Ed Zitron, who writes the Where's Your Ed At newsletter and hosts the Better Offline podcast, has long warned that the AI bubble will never pay for itself. Memory prices have doubled, Macs and iPads have gone up, and iPhones are expected to follow. Apple’s price hikes have been called *unavoidable* by Tim Cook, a clear sign that the cost of building AI infrastructure is being passed on to consumers.

At their core, Large Language Models burn tokens at a per‑million rate regardless of output quality. Consumers expect a monthly fee, not a metered service that can cost hundreds or thousands of dollars for a single subscription. OpenAI lost $20.9 billion on $13.07 billion in revenue in 2025, and Uber spent its entire annual token budget in a quarter after moving to token‑based billing.

Building AI data centers is expensive – it takes 18 to 36 months and costs billions. Formation of private‑credit funds backed by pension funds means that a collapse would hit institutional investors worldwide. Oracle’s $340 billion debt and the reliance of Taiwanese ODMs like Quanta and Hon Hai on AI server sales add further contagion risk.

Apple is largely keeping its hands off large‑scale AI. The Vision Pro remains a dud, and the company is spending only $14 billion on AI infrastructure compared to $650 billion by hyperscalers. If the bubble deflates, Apple will probably sit on the sidelines, watching everything burn.