For years, corporate executives were promised a future in which artificial intelligence would handle the work of human employees around the clock, slash payroll expenses, and make businesses more profitable. Companies rushed to embrace that vision, with some even cutting jobs to accelerate their transition. However, an uncomfortable problem is now emerging: running AI systems can cost far more than many businesses anticipated.
From exhausted technology budgets to executives questioning their own AI strategies, the corporate enthusiasm surrounding artificial intelligence is facing a serious financial test. Some companies are discovering that their supposedly cheaper digital workforce comes with enormous bills, questionable productivity gains, and consequences that could reshape the future of employment.
Executives Are Discovering The True Cost Of AI Adoption
The financial reality of artificial intelligence is becoming increasingly difficult for corporate leaders to ignore. A growing number of businesses are discovering that running powerful AI systems can cost substantially more than they expected, particularly when employees begin using them across entire departments.
A survey conducted by accounting firm KPMG, involving 2,145 senior executives across 20 countries, revealed that 29% of respondents had no clear understanding of where their organization’s growing AI expenses were coming from. Another third acknowledged that their lack of understanding about AI economics was preventing successful workplace adoption.
The report authors explained: “As usage-based pricing models become more common, many organizations are still building the capabilities required to forecast, monitor, and manage AI spending effectively.” Their findings suggest that companies have been adopting increasingly expensive technology without necessarily developing the financial systems needed to manage it.
For businesses accustomed to predictable software subscriptions, usage-based pricing can be an unpleasant surprise. Every request, generated response, and complicated task can consume computing resources, meaning costs can rise rapidly when thousands of employees start using powerful AI models throughout their working day.
Why AI Bills Can Increase So Quickly
Many companies initially benefited from subscription arrangements that allowed employees to access AI services for relatively predictable monthly fees. However, providers are increasingly introducing pricing structures that reflect actual computing consumption, exposing businesses to expenses that were previously less visible.
Large language models require substantial computing power to process requests and generate responses. More complicated tasks can consume considerably more resources, especially when employees repeatedly ask models to analyze documents, write code, process information, or perform lengthy workflows.
The problem becomes particularly serious when organizations encourage widespread adoption without establishing usage limits. Executives who expected AI to reduce operational expenses are now discovering that encouraging employees to use expensive models for virtually everything can produce the opposite financial result.
Even Major Technology Companies Are Struggling With AI Expenses
If anyone should understand the economics of artificial intelligence, it would be the companies building and selling the technology. Yet even major technology businesses are experiencing unexpected expenses as their employees increasingly incorporate AI into everyday operations.
Uber offers one particularly striking example. The company reportedly exhausted its entire 2026 AI coding budget within four months after encouraging engineers to adopt Anthropic’s Claude Code, an AI assistant designed to help developers write and manage software.
By March, approximately 84% of Uber’s engineers had adopted the coding assistant, with around 70% of committed code reportedly coming from AI. The adoption numbers were impressive, but the financial results proved considerably more complicated than executives had anticipated.
Uber’s chief technology officer, Praveen Neppalli Naga, told The Information that the company had exhausted its planned budget much earlier than expected. He described the situation by saying: “I’m back to the drawing board because the budget I thought I would need is blown away already.”
The problem was not simply that employees were using AI. The company had actively encouraged adoption, including through initiatives designed to motivate engineers to incorporate the technology into their work. Once usage accelerated, however, the financial consequences became difficult to ignore.
Microsoft has also reconsidered its use of AI coding assistants after costs became difficult to justify. Reports described the company moving away from widespread Claude Code licensing in favor of alternative tools, demonstrating that even businesses with enormous technology budgets are paying closer attention to AI spending.
Nvidia Has Admitted That AI Can Cost More Than Employees
Perhaps the most revealing observation came from Bryan Catanzaro, Nvidia’s vice president of applied deep learning. Nvidia is one of the most important companies supplying the hardware used to power modern AI systems, making his comments particularly significant.
“For my team, the cost of compute is far beyond the costs of the employees,” Catanzaro told Axios in April. His statement directly challenges the assumption that artificial intelligence automatically provides a cheaper alternative to human labor.
The comparison becomes especially interesting when considering Nvidia CEO Jensen Huang’s position on workplace AI adoption. Huang has encouraged companies to invest heavily in AI tokens, suggesting that a $500,000 engineer should consume at least $250,000 worth of AI tokens annually.
