Apple is confronting a surge in component costs, with memory prices reportedly doubling within a year and recent price hikes hitting Macs and iPads. Industry watchers expect iPhone price increases to follow as the company grapples with the fallout of an AI-driven infrastructure boom that has consumed vast supplies of DRAM and other memory chips. Ed Zitron, a writer, podcaster, and long-time critic of the AI industry, argues that this is only the surface of a much larger economic problem.

Zitron has been calling the AI boom a bubble for years. In a recent discussion, he laid out why the current business models for large language models are unsustainable, how data center debt could spread through the financial system, and why Apple, despite its $14 billion annual spending on AI, is well positioned to sit back and watch the collapse.

The Core Problem With AI Economics

Large language models operate on a cost structure that is fundamentally different from traditional software. Customers are used to paying a monthly subscription fee for services with predictable limits. AI, by contrast, is metered by tokens, and the costs can be opaque and volatile. Even when a user receives a bad response, the tokens are consumed and the provider incurs expenses. This creates an uncomfortable mismatch between what customers pay and what the service actually costs to run.

Zitron points out that Anthropic and OpenAI were forced to offer their consumer subscriptions with vague rate limits and give away far more tokens than the subscription price justifies. He cites an analysis from SemiAnalysis showing that a $20-per-month subscription could cost a provider hundreds of dollars in computing resources, while a $200-per-month plan could cost thousands. Despite claims from AI companies that they enjoy 70% gross margins on tokens, Zitron argues that the evidence is thin. His own reporting indicates that OpenAI lost $20.9 billion in 2025 despite generating $13.07 billion in revenue.

In March, both Anthropic and OpenAI moved their enterprise customers to token-based billing. Shortly afterward, Uber reportedly exhausted its annual token budget within a single quarter, and its COO said it was getting harder to justify AI costs because it was difficult to connect them to useful features. OpenAI's Sam Altman eventually acknowledged that the situation was a 'huge issue' but did not offer a clear solution.

The same financial strain affects nearly every AI startup. Perplexity, Cursor, GitHub Copilot, and other AI products rely on token-based pricing from a handful of foundation model providers. According to Zitron, these startups are unprofitable because their users refuse to pay the true cost of AI. He notes that 89% of all AI revenue flows to Anthropic and OpenAI, and that many AI startups trumpet 'annualized revenue' numbers that are simply monthly revenue multiplied by twelve, masking disappointing actual results.

Data Centers and Debt: A Contagion Waiting to Happen

The AI bubble is not just about software. The infrastructure needed to train and run AI models is enormously expensive. A single AI data center can take 18 to 36 months to build, cost billions of dollars, and is typically financed through debt. Zitron explains that the revenue from such facilities depends entirely on a few major customers, principally OpenAI and Anthropic. Both are deeply unprofitable and have raised hundreds of billions of dollars even while Microsoft, Google, and Amazon continued to build infrastructure on their behalf.

This has created a fragile ecosystem. If the AI bubble deflates, the losses could spread to private credit funds and pension funds that have financed these data centers through special purpose vehicles. Zitron mentions the SF teachers fund and CalPERS as examples of institutional investors that could be exposed to contagion. Governments might be pressured to offer bailouts, but the structure of project financing makes it politically difficult and financially uncertain.

Oracle is among the most exposed. The company's core revenues have stagnated for decades, and it has relied on more than $85 billion in acquisitions to stay afloat. Its massive bet on AI data centers, reportedly exceeding $340 billion in commitments with hundreds of billions in debt, requires OpenAI to become the most profitable company in the world by 2030. Zitron calls that an extremely unlikely outcome and predicts that Oracle could be destroyed by OpenAI's inevitable collapse.

Global markets would also feel the pain. Taiwan's TWSE index is heavily dependent on ODM manufacturers like Quanta and Hon Hai, which have seen revenue boosts from selling AI servers. A sharp decline in AI spending would hurt their stock prices. South Korea's KOSPI would face similar headwinds, and American investors would be exposed through hyperscalers and semiconductor companies that have become dependent on AI-related growth.

Apple's 'Unavoidable' Price Increases

Consumers are already feeling the squeeze. DRAM prices have roughly doubled this year, and Apple's Tim Cook called the resulting price increases 'unavoidable.' Macs and iPads have gone up, and future iPhone models are expected to carry higher prices. This has led to an uncomfortable conclusion: ordinary consumers are paying more for hardware to subsidize speculative data center construction that may never generate a profit.

Zitron believes none of the AI data center projects will turn a profit. The hyperscalers have spent over $1 trillion in capital expenditures since 2022, and even if half of that went to AI, they would need to generate more than $1.5 trillion in entirely new profits, not just revenue, to justify the investment. The chance of that happening is, in his view, virtually zero.

Apple's Cautious Approach: Smart or Lucky?

Apple's response stands in sharp contrast to the hyperscalers. While Google, Microsoft, Amazon, and Meta are spending a combined $650 billion on AI infrastructure this year, Apple is spending roughly $14 billion. The company has chosen to rent some AI capabilities, paying Google around $1 billion per year for Gemini to power Siri, while also doing as much as possible on-device.

Apple Intelligence has been widely criticized. Zitron calls it a 'mass-radicalization' of users against AI, pointing to the barely-functional features, mocked summaries, and an updated Siri that many found worse than the old one. He suggests that the backlash may have convinced Apple to pull back. Despite endless headlines about Apple falling behind, there is no clear definition of what Apple is losing. Apple Intelligence is disliked, and Apple seems to know it, choosing to attach the AI label to products without making a full commitment.

What Happens When the Bubble Bursts?

Zitron expects Apple to weather the storm comfortably. 'I think things would look much the same for Apple,' he says. 'They will sit on the sidelines and watch everything burn.' He sees the possibility of strategic acquisitions as rivals struggle, but also acknowledges that Apple could simply do nothing.

Apple is in an odd position. The Vision Pro has been a commercial disappointment, but Zitron regards it as one of the most forward-thinking products in years. If Apple is smart, he argues, it will invest in making the headset smaller, lighter, and more practical, regardless of how long that takes. The entire AI bubble, in his view, is a symptom of an industry chasing hypergrowth because it has run out of new interfaces.

The Vision Pro's promise, however, is constrained by its physical limitations. It needs to be weightless and invisible, according to Zitron, and it still requires constant adjustments to stay in focus. He says he can no longer use his own unit because a software update forces the user to wear the device throughout the entire upgrade process. It was a product released too early, he says, and pushed out by a CEO on his way out.

Source: MacRumors News