What Happens When AI Costs More Than Workers?

The promise of AI as a cheap replacement for human labor is being upended by soaring operational costs, with some companies seeing AI bills that far exceed employee salaries [1]. Nvidia executives have noted that AI computing costs are "far beyond the costs of the employees" [1]. A stark example comes from Swan AI, which reported a monthly AI bill of $113,000 for a four-person team [1]. While entry-level subscriptions can start around $20 a month, costs escalate quickly as companies adopt multiple tools. "A $30-a-month tool usually becomes six or seven of them, plus API overages, plus a workflow platform tying it together," said Scott Tobin, founder of Delegate [1]. Shifting to pay-per-data models can cause monthly costs to jump from $10,000 to $50,000 [1]. This cost dynamic challenges the core assumption behind technological unemployment, where labor-saving automation is expected to displace workers [7]. Instead, AI may be creating a new cost center. As Nils Gilman of the Berggruen Institute notes, "You may be getting more productivity, but you may not be able to get it out of fewer people" [1]. This aligns with historical debates where some economists argued technological change causes only a "temporary phase of maladjustment," not lasting mass unemployment [7]. The complexity of modern AI systems contributes to the expense. Many operate as "compound AI systems" or agents that pursue goals and use tools, requiring significant computational resources [2]. Furthermore, the industry's drive toward more capable artificial general intelligence (AGI), which aims to match or surpass human capabilities across cognitive tasks, represents a massive and costly research endeavor [3]. This long-term ambition, pursued by firms like OpenAI and Google, requires investment far beyond simple task automation [3]. The result is a shifting labor landscape. While some junior roles are automated, mid-level positions requiring judgment and management are becoming more valuable [1]. Companies must now balance the high cost of AI against human labor, managing teams where, as Gilman observes, "Everybody is becoming a middle manager of their team of bots" [1].