The message from one of the industry’s most influential executives is clear: businesses should be spending heavily on artificial intelligence. However, Catanzaro’s comments reveal that the actual costs of running these systems can already exceed employee expenses in some circumstances.
Companies Are Cutting Thousands Of Jobs While Spending Billions On AI
The growing AI bill becomes even more controversial when compared with what is happening to workers. Technology companies have announced enormous workforce reductions while simultaneously committing unprecedented amounts of money to artificial intelligence infrastructure.
According to figures cited in the supplied reports, Big Tech companies announced approximately $740 billion in capital expenditure for 2026, representing a 69% increase from 2025. The spending reflects the industry’s determination to expand data centers, acquire advanced computing equipment, and develop increasingly powerful AI systems.
At the same time, more than 115,000 technology workers have reportedly lost their jobs during 2026. The figures illustrate the growing tension between corporate investment in artificial intelligence and the financial security of the people whose work these technologies are expected to transform.
Several prominent businesses have announced substantial workforce reductions this year. Their decisions have frequently been associated with operational efficiency, restructuring, and redirecting resources toward artificial intelligence.
Some of the most significant reported cuts include:
- Meta: The company announced plans to eliminate approximately 8,000 positions while also reducing its hiring plans.
- Wix: The website-building company reportedly cut around one-fifth of its workforce as it shifted its operations toward AI.
- Block: The financial technology company reduced its headcount by approximately half as part of a major restructuring.
- Atlassian: The software company announced 1,600 job cuts while reorganizing its business.
- SentinelOne: The cybersecurity company reduced its workforce by 8% while redirecting resources toward artificial intelligence.
These decisions raise an uncomfortable question about the financial logic behind replacing employees with AI. Businesses may reduce payroll expenses through layoffs, but those savings become less impressive if the technology they introduce creates substantial new operating costs.
A 2024 MIT study provides another important perspective. Researchers examining the technical requirements of AI automation found that replacing humans was economically viable in only around 23% of the roles they studied where computer vision was a primary component of the work.
In the remaining 77% of cases, continuing to employ humans was cheaper. Although the research focused on a particular category of work, it demonstrates why businesses should be careful about assuming that technological capability automatically translates into financial savings.
The AI Spending Frenzy Has Created A Strange Workplace Culture
Part of the problem appears to come from the way companies have encouraged employees to use artificial intelligence. Rather than identifying specific tasks where AI could deliver measurable improvements, some organizations have treated adoption itself as evidence of progress.
This has contributed to a workplace phenomenon sometimes described as tokenmaxxing. AI models process information using tokens, which are units of text or other data, and many AI services charge businesses according to how many tokens their employees consume.
Consequently, widespread usage can translate into substantial expenses. The more employees interact with expensive models, the more computing resources their organizations must purchase.
Amazon reportedly introduced an internal leaderboard called KiroRank to track AI usage among engineering teams. The system was eventually taken down after employees began gaming the rankings by consuming tokens on meaningless tasks simply to improve their positions.
Meta reportedly developed a similar tracking system called Claudeonomics. These initiatives reveal how enthusiasm surrounding AI adoption can sometimes create incentives that have little connection to actual business performance.
When employees are rewarded for using more AI, consumption can become the objective rather than producing better results. Companies may end up paying for thousands of unnecessary requests while believing their workforce is becoming more technologically advanced.
More AI Usage Does Not Always Mean More Productivity
The problem extends beyond expensive subscriptions and unnecessary token consumption. Businesses must also determine whether AI-generated work actually improves productivity once the time required to review and correct it is considered.
Uber’s experience offers a useful example. Although AI reportedly generated a substantial proportion of its committed code, the company’s leadership acknowledged that token usage did not directly correspond with useful features being delivered to customers.
Research from Faros AI has also raised questions about the quality of AI-assisted software development. A study cited in the supplied reference material found that code churn, measured by lines of code deleted compared with lines added, increased by more than 800% under high AI adoption.
That finding suggests that increased AI-generated output can create additional work when developers must review, replace, or remove unnecessary code. Businesses that measure success through the volume of generated material could therefore overlook the time and money spent correcting it.
The financial consequences become even more significant when companies rely on their most expensive models for relatively simple tasks. According to the supplied reference material, approximately 95% of enterprise AI usage still runs through costly frontier models, even when less sophisticated alternatives could potentially handle the work.
AI Is Also Creating An Unexpected Crisis Of Workplace Trust
The financial costs are only part of the corporate AI problem. Another study suggests that employees who use AI-generated work could face a surprising disadvantage when it comes to gaining their managers’ trust.
