O "Grande Roll-back": Por que a IA não é a bala de prata que prometeram? A cruel realidade matemática de que a IA custa muito mais que o humano.
Dev.to / 6/3/2026
💬 OpinionSignals & Early TrendsIdeas & Deep AnalysisIndustry & Market Moves
Key Points
- The article argues that the long-promised myth of AI drastically lowering technology operating costs is colliding with real compute economics, noting that AI model compute costs in major firms can now exceed payroll expenses.
- It cites the Uber CTO’s admission that Uber’s annual AI budget was consumed in just four months, attributing the blowout to uncontrolled usage of Claude Code and “tokenmaxxing” behavior without governance.
- The piece highlights that only about 18%–25% of AI projects manage to beat the cost of capital, because development, integration, ongoing maintenance, and token costs create a high barrier to sustainable ROI.
- It points to massive planned spending by Big Tech—US$725B this year (up 92%)—while questioning whether token-driven AI infrastructure costs can remain sustainable and justify the investment.
- It adds supporting signals: a large share of GitHub code is reportedly AI-generated, and Meta’s projected 2026 AI infrastructure spend (US$145B) is rising due to underestimated computational capacity needs, triggering investor concerns and likely layoffs.
Continue reading this article on the original site.
Read original →Related Articles
Black Hat USA
AI Business
Microsoft created the mini Surface dev box that Qualcomm couldn’t
The Verge
Microsoft debuts Surface RTX Spark Dev Box to run large AI models without cloud costs
VentureBeat
Code Is Cheap. Engineering Judgement Is Now the Scarce Resource
Towards Data Science
Microsoft launches MXC, an OS-level sandbox for AI agents, with OpenAI and Nvidia already on board
VentureBeat
