The Case Against Generative AI

Source: Where’s Your Ed At Date: 2025-09-29 URL: https://www.wheresyoured.at/the-case-against-generative-ai/

Summary

Ed’s comprehensive prosecution of the AI bubble: $500B+ invested with minimal profitable revenue; OpenAI needs $290B+ in additional funding while burning billions; NVIDIA funds neoclouds (CoreWeave, Lambda, Nebius) that purchase NVIDIA GPUs using debt, creating artificial demand (“barely $1B in revenue combined” across those three). AI software products are structurally loss-making per user — Anthropic loses on Claude Code subscribers; GitHub Copilot loses $20/month/user on average. Interviews with experienced developers confirm LLMs produce unreliable code requiring extensive human verification. Conclusion: “a fiery apocalypse.”

Implications

This is Ed’s most detailed technical and financial dissection of AI economics. The neocloud circular-demand finding is the most significant new data point: if NVIDIA’s “AI infrastructure demand” is partially NVIDIA-funded debt purchases, the demand signal underlying the entire AI capex thesis is fabricated. The per-user loss figures for AI coding products are directly useful for evaluating AI SaaS economics — any AI product that loses money on each subscriber requires either price increases (churn risk) or subsidy from investors (unsustainable). The developer interview methodology is also notable: primary source reporting over vendor claims.