Research published in The Leadership Quarterly, led by Hebrew University of Jerusalem PhD student Samuel Meyers, examined how supervisors judged employees who used AI tools. The experiments investigated whether simply knowing that an employee had used AI influenced a manager’s willingness to trust their work.
The researchers found that employees who did not use AI had a 68% chance of earning their supervisors’ trust. When AI was involved, that figure dropped to just 24%.
Coauthor Yochanan Bigman explained the findings in a statement: “AI is changing not only how we work, but also how we judge the people we work with.”
He added that employees could use AI to produce better work more efficiently while still losing their managers’ trust simply because the technology was involved. The findings reveal a complicated workplace contradiction that companies may struggle to resolve.
Some organizations actively encourage employees to use AI, while other executives are becoming increasingly frustrated with generic AI-generated communications and poorly reviewed output. One CEO reportedly threatened to “fire the next person” who sent unedited ChatGPT-generated material.
Another executive introduced restrictions after receiving an influx of AI-generated Slack messages. These examples demonstrate how quickly corporate enthusiasm can turn into frustration when employees begin using AI without considering whether the resulting work is actually useful.
Workers can consequently find themselves caught between conflicting expectations. Management may demand greater AI adoption while simultaneously questioning the judgment, originality, and professional reliability of employees who depend heavily on these tools.
Researchers have also warned that this tension could encourage shadow AI, where employees secretly use artificial intelligence without informing their managers. Such behavior could create additional problems for organizations attempting to establish clear workplace policies.
The Business Model Behind AI Is Facing A Serious Test
For years, many AI companies attracted customers through relatively inexpensive subscription arrangements. Those prices allowed businesses to experiment with powerful tools without immediately confronting the full cost of operating increasingly demanding systems.
However, the economics become more complicated when customers start using these services constantly. Providers must pay for computing infrastructure, electricity, specialized hardware, and the technical resources required to keep their models running.
The supplied reference material describes Anthropic and GitHub moving enterprise customers toward pricing arrangements more closely connected to actual usage. This shift could make AI considerably more expensive for businesses with employees who rely heavily on advanced models.
There is another complication. Falling prices per token do not necessarily mean falling overall expenses because consumption can increase faster than individual prices decline.
Gartner has forecast that running the largest AI models could become nearly 90% cheaper by 2030. That sounds encouraging, but businesses will still need to account for the possibility that employees and automated agents consume substantially more computing resources as AI becomes increasingly integrated into everyday operations.
The profitability challenge is particularly significant for companies operating expensive infrastructure. OpenAI’s financial projections, cited in the supplied reference material, indicate substantial expected losses before the company reaches its projected profitability timeline.
This does not mean artificial intelligence has no commercial future. It does, however, expose the difficulty of building a profitable business when the cost of providing the technology can exceed the revenue generated from customers.
Corporate Leaders May Have To Rethink Their AI Strategy
Despite the growing financial problems, artificial intelligence is unlikely to disappear from the workplace. The technology can already assist with coding, research, customer service, document processing, and numerous other tasks that businesses previously handled entirely through human employees.
However, companies may need to become considerably more selective about how they deploy it. Instead of encouraging employees to use the most expensive models for every available task, organizations could benefit from matching different AI systems to specific requirements.
Keith Lee, an AI and finance professor at the Swiss Institute of Artificial Intelligence’s Gordon School of Business, described the current situation as a short-term mismatch between AI expenses and the economic benefits companies hope to achieve.
He also emphasized that AI must become cheaper and more predictable before businesses can confidently rely on it at scale. That means companies must evaluate reliability, computing expenses, human oversight, and measurable productivity rather than simply counting how many employees have adopted the technology.
A business might save money by eliminating several positions, but those savings can quickly become less impressive if it must spend heavily on AI subscriptions, computing infrastructure, employee oversight, and correcting generated mistakes.
The companies that eventually benefit most from artificial intelligence may be those that identify exactly where the technology delivers measurable value. Spending billions to demonstrate enthusiasm for AI is considerably easier than proving that the investment produces sustainable profits.
For now, corporate leaders are facing an awkward reality. The promise of replacing expensive human workers with cheaper digital alternatives has collided with the actual cost of running those alternatives, leaving businesses with financial questions they should have asked before launching their AI transformation plans.
And as more companies confront their monthly invoices, the biggest test for the AI industry may no longer be how many jobs its technology can replace. It may be whether the businesses paying for it can afford to keep using it.