We must pop the AI bubble sooner rather than later | Cory Doctorow
Sep 10, 2026 · 56m
More Technology podcasts · More News podcasts
Summary
Host Isaac and author Corey Doctorow discuss the AI bubble, arguing that current LLMs are statistical tools lacking true understanding and that the industry’s massive capital expenditure is unsustainable. They critique the "world without people" narrative driven by billionaires, highlighting how replacing skilled workers with chatbots destroys process knowledge and creates a death spiral. The episode explores the financial fragility of data centers, the misleading nature of OpenAI’s reported costs, and the importance of labor unions in resisting AI-driven job displacement.
Topics discussed
Sponsor: Rubric AI security platform
Sponsor: Midas tire guarantee and insidification
The death spiral of replacing skilled workers
Why the AI narrative captured public attention
The desire for a world without human friction
Replacing human friends with chatbots
Billionaire beliefs and the ketamine analogy
AI as a cynical sales pitch to capital
The limits of statistical probability in AI
Understanding vs. correlation in machine learning
AI does not understand, it only predicts
Autocorrect vs. true human understanding
Diminishing returns and the hard limit of LLMs
AI leaders' claims of god-like power
The nuke analogy and corporate culture
The narrative of AI inevitability
Fiscal constraints and data center resource use
The AI bubble and the railroad analogy
Bailing out the AI sector repeatedly
Inference costs and OpenAI financial leaks
Marketing expenses masking inference costs
The need to constantly retrain models
Comparing AI capex to railway mania
Morgan Stanley predictions and the metaverse
Stock as currency for growth firms
Sustainability of market share and growth
Keeping the investor story alive
The failure of early AI thought experiments
Blockchain vs. AI unpredictability
The appeal of a world without people
The reverse centaur and labor discipline
The importance of precise language
Screenwriters and the value of craft
Creating class enemies with AI output
Hollywood unions and sectoral bargaining
Historical obsolescence and the Luddites
Luddites as skilled technologists
Steam looms and material quality
Revising historical accounts of bubbles
The circular debt of AI firms
Sponsor: Thomson Reuters and One Times Square
Sponsor: Midas tire guarantee
Unstable founders and data center debt
Break clauses in data center financing
Mysterious infrastructure failures in the Gulf
Casualty bias in post-bubble analysis
LLM limitations in playing chess
Using a blowtorch to shovel snow
The cost of LLM swarms
The Concord analogy for AI applications
DeepSeek and the open source threat
The long road to smartphones
Process knowledge as an intangible asset
Tacit knowledge in corporate culture
Apple's MacBook Neo and process knowledge
Post-bubble infrastructure and demand
Strategies for national AI development
The death spiral of fixed costs
Nvidia's rapid hardware obsolescence
Data center retrofit costs
Sponsor: One Times Square and Rubric
Sponsor: Midas tire guarantee
AI as a tool for customer squeezing
Surveillance data and price optimization
Hallucinations as plausible deniability
Targeted hallucinations for profit
Outro and sponsor reads
Sponsor: Midas tire guarantee
More from The Tech Report
Listen ad-free on Castria