$300/Day for AI? Study Shows Costs Can Exceed Employees' Salaries

AI was pitched as a way to ease pressure, but new research suggests it often increases it.

In an eight-month study published by Harvard Business Review, UC-Berkeley researchers tracked how generative AI altered daily work.

The U.S. tech company with about 200 employees offered enterprise subscriptions to commercially available AI tools but did not require their use. Employees adopted them voluntarily.

As generative AI use spread, workers began stepping into tasks that once belonged to other roles.

Product managers wrote code, and researchers handled engineering work. Engineers then spent additional time reviewing AI-assisted output from colleagues.

Oversight expanded quietly across teams, and Harvard Business Review researchers found that their work didn't actually shrink, but instead, intensified.

Employees worked faster, handled a wider scope, and filled more of the day with small bursts of AI-assisted output.

The Price of AI Acceleration

How much it actually costs to deploy AI daily has now entered the conversation.

On the "All-In" podcast, investor Jason Calacanis said he quickly reached roughly $300 per day per Claude AI agent through API usage.

And this is true even when the system was operating at only a portion of its capacity.

At sustained usage, this level of spending adds up to around $100,000 annually per agent per month.

In the podcast, investment firm Social Capital CEO Chamath Palihapitiya said companies now need to ask themselves, "What is the token budget that we're willing to give our best devs?"

“[I]f you aggregate it across all people, you can clearly see a trend where you're like, ‘Well, hold on a second, now they need to be at least 2x as productive as another employee,’” Chamath explained.

“That is actively happening inside my business, because otherwise I'll run out of money.”

Entrepreneur and investor Mark Cuban later described the token-cost argument as a serious counterpoint to claims that AI will automatically replace workers.

At the infrastructure level, usage is accelerating at historic speed.

During its Q4 2025 earnings call, Google said its Gemini models now process more than 10 billion tokens per minute through direct API usage, up from 7 billion the previous quarter.

On a year-over-year basis, token volume increased 52x, from about 8.3 trillion tokens per month in late 2024 to an annualized run rate above 430 trillion.

Nearly 350 customers now process more than 100 billion tokens each. Gemini Enterprise has also sold more than 8 million paid seats within four months of launch.

Once embedded into daily operations, organizations would have to commit to rising AI token spend and deeper vendor dependence.

Here is how companies can slow down and apply financial and operational discipline to AI usage:

• Define scope before tools expand it. Set clear limits on how AI-enabled tasks should and should not widen job responsibilities.
• Measure cost against output. Tie token spend directly to revenue impact, efficiency gains, or risk reduction so usage maps to business value.
• Protect cognitive bandwidth. Build structured pauses and sequencing norms into workflows so speed doesn't quietly erode judgment.

Remember that lower serving costs at the model level can still coincide with rising oversight demands, higher expectations, and growing capital commitments.

Our Take: Is AI Actually Cheaper Than People?

We don't think this is a simple substitution equation.

The Harvard Business Review article shows that AI adoption can widen workloads and extend the day. And as usage grows, token bills grow with it.

Infrastructure spending across the industry is hitting levels that signal long-term commitment. 

We think that discipline is key. Those who define limits early will have a clearer answer to whether the math truly works in their favor.

In other news, Ring’s Super Bowl ad recently drew privacy backlash, highlighting how quickly public trust can erode when tech outpaces consumer comfort.

Executives evaluating automation and workforce strategy often consult advisory firms to assess long-term cost exposure and operational risk.

Explore the top AI consulting companies in our directory.