will help bring costs back within budget as quickly as possible; others believe that continued investment is still worthwhile as long as employees can translate AI into higher productivity.

However, the core issue remains unchanged. Companies were previously willing to increase AI spending primarily based on expectations of increased productivity. A significant reduction in usage could now diminish those returns.

The demand for cost control tools is rising.

Tighter budgets are also changing the beneficiaries across the AI industry chain. More and more companies are shifting from expensive, cutting-edge models to cheaper or open-source alternatives for simple tasks in order to reduce spending without significantly decreasing usage.

Against this backdrop, demand for model gateways, usage monitoring, and model routing tools has increased. Microsoft and Databricks have launched related gateway products to help enterprises monitor employee usage and set spending limits. Factory, an AI software company invested in by Nvidia, also released a model router this month, aiming to automatically assign low-complexity tasks to lower-cost models.

Microsoft CEO Satya Nadella recently stated publicly that AI models should function like replaceable goods, rather than concentrating value in the hands of a few model providers. This statement aligns with the current cost-reduction demands of enterprise customers.

Microsoft emphasizes controllable costs when launching new products.

In response to customers' budget constraints, Microsoft this week revealed the pricing strategy for its new product, Copilot Cowork. This product primarily relies on the Anthropic model to automate more complex, multi-step tasks within Office 365.

Its pricing model is a "license fee plus pay-as-you-go" system. Users first need to purchase a 365 Copilot license starting at $30 per month, and then pay additional fees based on the actual usage of Copilot Cowork. This is similar to the billing method Anthropic previously offered to enterprise customers.

Microsoft also emphasized that customers can set usage limits for Copilot Cowork and plans to support switching between OpenAI, Microsoft's own models, or other models within the product to reduce costs. The report also stated that Microsoft is testing solutions that allow replacing the Anthropic model with an open-source model in certain scenarios.

This indicates that the new competitive focus in the enterprise software market is shifting from simply competing on model capabilities to who can ensure effectiveness while keeping AI costs within an acceptable range for enterprises.