Microsoft is taking a closer look at how its employees use artificial intelligence after one employee reportedly generated $28,000 worth of AI usage in just 28 days.
The unusually high bill has exposed a growing problem for companies that have encouraged employees to use generative AI aggressively: while artificial intelligence can significantly improve productivity, unrestricted access to expensive models can also create unpredictable costs.
Microsoft’s response has been to introduce more detailed tracking of AI consumption, including token budgets and internal visibility into individual usage. The company has also reportedly shifted its default internal AI workloads to GPT-5.6 Sol, aiming to get greater value from the tokens employees consume.
The incident comes at a time when Microsoft is among the world’s biggest corporate users and developers of artificial intelligence. The company has invested billions of dollars in AI infrastructure and has encouraged its workforce to incorporate AI tools into everyday work.
But the $28,000 figure demonstrates just how quickly those costs can grow when usage is not carefully managed.
A $28,000 AI bill raises questions
The employee, working in Microsoft’s Customer and Partner Solutions organisation, reportedly recorded approximately $28,000 in AI usage during a single 28-day period.
That figure stands far above the reported company-wide median of roughly $300 for the same period. Data voluntarily submitted by around 350 US employees showed significant differences between teams, with some AI-focused divisions recording considerably higher median usage.
The figures do not represent Microsoft’s entire workforce and were based on voluntary reporting. Nevertheless, the enormous difference between typical usage and the highest individual spending highlighted the difficulty of controlling AI costs across a large organisation.
Unlike traditional software subscriptions, AI services can involve usage-based costs. Every interaction with a model consumes computational resources, measured in tokens. Longer prompts, large files, extended conversations and complex reasoning tasks can all increase consumption.
As employees rely more heavily on AI for coding, research, writing, analysis and other tasks, those individual requests can add up quickly.
When AI usage becomes a competition
Microsoft’s internal tracking also revealed another unexpected problem: employees were reportedly competing over AI usage.
The practice, described as “tokenmaxxing”, involved employees deliberately consuming large quantities of AI tokens, sometimes through low-value or unnecessary queries, in an effort to increase their position on internal usage dashboards.
The phenomenon highlights an unusual side effect of making AI consumption visible.
What begins as a measurement tool can become a performance metric. Once employees see usage numbers displayed publicly or internally, those numbers can become something to compete over, even if high consumption does not necessarily translate into better work.
Microsoft’s leadership has reportedly pushed back against this behaviour, stressing that the goal should not be to maximise the number of tokens used but to maximise the business results produced with those tokens.
That distinction could become increasingly important as companies attempt to measure the return they receive from their AI investments.
Microsoft introduces tighter controls
The company has responded by introducing division-level AI token budgets and increasing visibility into individual spending.
The idea is not necessarily to prevent employees from using AI. Instead, Microsoft appears to be moving towards a system in which AI consumption can be monitored and managed more closely.
For managers, token tracking can help identify unusually high usage and determine whether that consumption is generating meaningful results.
For employees, it means that AI activity is becoming another measurable part of workplace technology use.
This represents a significant change from the early days of workplace AI adoption, when companies were primarily encouraging employees to experiment with tools such as chatbots and coding assistants.
The focus is now shifting from adoption to efficiency.
Why the model matters
Microsoft’s reported decision to make GPT-5.6 Sol the default model is another important part of the company’s strategy.
Not every task requires the most computationally intensive AI model available. A simple summary, routine piece of code or straightforward workplace question may not require the same level of processing as a complicated research or reasoning task.
Using an appropriate model for each task can therefore reduce costs while maintaining productivity.
Microsoft’s reported shift suggests that the company is thinking about AI spending in much the same way businesses have historically approached cloud computing.
The objective is not simply to use less technology. It is to make sure that the resources being consumed produce enough value to justify their cost.
A new phase of corporate AI
The episode reflects a broader change taking place across the technology industry.
For years, companies competed to demonstrate how quickly they could integrate AI into their workplaces. Employees were encouraged to experiment with generative AI, automate repetitive tasks and use increasingly sophisticated models.
Now businesses are beginning to confront the financial consequences of that adoption.
AI infrastructure is expensive, and increasingly powerful models can require significant computing resources. At large companies with tens or hundreds of thousands of employees, even modest individual usage can become a substantial expense when multiplied across the workforce.
That makes AI budgeting an emerging part of corporate financial management.
Microsoft’s experience shows that companies may need to establish clear rules around model selection, token consumption and acceptable use before AI spending gets out of control.
From more AI to better AI
The lesson from Microsoft’s $28,000 incident is not necessarily that employees should use less artificial intelligence.
Instead, it demonstrates that organisations need to understand how AI is being used and whether that usage is producing meaningful results.
A worker who spends thousands of dollars on AI but saves a company millions of dollars through automation could represent an excellent investment. Conversely, thousands of dollars spent generating unnecessary responses or chasing usage rankings would provide little value.
That is why Microsoft’s new approach appears to focus on efficiency rather than simply restricting access.
The next stage of enterprise AI may therefore be less about asking how many tokens employees can consume and more about determining how much value they can generate from every token.
For Microsoft, one extraordinary $28,000 bill has turned AI usage into a much more closely watched resource. And as businesses everywhere expand their dependence on generative AI, that kind of scrutiny could soon become normal across the corporate world.
