Critics of generative AI have for the most part been obsessed with a single question: What if the several hundred billion–dollar bet on the future of the world economy fails? This isn’t just a concern about the benefits of the technology. Bottlenecks exist at seemingly every stage. Energy supply is severely constrained by regional war in West Asia; information is limited by copyright laws; fewer than half of planned data centers are actually being built; and chips may too be in short supply.

Meanwhile, the usefulness of actually existing AI has proved hard to calculate. A paper by Nobel Prize–winning economist Daron Acemoglu calculated that the new technology has had little effect on productivity and is unlikely to do so in the future. For day-to-day users, who employ large language models at work, their experience is often one of having to pick through inaccuracies and confusions caused by machine “hallucinations.”

Given the hype surrounding AI, it is hard to avoid the feeling that the whole US economy is balancing rather precariously on a house of cards.