Summary

A long interview with Ed Zitron, a technology PR veteran and the industry's most prominent financial sceptic, arguing that generative AI is — his word — 'a con': that the capability is oversold, the economics do not work, and the revenue is circular. The host pushes back repeatedly on adoption figures and on the 'investing ahead of monetisation' argument, so the counter-case is present rather than absent. EVERY FIGURE BELOW IS ZITRON'S CLAIM AS STATED ON THIS PODCAST and none is independently verified here. Note also that this is a third-party clips channel, not the original publisher.

• The circularity claim, which is the core of his case: he says roughly 70% of all AI revenue across the major cloud providers comes from OpenAI and Anthropic — two companies he describes as unprofitable and unable to exist without money from those same providers. He cites Amazon sending $50bn to OpenAI and $5bn to Anthropic this year, and Google $10bn to Anthropic.
• On disclosure: he says these companies do not break out AI revenue, and when they do they use an 'annualised run rate' they never define — 'it can mean month times 12, it can mean month times 13' — and argues that selective silence from companies that publicise good news is itself evidence.
• His Microsoft arithmetic, attributed to Bloomberg: FY2026 AI revenue of about $34.33bn, of which $24.1bn came from OpenAI, leaving roughly $10bn — against $115bn of capital expenditure that year and an intended $175bn next year.
• On unit economics he cites Semi Analysis finding that a $200/month ChatGPT subscription can consume $14,000 of tokens, and Anthropic's $8,000 for the same $200. He says OpenAI lost $20.9bn last year, and that Uber exhausted its annual token budget in three months.
• The scale illustration: Stargate Abilene in Texas at 1.2 gigawatts, eight buildings of 50,000 NVIDIA GB200 GPUs each — more power than the city of Bristol, in a footprint he puts at roughly a thousandth of the area.
• On adoption he does not deny usage; he denies it is voluntary. He calls it 'the largest non-consensual push of technology in history' — Gemini in Google Docs, Copilot in Word — and argues people use chatbots mainly as search, and mainly because they have been told for three years that they will fall behind if they don't.
• His strongest structural point, and the one least dependent on any single figure: unlike railways or fibre, he argues there is no post-bubble use for the asset, because AI GPUs are not repurposable for anything else.
• The host's counter-figures, given on the show: 88% of organisations use AI for at least one business function, and 95% of his own company's staff use a chatbot daily. Zitron's reply is that adoption under default-on pressure is not evidence of value, and that enterprises started objecting the moment they were asked to pay actual cost — he quotes Sam Altman calling it 'a huge issue'.

Why it matters

Worth keeping as the strongest available statement of the bear case, and worth keeping clearly labelled as one man's argument on a podcast. Two parts deserve different weight. The financial claims are specific, sourced to named analysts, and checkable — and should be checked before any of them is repeated as fact. The structural claim is harder to dismiss and does not depend on the arithmetic: if AI GPUs have no second life, then the usual consolation that a burst bubble leaves useful infrastructure behind does not apply here. Against that, the host's adoption figures are real and Zitron's answer to them is an assertion about motive, which is the weakest link in his case. File it as a serious counterweight to be verified, not as a finding.

TNc0aTIZBcc-transcript.txt