Ed Zitron has spent years warning that the artificial-intelligence boom is an economic bubble. In a new interview, the tech commentator and podcaster lays out why the large language model industry is fundamentally broken, who will get hurt when the collapse comes, and why Apple is likely to emerge largely unscathed.

Key Facts

• Ed Zitron argues LLM costs run against every software business model.
• AI companies give subscribers 20 to 40 times more tokens than subscriptions cover.
• Hyperscalers have spent over $1 trillion on capex since 2022; Zitron says AI data centers won't turn a profit.
• DRAM prices have doubled, pushing Apple to raise Mac, iPad, and iPhone prices.
• Zitron expects Apple to sit on the sidelines and "watch everything burn" if the bubble deflates.

The Broken Economics of AI

At the core of Zitron's argument is a simple mismatch: LLMs burn tokens by the million, regardless of whether a user gets a useful answer. Unlike traditional software, which has fixed subscription costs, AI services have metered and hard-to-measure costs. A coding agent can spin in a loop for hours and customers still pay for every token consumed.

Most AI companies know consumers would never pay the real cost of these services. Instead, they sell monthly subscriptions with vague rate limits, then burn far more in compute than the subscription price covers. SemiAnalysis, according to Zitron, found that a $20-a-month subscription can burn hundreds of dollars in tokens, while a $200-a-month plan can burn thousands. AI boosters claim gross margins of 70% on tokens, but Zitron says there is little proof. His reporting indicates OpenAI lost $20.9 billion on $13.07 billion of revenue in 2025.

The basic economics are therefore broken. If Anthropic and OpenAI believed customers would pay real costs, they would not have to give away 20 to 40 times the amount of tokens in subscriptions. In March 2026, both companies moved enterprise customers to token-based billing. Within weeks, Uber reportedly spent its entire annual token budget in a single quarter. Its COO said it was getting "harder to justify" AI costs because tracking them to actual shipping features was difficult. Sam Altman acknowledged it was a "huge issue" but offered no fix.

Everywhere the Model Fails

Zitron says this problem extends across all AI startups. Perplexity, Cursor, GitHub Copilot — which moved to token-based billing in June 2026 — are unprofitable because users refuse to pay the true cost of inference. At the same time, AI services are not that useful or differentiated. A coding assistant might help in some cases, but Zitron says it often makes developers slower and fills codebases with slop. An LLM can generate, summarize, and search, but little else. That leaves every AI service looking essentially the same, which is why 89% of all AI revenues go to Anthropic and OpenAI and why startups talk about "annualized revenue" — actual revenue is depressing.

Data centers make matters worse. An AI data center costs billions, takes 18 to 36 months to build, and is usually financed with debt. The only real customers are Anthropic and OpenAI, both deeply unprofitable. Even with Microsoft, Google, and Amazon building infrastructure, OpenAI and Anthropic have had to raise hundreds of billions. The market is calm only because AI-related stocks have performed well, even though the hyperscalers do not clearly disclose AI revenue.

Who Bears the Cost?

If the AI bubble unravels, Zitron says the losses will fall on private credit funds and ultimately pension funds. Data center projects are financed through special purpose vehicles, often backed by pensions like the SF teachers fund or CalPERS. That makes the potential spread of contagion genuinely dangerous. A bailout would be politically toxic because it would require hundreds of billions to make the SPVs whole, and there is no obvious business to save.

Zitron also predicts Oracle will be killed by OpenAI. Oracle's revenues have stagnated for 20 years; it has used more than $85 billion in acquisitions just to stay flat. Its AI data center bet, reportedly $340 billion plus with hundreds of billions in debt, requires OpenAI to become the world's largest and most profitable company by 2030. Zitron is skeptical that will happen.

The broader fallout would hit global markets. The TWSE depends on Taiwanese ODMs like Quanta and Hon Hai, whose revenues have been lifted by AI server sales. Those companies will probably survive, especially because Hon Hai makes much of Apple's hardware, but their stock prices will suffer. Korean investors on the KOSPI and American investors in hyperscalers and semiconductor companies would also feel the pain.

Consumers Pay for the Bubble

The speculative buildout is already raising hardware prices. Hyperscalers have consumed so much memory supply that DRAM prices have roughly doubled in a year. Apple CEO Tim Cook called recent price increases "unavoidable." Macs and iPads are already more expensive, and iPhones are expected to follow. Zitron says consumers are paying more for hardware to subsidize data centers that may never turn a profit. The hyperscalers need more than $1.5 trillion in new profit — not revenue, but actual profit — to justify even half of the $1 trillion in capex spent since 2022.

Apple's Strategy

Apple is spending about $14 billion on AI infrastructure this year while the hyperscalers plan to spend more than $650 billion. Apple pays Google around $1 billion a year for Gemini to run Siri and does what it can on-device. Zitron sees Apple Intelligence as both the worst and best thing to happen to Apple during the AI bubble.

It was, he says, a mass-radicalization of users against AI. Apple tried to cram a barely functional series of add-ons nobody asked for into its devices. Summaries became memes, and the redesigned Siri was somehow worse than the old one. That reaction told Apple to pump the brakes. The company has barely spent on capex and has done little with AI. Headlines keep saying Apple is falling behind, but nobody can explain what it is falling behind on or why it matters. People hate Apple Intelligence, and Zitron believes Apple knows it. So Apple will "jingle the keys" for the markets by attaching AI labels to products without fully committing.

What Happens to Apple?

If the bubble deflates, Zitron expects Apple to look much the same. "I think they will sit on the sidelines and watch everything burn," he said. Apple could make some choice acquisitions as the collapse unfolds, or it could do nothing.

He notes that Apple is in a strange place. The Vision Pro was a dud, but it was also the most interesting and future-forward product in a while. If Apple were smart, Zitron argues, it would tread water and pour money into making the Vision Pro as small and as light as possible. The entire AI bubble, in his view, happened because tech companies ran out of hypergrowth ideas after exhausting new interfaces.

Zitron is careful to note that the Vision Pro's potential requires it to be weightless, invisible, and free of constant adjustments. When it works, he says it is genuinely awesome, but one slight movement can throw the whole thing out of focus. He says he can no longer use his because an update requires wearing the device during the entire installation. It was promising technology released too early and shoved out by a CEO on his way out.

Source: MacRumors News