Walmart, Uber, Amazon: Rising AI costs are forcing companies to put limits on employee usage

For the past two years, companies have pushed employees to use artificial intelligence wherever possible. AI assistants have been promoted as productivity boosters capable of writing code, analysing data, creating presentations and automating repetitive tasks. The message from management was clear: use more AI.

Now, a growing number of companies are discovering that widespread AI adoption comes with a hefty price tag.

As usage rises, so do the costs associated with running advanced AI models, many of which operate on token-based pricing systems where every prompt, response and interaction carries a financial cost. The result is a growing trend among major employers: encouraging AI use while simultaneously placing limits on how much employees can consume.

More from Tech

Walmart, Uber, Amazon and Microsoft have all recently introduced measures aimed at controlling AI-related expenses, signalling that the industry's enthusiasm for AI is increasingly being tempered by economic realities.

Walmart limits AI use as demand drives up costs

The latest company to tighten controls is Walmart.

According to Bloomberg, the retail giant has begun restricting usage of its internal AI assistant, known as Code Puppy. The tool is designed to help employees perform workplace tasks such as creating presentations, filling out spreadsheets and handling other routine business activities.

Previously, workers reportedly had unrestricted access to the platform. That has now changed.

Employees are being assigned token limits, effectively placing a cap on how much they can use the AI system. Token-based systems measure the amount of data processed by AI models, meaning higher usage translates directly into higher costs.

A Walmart spokesperson told Bloomberg that the company wants employees to use AI in ways that create meaningful value and ensure the right tools are being applied to the right tasks.

The retailer reportedly provides access to multiple AI platforms, including Anthropic's Claude and OpenAI's ChatGPT, alongside its own internal systems. However, growing demand for Code Puppy appears to have contributed to rising operational costs, prompting the move towards stricter controls.

Uber introduces spending limits after exhausting its AI budget

Uber has encountered similar challenges. The ride-hailing company spent much of the past year encouraging employees to integrate AI into their daily workflows. Internal rankings reportedly tracked AI usage, and staff were encouraged to experiment extensively with new tools.

That enthusiasm came at a cost.

Earlier this year, Chief Technology Officer Praveen Neppalli Naga revealed that Uber had consumed its entire annual AI budget within the first few months of the year. According to Bloomberg, the company subsequently introduced spending limits for AI coding tools.

Under the new rules, employees receive a monthly allowance of $1,500 per AI coding platform. The restrictions focus primarily on agentic coding tools capable of generating, reviewing and modifying software code with minimal human oversight.

The spending cap applies separately to each platform, and employees requiring additional access can seek approval for higher limits.

Uber has also introduced dashboards allowing workers to monitor their AI spending in real time.

The move is particularly notable given how deeply AI has become embedded within the company. Chief executive Dara Khosrowshahi recently said AI agents now generate roughly 10 per cent of Uber's submitted code. Beyond engineering teams, departments such as legal and marketing have also embraced AI tools.

Yet executives have acknowledged that measuring the actual business return on this investment remains difficult, even as internal productivity metrics improve.

Amazon shuts internal AI leaderboard after misuse concerns

Amazon's experience highlights another challenge associated with large-scale AI adoption: incentives.

According to the Financial Times, the company recently discontinued an internal dashboard called Kirorank, which ranked employees based on their activity within Amazon's Kiro developer platform.

The system was originally designed to encourage awareness and adoption of AI tools among developers. Instead, it reportedly produced unintended consequences.

Some employees allegedly began increasing their AI activity primarily to improve their rankings, a practice that became known internally as "tokenmaxxing". The behaviour led to higher infrastructure costs without necessarily generating meaningful productivity gains.

Amazon senior vice-president Dave Treadwell reportedly acknowledged the issue, telling employees, "Please don't use AI just for the sake of using AI."

The company later confirmed that the dashboard had been removed. Amazon said the tool was never formally approved and had since been deprecated.

Microsoft, meanwhile, has reportedly restricted some employees' access to Anthropic's Claude Code and encouraged engineers to migrate towards internal alternatives such as GitHub Copilot CLI. While the company has not explicitly linked the move to cost concerns, reports suggest rising token expenses may be playing a role.

Together, these developments reveal an emerging reality of the AI era. While companies remain committed to deploying AI across their organisations, unlimited access is proving difficult to sustain. The challenge is no longer convincing employees to use AI, but ensuring that the productivity gains justify the growing bill.