Ed Zitron, the blunt-talking columnist and podcaster known for his commentary on the rise of artificial intelligence, warns that the economy built around large language models (LLMs) is fundamentally broken. In his view, rising costs — memory, data centers and usage-based billing — have created a bubble that, if it bursts, would leave a cascade of heavy losses. For Apple, he says, the company would mostly be “watching everything burn” on the margin: little immediate impact for users, but widespread financial consequences across the sector.
Why the LLM economy is fragile
Zitron argues that the cost model for LLMs runs counter to traditional software purchasing habits: instead of predictable subscriptions, vendors charge by the token consumed — a usage metric whose cost is decoupled from outcome. Individual users and enterprises can burn tens or hundreds of dollars worth of tokens on cheap plans, while providers often give vague caps that mask the real bill. He sees this practice as one reason OpenAI reported massive losses — $20.9 billion on $13.07 billion of revenue in 2025 — and why many services remain unprofitable.
Enterprise migration to token-based billing has revealed the model’s limits: customers like Uber reportedly spent their annual token budgets in a single quarter, undermining the case that benefits justify the expense. Moreover, the low functional differentiation between LLMs — all capable of generation, summarization and search — concentrates the bulk of revenue among a few players (Zitron cites 89% for Anthropic and OpenAI), while startups struggle to build meaningful recurring revenue.
Who will foot the bill if the AI bubble bursts
Zitron says the debt tied to building data centers is a heavy weight: these facilities cost billions and are often financed through special-purpose vehicles (SPVs). The private credit funds that back those projects, themselves funded by pension pools (e.g., San Francisco teachers, CalPERS), would be the first to be exposed. A collapse would leave few easy public policy levers for a bailout: buying the debt or generating revenue for those SPVs would involve politically sensitive and substantial sums.
He also flags sectoral risks: server and component suppliers — Taiwanese manufacturers such as Quanta and Hon Hai (Foxconn) — have seen revenue boosts from AI server sales; a drop in orders would hit the Taiwan Stock Exchange (TWSE) and ripple into markets like the KOSPI and semiconductor and hyperscaler stocks tied to AI. He further notes the vulnerability of firms such as Oracle, whose large infrastructure bets on AI, he argues, depend on extraordinary outcomes from players like OpenAI.
Consequences for Apple if the AI bubble bursts
According to Zitron, Apple occupies a particular position: it has spent relatively little on AI capex (around $14 billion in the cited year) and outsources part of its needs — for example, paying Google roughly $1 billion per year. As a result, Apple would likely face limited immediate product disruption but could still suffer significant margin pressure and broader market fallout if the AI financing model collapses.
