April 29, 2026 · 3 min read
AI is turning out to be more expensive than human labor
Why more teams are discovering that AI compute costs can outpace payroll, and what that means for real ROI decisions.
I have been hearing for years that AI would slash operational costs and replace expensive talent.
Reality is starting to look quite different.
When compute bills outrun expectations
Take Uber as a prime example. The company pushed engineers to use AI coding tools, especially Claude, to speed up development, and teams adopted them quickly.
Despite a massive $3.4 billion R&D budget, Uber reportedly burned through its AI compute allocation early this year.
Not an isolated signal
Uber is not alone. Brian Catanzaro, VP of Applied Deep Learning at NVIDIA, recently said that for his team, compute costs now significantly exceed the cost of human workers.
In some scenarios, hiring people is literally cheaper than running AI models.
From hype to unit economics
There are still bright spots. Uber says around 11% of updates in its ecosystem are now handled by AI agents without human intervention, and leadership still believes in autonomous digital engineers.
But rising bills are shifting the discussion from “AI will replace everyone” to “What is the actual ROI, and when does a human make more economic sense than a model?”
A more honest phase of adoption
At the same time, OpenAI and its investors are positioning Codex as a more efficient, cost-effective alternative with better token optimization.
It feels like AI adoption is moving into a more mature and more honest phase.
Final take
The central question is no longer whether AI can produce output. It is whether that output is economically superior once full compute costs are included.
Links: