Companies Rein In AI Spending as Workplace Costs Rise
AI spending is moving from a workplace experiment to a budget problem as companies place tighter controls on employee use of generative AI tools.
After more than a year of encouraging workers to test AI assistants for writing, coding, research, customer service, and internal operations, finance and technology leaders are now asking a sharper question. Which tasks are worth the cost?
The shift comes as AI adoption expands across corporate departments. McKinsey’s latest global AI survey found that 71% of respondents said their organizations regularly use generative AI in at least one business function. Stanford’s 2025 AI Index also reported that 78% of organizations used AI in 2024, up from 55% a year earlier.
That wider use is now showing up in corporate technology budgets. Gartner forecasts worldwide AI spending will reach $2.52 trillion in 2026, a 44% increase from the prior year. Boston Consulting Group reported that companies expect AI spending to rise from about 0.8% of revenue in 2025 to roughly 1.7% in 2026.
The result is a new phase of workplace AI adoption. Companies are not walking away from AI. They are beginning to ration it.
AI Spending Moves From Pilot Mode to Budget Discipline
Many employers initially treated generative AI as a low friction productivity tool. Employees were encouraged to test chatbots, coding assistants, document summarizers, meeting tools, and research platforms. The early goal was simple. Find out where AI could save time.
That approach helped adoption spread quickly, but it also made spending harder to track. A single prompt may appear inexpensive, but thousands of workers using premium AI services throughout the day can create a much larger expense.
Companies are now reviewing who has access to higher cost models, which departments use them most often, and whether routine tasks justify the bill. Some organizations are requiring approval for advanced AI tools. Others are consolidating software subscriptions or routing employees toward approved platforms with centralized controls.
The reassessment has become more urgent because the cost of AI often extends beyond a monthly license. Enterprise AI can involve cloud infrastructure, data storage, cybersecurity reviews, system integration, employee training, governance, and compliance work.
That means AI is no longer a scattered software expense. It is becoming a recurring operating cost that must be measured against productivity gains.
Token Bills Turn Small Prompts Into Large Expenses
One reason AI costs can rise quickly is token based billing. Tokens are the units of text that AI systems process when users submit prompts and receive responses. Longer documents, repeated revisions, code generation, research tasks, and agent based workflows can all increase token use.
That model has created new budget pressure for companies that scaled AI tools before fully understanding usage patterns.
Uber reportedly exhausted its 2026 AI coding tools budget in four months after encouraging employees to use the technology. The case drew attention because it showed how fast AI usage can outpace budget assumptions when adoption is pushed across technical teams.
Legal AI startup Harvey also offered a visible example of the scale involved. Its chief executive said the company used about 1 trillion tokens in January and was on pace to use 12 trillion to 13 trillion tokens in May, according to Business Insider.
Those figures show why companies are asking employees to match the tool to the task. A complex legal review, software debugging assignment, or large data analysis project may warrant advanced AI. A short email, basic summary, or simple formatting request may not.
Companies Sort High Value AI From Routine Tasks
The new controls are not aimed at stopping AI use. They are aimed at separating high value work from low value usage.
Software development remains one of the clearest areas where companies see practical value. Coding assistants can help generate boilerplate code, identify errors, write documentation, and support testing. Customer service teams are also using AI for automated responses, internal knowledge retrieval, and support documentation.
Other use cases are receiving more scrutiny. Routine email drafting, simple document summaries, basic internet style research, and minor editing tasks are increasingly being reviewed as areas where premium AI may not always be needed.
This is changing workplace guidance. Employees may be asked to use AI for complex reasoning, technical work, or high volume content tasks, while completing simple administrative work without calling on advanced models.
The difference matters because AI agents can consume far more tokens than basic chatbot interactions. A recent academic paper on agentic coding tasks found that those tasks can consume roughly 1,000 times more tokens than code reasoning or code chat. The same study found that runs on the same task could vary by as much as 30 times in total token use.
That variability makes budgeting difficult. A company may approve an AI workflow expecting one level of cost, then discover that repeated prompts, long context windows, and automated agent steps produce a much larger bill.
Finance Teams Push for Proof Before Wider Rollouts
The pressure to control AI spending is also tied to uneven returns.
McKinsey reported that more than 80% of respondents said their organizations were not yet seeing a tangible enterprise level EBIT impact from generative AI. MIT NANDA’s 2025 report found that most enterprise generative AI pilots had not produced measurable profit and loss impact, while a small group of integrated pilots produced significant value.
Those findings do not mean AI is failing inside companies. They show that adoption alone does not automatically produce measurable financial results.
This distinction is shaping how executives approve AI budgets. Finance teams are asking for clearer reporting on usage, output, cost per department, and measurable productivity gains. Technology leaders are reviewing whether teams are using multiple overlapping AI subscriptions. Procurement teams are looking for enterprise agreements that improve oversight and reduce duplicated spending.
Some companies are also testing cheaper models for lower risk tasks while reserving advanced models for complex work. Reuters reported that companies including Siemens, Renault, Orange, and ChapsVision use a mix of U.S., Chinese, and European models to avoid depending on one provider. Cost control has become part of that wider provider strategy.