Companies Face Escalating AI Token Costs That Surpass Employee Salaries

Recent reports show several tech firms spending more on AI token usage than on the salaries of the engineers generating the requests. Engineers running multiple AI agents for code generation incur token fees that, in total, exceed payroll costs for their teams. The trend has grown as AI-driven development scales.

Futurism

AI agents now produce software code faster than humans, often running several agents concurrently. Each task consumes tokens billed to the company. The practice, dubbed “tokenmaxxing,” has led some users to incur monthly token bills over $150,000. Uber engineers have reportedly exhausted their 2026 AI budget on Claude Code. Meta has added AI usage metrics to employee performance reviews, further encouraging high token consumption.

Futurism

• Monthly token bills for power users can exceed $150,000.

Futurism

• Uber’s Claude Code usage depleted its projected 2026 AI budget.

Futurism

• Anthropic claims “pretty much 100 %” of its code is AI-generated; Google and Microsoft estimate roughly 25 % AI-generated code.

Futurism

• Nvidia reports compute costs for some teams surpass employee salary costs.

Futurism

Nvidia’s vice president, Bryan Catanzaro, said compute costs for his team “are far beyond the costs of the employees.” Anthropic’s head of Claude Code, Boris Cherny, asserted that most of the company’s code now comes from AI. Nvidia CEO Jensen Huang proposed allocating AI tokens worth about half an engineer’s salary as a recruiting incentive. Meta’s internal review now includes AI usage as a performance metric. An OpenAI investor noted that token-cost concerns could favor providers claiming more efficient models, such as Codex.

Futurism

Analysts caution that high token consumption may not yield productivity gains. Error-prone AI outputs have caused internal disruptions at Meta and Amazon. Studies cited suggest mandatory AI tools can make tasks harder for workers. The OpenAI investor’s claim that Codex uses tokens more efficiently than Claude underscores competitive worries about pricing and efficiency.

Futurism

Anthropic’s near-total AI-generated code claim conflicts with Google and Microsoft’s estimate of a 25 % share, indicating no industry-wide measurement standard. Detailed token-cost versus productivity data remain unavailable, limiting assessment of the true financial impact.

Futurism

• “For my team, the cost of compute is far beyond the costs of the employees,” — Bryan Catanzaro, vice president, Nvidia

Futurism

• “Pretty much 100 percent” — Boris Cherny, head of Claude Code, Anthropic

Futurism

• “I probably spend more than my salary on Claude,” — Max Linder, software engineer, Stockholm

Futurism

• “tokenmaxxing,” — Futurism, publication

Futurism

Nvidia plans to distribute AI tokens equal to half of engineers’ salaries, formalizing token budgeting. Anthropic has raised pricing, and OpenAI is promoting its models as more token-efficient. Companies are expected to tighten AI usage policies and monitor token spend as part of broader cost-control efforts.

Futurism