Is AI Too Expensive to Replace Humans?

According to Stanford’s latest AI Index Report, US private AI investment reached $285.9 billion in 2025. 1 Businesses are continuing to pour millions of dollars into AI initiatives. Meta alone is projected to spend between $125 and $145 billion on it in 2026.

Many companies believe that AI can automate repetitive and lower-level tasks normally reserved for entry-level staff, while making the remaining employees more productive. For instance, Meta has been relocating employees to new AI-focused roles and encouraging them to use the technology in their workflows, even if this makes them less productive in the short term.

One of the biggest potential benefits in the eyes of employers is the potential for reduced costs. Businesses believe that they can streamline their staff by replacing employees with AI. 2026 has seen hundreds of companies lay off staff, and many of the CEOs have cited AI as the reason.

How do workers feel about AI?

Bosses want their employees to adopt this “AI-first” mindset, but many are reluctant to do so. A 2025 survey from the Pew Research Center found that around half (52%) of workers were “worried” about how AI would be used in the workplace. 2

In an early 2026 survey from AI consulting firm Section, those concerns were still present. 68% of people reported feeling “anxious or overwhelmed” about AI.

For many workers, AI poses a threat to their jobs. Freelancers are already claiming that they’re being displaced by technology. In an early-2026 SideHustles survey, more than 1 in 3 independent workers said they’d lost work to AI.

Employees aren’t uneasy about AI just because they’re scared of losing their jobs. Many workers feel that it actually makes their jobs harder. A software engineer working for Amazon told The Guardian that the company’s push for employees to use AI has made things “worse now.”

He argued that while AI tools may save developers time, the end result is usually much worse. “This pressure to use [AI] has resulted in worse quality code, but also just more work for everyone,” he said.

While many employers think that AI will help them cut back on costs, recent data may suggest otherwise. Microsoft recently cancelled the majority of its Claude Code licenses only a few months after they told employees to embrace the technology.

In some cases, it has become more expensive to use AI than to pay human workers.

The dangers of “tokenmaxxing”

To use AI systems like Claude Code, companies need to use “tokens.” Each action costs a set number of tokens, and the more complicated a task is, the more tokens are needed.

Many firms, including Microsoft, Amazon, and Meta, encouraged workers to practice “tokenmaxxing”—in other words, using AI as much as possible. Some companies even gamified the process. The New York Times reported that Meta and OpenAI created “token leaderboards” and rewarded heavy users of AI. As a result, their engineers were burning through tokens and racking up costs.

Microsoft’s push to encourage employees to code with Anthropic’s Claude Code may have proven too popular. The Verge reported that engineers in Microsoft’s Experiences + Devices team, which includes Microsoft 365, Outlook, Microsoft Teams, and Surface, are being told to scale back their use of AI by the end of June due to unexpectedly high costs.

It isn’t just Microsoft that has had to curb its use of AI. In April, Uber’s CTO Praveen Neppalli Naga told The Information that they’d already used their entire 2026 AI coding tools budget in just 4 months.

Technology blog The Next Web reported that Uber’s engineers were spending between $500 and $2,000 a month on Claude Code.

Nvidia’s vice president of applied deep learning, Bryan Catanzaro, had a similar take in an interview with Axios. “For my team, the cost of computing is far beyond the costs of the employees,” he said.

Estimated costs of heavy AI use

Anthropic recently updated its estimate of how much using Claude Code costs. The company quietly increased the average cost for each developer to $13 per day and $150–$250 per month. 3

According to Indeed, the average annual developer salary in the US is $102,110. Based on Anthropic’s estimates, using Claude Code would cost companies an extra $1,800–$3,000 a year per developer.

While that figure may not sound like much, compounded over a team of developers, it adds up if the company isn’t able to leverage AI to save greater costs elsewhere.

AI vs. human efficiency

Another reason AI implementation can become expensive is that the productivity gains aren’t always immediate (and that’s assuming they ever show up at all).

As mentioned, many employees feel that using AI tools actually results in more work in the end.

A February 2026 study from Harvard Business Review backed up these claims. Researchers found that AI tools intensified workloads for employees and managers. For instance, engineers spent more time reviewing, correcting, and guiding AI-generated or AI-assisted work produced by colleagues.

Costs also increase when workers have to repeatedly refine AI outputs. In a recent post on the subreddit r/ArtificialIntelligence, one user claimed that it takes a lot of “wasted tokens” to get reliable results from AI. “For every output token of the final result, we spend 10 times (actually more than that) more in the backend to make sure that the data and everything else are right,” they said.

Things could change in the near future. AI tools like Claude Code are expensive to use primarily because of token pricing.

However, as consumption increases, the price of tokens is likely to fall. A recent report from Gartner predicted that token costs will decrease by 95% through 2030.

Cheaper tokens may not necessarily translate to reduced AI costs. Gartner notes that agentic AI models require far more tokens than standard models.

Standard AI models are reactive, working off a user’s input and usually only specializing in a single task. On the other hand, agentic AI models act autonomously and are able to generalize from training data and adapt to new situations. 4

As an example, an agentic AI model could access a customer’s account history, issue a refund, update the inventory list, and send a personalized email, all without human supervision.

As more businesses integrate AI into their workflows, especially for more complicated tasks, it’s possible that increased consumption will outpace the dropping price of individual tokens.

“Chief Product Officers (CPOs) should not confuse the deflation of commodity tokens with the democratization of frontier reasoning,” said Will Sommer, the Senior Director Analyst at Gartner. Sommer argued that while AI tools move toward a “near-zero cost,” the real challenge will be having enough computing power and efficient systems to support them.

As companies confront the true cost of AI tools, the once-firm narrative that technology is going to replace human workers feels shakier.

For now, having employees tokenmaxxing and experimenting with AI tools isn’t particularly cost-effective.

However, even if AI tools remain expensive, companies may still view them as worthwhile if they ultimately allow smaller teams to handle larger workloads, even if that day doesn’t come for years.

Many companies seem determined to stick to the “AI-first” mindset, and in the short term, may cut staff instead of scaling back AI integration. Whether or not that pays off in the long run remains to be seen.