NextFin News - The cost of artificial intelligence has crossed a historic threshold, with major enterprises now reporting that their spending on computing power and AI tokens has eclipsed the total payroll for the teams managing those systems. According to data released by Gartner on April 22, worldwide IT spending is projected to reach $6.31 trillion in 2026, a 13.5% surge from the previous year. This acceleration is driven almost entirely by the "sustained momentum" of AI infrastructure and cloud services, marking a fundamental shift in how corporate capital is allocated between human talent and digital labor.
The financial strain of this transition is already hitting the C-suite. Uber’s chief technology officer has reportedly exhausted the company’s entire 2026 AI budget in just four months, primarily due to the unforeseen volume of token costs associated with autonomous coding tools. This is not an isolated incident. Bryan Catanzaro, vice president of applied deep learning at Nvidia, told Axios that for his specific teams, the cost of compute now "far exceeds" the cost of the employees themselves. While Nvidia sits at the center of the AI boom, the trend is spreading to software-heavy firms where the price of "intelligence" is becoming a larger line item than the price of the people who prompt it.
Amos Bar-Joseph, CEO of Swan AI, recently highlighted this shift by publicizing a massive bill from Anthropic, the developer of the Claude AI models. Bar-Joseph framed the expense as a strategic choice, stating that his firm is building an "autonomous business" that scales with intelligence rather than headcount. This perspective, while gaining traction among venture-backed startups, remains a point of intense debate among institutional investors. The core question is whether these massive compute bills represent a temporary "build phase" or a permanent new cost of doing business that could erode margins if productivity gains fail to materialize.
The shift in spending is also forcing a radical change in how software is priced. According to a Goldman Sachs research note published this month, AI companies are moving away from traditional "per-seat" licensing—where a company pays based on the number of employees using a tool—toward "unit of labor" pricing. Goldman Sachs analysts, who recently met with 40 industry leaders, noted that software providers are increasingly selling tokens or specific productivity outcomes. This allows AI labs to capture a larger share of corporate budgets that were previously reserved for human salaries, effectively turning software expenses into a direct substitute for payroll.
However, this "compute-first" strategy carries significant risks. Brad Owens, vice president of digital labor strategy at Asymbl, cautioned that the market is entering a phase where the "true value" of a worker—human or digital—must be rigorously audited. Companies that slash hiring and training budgets to fund AI tokens may find themselves with expensive digital tools but no human expertise to steer them. Fortune recently reported that while 75% of knowledge workers now use AI, 60% have received no formal training, leading to "quiet costs" such as higher turnover and lost productivity that do not appear on a compute bill but weigh heavily on the bottom line.
The sustainability of this spending model will likely be tested in the upcoming quarterly earnings cycle. As U.S. President Trump’s administration continues to emphasize domestic industrial efficiency, shareholders are expected to demand clearer metrics on the return on AI investment (ROAI). If the cost of tokens continues to rise while productivity gains remain incremental, the current "flex" of high AI spending could quickly be reclassified as a liability. For now, the corporate world is betting that silicon is more scalable than soul, even if the invoice for the former is starting to look more expensive than the latter.
Explore more exclusive insights at nextfin.ai.
