The adoption of artificial intelligence (AI) has become widespread, with both businesses and individuals increasingly integrating it into their daily routines. The proportion of individuals utilizing AI in their work has seen a dramatic increase in recent years. According to survey data, in 2023, 21% of US workers used AI, a figure that has since risen to 50% by 2026. Concurrently, job insecurity is escalating, and companies are expressing intentions to replace a significant number of their workforce. Now, we have statements from senior executives, such as a director at NVIDIA, who believes that AI is more expensive than human workers, asserting that companies are failing to account for the true cost of tokens.
When generative AI first emerged a few years ago, some companies swiftly embraced it, laying off their human employees and adopting artificial intelligence in their place. At that time, AI was prone to numerous errors and lacked the sophistication it possesses today. It’s likely that many of these companies came to regret their decision, but it may have been too late, and their attempts to cut costs ultimately yielded poorer results.
NVIDIA Executive, Uber CTO, and Swan AI CEO Agree: AI Can Be Costlier Than Employees, But Businesses Ignore Additional Expenses
The primary motivation for most companies seeking to implement AI is cost savings. When comparing the subscription fees for AI services to employee salaries, a substantial difference is apparent. However, the reality is more complex than simply stating that a monthly fee of €100 or €200 will ensure AI operates 24/7. The true cost is never the same when the price of tokens is factored in. Furthermore, token costs can vary significantly between different AI models and agents, depending on their complexity.
As stated by Bryan Catanzaro, Vice President of Deep Learning at NVIDIA, there are instances where the cost of AI surpasses the expense of employing human workers, a situation he has observed within his own team at NVIDIA. Praveen Naga, the Chief Technology Officer at Uber, shares a similar perspective, having firsthand experience with AI becoming unmanageable and incurring expenses far exceeding initial projections. He notes that in his organization, 11% of real-time code updates are performed by AI agents.
The Underrated Cost of Tokens: Escalating to Thousands, Even Tens of Thousands of Dollars Monthly for Small Teams
Another pertinent example comes from Amos Bar-Joseph of Swan AI, who reported on LinkedIn spending $113,000 on Anthropic Claude with a team of just 4 people, equating to $28,000 per person per month. This is not comparable to a ChatGPT Pro subscription costing around €20 per month. Instead, it refers to AI models like Claude Code or GitHub Copilot where payment is based on token usage. The specific functionalities utilized also have a significant impact. For instance, a recently showcased feature called Bugcrawl consumed substantially more tokens but offered the benefit of analyzing entire repositories using Claude AI to identify errors and bugs with proposed solutions.
Additionally, consider companies at the forefront of the AI sector, such as NVIDIA, where employees are compensated with tokens as ‘part of their salary.’ Given their freedom to use these tokens as they wish, it wouldn’t be surprising if they established businesses utilizing any surplus. This highlights that the AI sector requires a focus on long-term results and investments, which often clashes with the short-term profit and gain objectives that drive many companies to adopt AI.
