Cory Doctorow argues that today's AI boom rests on a category error: LLMs find statistical correlations but don't understand anything, so they hit hard limits and steep diminishing returns. The money behind it is worse — roughly a trillion dollars spent this year against about $50 billion in revenue, with Nvidia lending its customers the money they spend at Nvidia, and OpenAI apparently booking giveaway inference as 'marketing.' He expects the bubble to pop from some brittle link (a margin call, a data center financing tranche walking away, Gulf money redirected to rebuilding oil infrastructure) and wants it popped sooner rather than later, because every skilled worker replaced by a chatbot destroys 'process knowledge' that takes a generation to rebuild.
- The AI story sells because it promises 'a world without people': bosses who never face expert pushback, and workers whose know-how buys them no leverage — a pitch that works on investors whether or not the product works.
- LLMs can predict what people will say because people repeat themselves, not because they understand; the difference shows up exactly when something novel happens, and it means there's a ceiling that bigger models can't break through.
- The economics don't close: about a trillion dollars in spending produced roughly $50 billion in revenue, inference costs aren't actually falling, and there are almost no switching costs, so firms must keep retraining models — 'building the same railroad over and over again on the same patch of ground.'
- Ed Zitron's leaked OpenAI financials show a marketing line item the size of Coca-Cola's with no visible advertising, which Doctorow reads as inference given away and relabeled as marketing.
- The bubble is brittle in several places at once: circular Nvidia vendor financing among firms that are 35% of the S&P 500, data center debt with break clauses that could trigger across many projects simultaneously, Gulf investors who may prefer rebuilding LNG terminals to honoring contracts, and founders unstable enough to blow up their own companies.
- Statistical inference stays useful after the crash, but mostly in cheaper forms — an LLM will make illegal chess moves that a 1950s relay computer never would, while LLMs are genuinely good at helping skilled programmers build the chess engine.
- Workers should insist the threat is 'my boss will give my job to an AI that can't do it,' which recruits the public as an ally; the screenwriters are the only workers to have beaten AI, and they did it with a union able to bargain across an entire sector.
- Firing skilled staff and switching off the chatbots later doesn't restore the status quo — 'process knowledge,' the unwritten ropes of a workplace, never appears on a balance sheet and takes a generation to regrow.
- AI also supercharges enshittification through algorithmic price and wage discrimination — US nurses offered lower shift wages based on how much credit card debt they carry — with hallucinations providing plausible deniability for rigged results.
A concrete, checkable accounting claim that undercuts the industry's central promise that inference costs are falling.
Their marketing budget is the size of Coca-Cola's... There's no evidence that OpenAI is marketing like that. Ed's theory, which I think is totally true, is that OpenAI is giving away shedloads of inference and calling it marketing. And that doesn't mean inference is getting cheaper.
The clearest one-sentence description of the circular financing that makes the whole sector fragile.
You have six companies that are losing money on AI, and a seventh one they're losing money to called Nvidia, and together they are 35% of the S&P 500, and Nvidia is loaning them the money that they're losing to Nvidia.
These people who think that teaching enough words to the word-guessing program makes it conscious are like people who think that if you breed horses to run fast enough, the mares will start giving birth to locomotives.
Never saying that your fear was that the AI was going to take your job, but rather that your boss was going to give your job to an AI that couldn't do it.
AI is the money-losingest thing anyone has ever done. They want more money and they'll make it worse too.