Notes ·
The AI Bubble Can Burst Without AI Going Away
Tim Hardwick interviews Ed Zitron for MacRumors about his argument that the economics behind the current AI boom are unsustainable.
Zitron’s core point is that large language models do not behave like conventional software businesses. Every request consumes computing resources, while customers are accustomed to flat monthly subscriptions and often use far more processing than their subscriptions cover. AI companies are therefore subsidizing usage while spending enormous amounts on chips, memory and data centers.
He argues that the industry is relying on future demand and profitability that may never arrive. Hyperscalers have committed vast amounts of capital to infrastructure, but they disclose little about the revenue or profit directly attributable to AI. Meanwhile, those investments are already affecting memory supplies and hardware prices for everyone else.
I agree that the bubble is real. Too much money is chasing products with unclear margins, weak differentiation and business models that depend on investors continuing to absorb the true cost.
That does not mean AI is fake or going away. The dot-com crash did not kill the internet. It destroyed companies whose valuations, spending and assumptions could not survive contact with reality.
The difficult part is that we do not know who wins.
Some model providers may become foundational infrastructure. Others may collapse under training and inference costs. Hardware companies could benefit from years of demand or be left exposed to excess capacity. Companies such as Apple, which spent comparatively little and rented access to outside models, may look cautious or remarkably disciplined depending on how the market develops. Zitron believes Apple could largely remain on the sidelines and potentially acquire useful assets after a correction.
The technology will remain. The current collection of companies, prices and promises probably will not.