Companies aren’t treating enterprise AI budgets like a contest to see who can burn the most tokens anymore. The conversation has shifted to control: keep token use from spiraling, set hard limits, and put guardrails around chatbots, coding assistants, and AI agents.
At the top end, those costs have reportedly reached $7,500 per employee each month. Agent-based workflows can drive the bill higher still, even while per-token prices keep dropping. Global spending is projected to hit $2.59 trillion in 2026, yet only 28% of finance leaders in a Deloitte survey said they see clear value. So the mood is changing. Companies are looking at budgets now, not chasing growth at any price.
You can already see it across the industry. AT&T has pulled back on GitHub Copilot. Meta has reduced its use of Anthropic and other models. Uber reportedly blew through its 2026 budget for coding tools by April, then put a $1,500 monthly cap per tool in place for employees.
Walmart is tightening control over its in-house agent. Amazon removed a usage leaderboard after employees started gaming it and pushing compute costs up. At Microsoft, some engineers were spending anywhere from $500 to $2,000 a month on Claude Code tokens.
If you’re in charge of rolling out AI at your company, keep an eye on this. The same shift shows up in the new controls from Microsoft, Databricks, and Nvidia-backed Factory, which routes simpler tasks to cheaper models.
And companies are adjusting in different ways. More of them are sending basic AI work to smaller or open-source systems, while Box says it never pushed heavy usage in the first place, and Databricks still gives engineers unlimited budgets.
