Apple has begun raising prices on Macs and iPads, and iPhone price increases are widely expected to follow. The immediate cause is a sharp rise in memory prices, driven by the artificial intelligence industry’s voracious appetite for data-center infrastructure. To critics of the AI boom, this is the clearest sign yet that speculative spending is spilling into ordinary consumers’ wallets.
Ed Zitron, a tech commentator and podcaster who has spent years arguing that the AI buildout will never pay for itself, believes Apple is in a unique position. While hyperscalers have spent staggering amounts on AI data centers, Apple has spent comparatively little. If the bubble bursts, Zitron says, Apple will be able to sit on the sidelines and watch the rest of the industry burn.
The economic problem inside large language models
Zitron’s central argument is that large language models don’t behave like normal software from an economic perspective. Most software is sold as a monthly subscription with predictable limits. Consumers and enterprises have been trained to pay a flat fee for services such as streaming, storage, or productivity suites. AI products, however, are metered by token usage, and that usage is both difficult to forecast and difficult to control.
Every request made to an LLM burns tokens at a per-million rate, whether or not the response is useful. If a coding agent gets stuck in a loop, the user still pays for those tokens. If a chatbot provides a wrong answer, the tokens are gone anyway. That is fundamentally different from traditional software, where the cost of running the service is largely hidden from the user behind a fixed subscription.
AI companies know that consumers would not pay the actual cost of AI services, so they have mostly sold monthly subscriptions with vague rate limits. This allows users to burn far more in tokens than the subscription fee justifies. Research from SemiAnalysis has suggested that a $20-a-month subscription can generate hundreds of dollars in token consumption, and a $200-a-month plan can generate thousands. Despite claims that AI companies enjoy gross margins of 70 percent on tokens, there is little evidence to support that. Zitron’s own reporting has pointed to OpenAI losing $20.9 billion on $13.07 billion in revenue in 2025.
For Zitron, this proves the basic economics are broken. If Anthropic and OpenAI truly believed their customers would pay the real cost of tokens, they would not give away 20 to 40 times the token allowance that the subscription price implies. The subsidy is a sign that real demand at profitable prices does not exist.
Enterprise token billing is already backfiring
The problem has become even more obvious in the enterprise market. In March, both OpenAI and Anthropic moved enterprise customers to token-based billing. Within weeks, reports emerged that Uber had spent its entire annual token budget in a single quarter. The company’s chief operating officer said it was getting harder to justify AI costs because it was difficult to connect those costs to actual features that shipped. Sam Altman eventually acknowledged that this was a huge issue, but he did not explain how it would be fixed.
This is not just a problem for the big AI labs. Startups such as Perplexity, Cursor, and GitHub Copilot are all exposed to token costs at the per-million level. GitHub Copilot moved to token-based billing in June. Almost every AI-powered startup is unprofitable because users are unwilling to pay the true cost of AI.
There is also a differentiation problem. While AI-generated code can be useful, Zitron argues that these tools often make developers slower and fill codebases with low-quality output. Beyond that, an LLM is an LLM: it can generate, summarize, and search, but not much else. Most AI services are effectively interchangeable. That helps explain why, according to Zitron, around 89 percent of all AI revenue goes to just two companies: Anthropic and OpenAI. Other startups are forced to report annualized revenue rather than actual revenue because the real numbers are depressing. Even then, most are barely crossing $100 million in annualized revenue.
Who will bear the cost when the bubble bursts?
The AI bubble is not just a software story. It is also a story about data centers, debt, and pension funds. An AI data center can take 18 to 36 months to build and cost billions of dollars. The companies building them usually take on significant debt and only get paid once a customer moves in. But outside of OpenAI and Anthropic, there are almost no customers for AI data centers. And both OpenAI and Anthropic are so unprofitable that they have had to raise hundreds of billions of dollars even after Microsoft, Google, and Amazon built much of their infrastructure.
Zitron worries that the real cost will fall on private credit funds, many of which are backed by pension funds such as the San Francisco teachers fund and CalPERS. If AI data center projects fail, those losses could create systemic contagion. He is skeptical that a bailout would work. Data centers are often funded through project financing, meaning the money is already gone. Making investors whole would require buying out huge amounts of debt or feeding revenues into empty facilities. That would cost hundreds of billions of dollars and would be politically toxic.
Oracle is a particular concern. The company’s revenues have stagnated for twenty years. It has spent more than $85 billion on acquisitions just to keep things flat. Its AI data center bet is enormous, reportedly exceeding $340 billion with hundreds of billions in debt. That bet only works if OpenAI becomes the largest, most profitable company in the world by 2030. Zitron is blunt about those odds: good luck, Larry.
The fallout would extend beyond the United States. The Taiwanese stock exchange is heavily dependent on ODM companies like Quanta and Hon Hai, also known as Foxconn, which have enjoyed a revenue boost from AI servers. A collapse would not necessarily kill those companies, especially since Foxconn earns so much from Apple, but it would hammer their stock prices. The same would be true for Korean investors on the KOSPI and American investors in hyperscalers and semiconductor companies that are being lifted by AI enthusiasm.
Consumers are already paying for the AI bubble
Apple’s recent price increases are one of the clearest signs that the bubble’s costs are being passed on. DRAM prices have roughly doubled this year because data center builders have absorbed so much of the world’s memory supply. Tim Cook has called Apple’s price increases unavoidable. Macs and iPads have already gone up, and iPhone models are expected to follow.
Zitron does not believe the AI data centers will ever turn a profit. Hyperscalers have spent more than $1 trillion in capital expenditures since 2022. Even if only half of that goes to AI data centers, they would need to generate over $1.5 trillion in entirely new profit, not revenue, to justify the spending. In the meantime, ordinary people are paying more for hardware for no reason other than tech companies’ collective desire to build as many data centers as possible. He has called this the Hater’s Guide to the Memory Crisis.
Why Apple may be more savvy than it looks
For years, Apple was accused of falling dangerously behind in AI. It responded with Apple Intelligence, a set of features bundled into iPhones, iPads, and Macs. The rollout was widely mocked. Summaries quickly became memes. The new Siri was seen as worse than the old Siri. Apple Intelligence became a mass-radicalizing experience that turned many users against AI, according to Zitron.
But it also told Apple to pump the brakes. The company has spent very little on AI capex compared with its peers. Around $14 billion in infrastructure spending is nothing next to the more than $650 billion that the hyperscalers are spending this year. Apple treats AI as a commodity: it pays Google around a billion dollars a year to run Gemini behind Siri and does as much as possible on-device.
Zitron thinks this caution is quiet intelligence. Despite endless headlines about Apple falling behind, nobody can really explain what Apple is falling behind on or why it matters. People hate Apple Intelligence, and Apple knows it. So Apple will continue to attach the letters AI to products to satisfy the market, without ever fully committing to the speculative boom.
What happens to Apple when the bubble bursts?
If AI spending collapses, Apple is unlikely to break. Zitron expects Apple to sit on the sidelines and watch everything burn. There could be some carefully chosen acquisitions as distressed AI companies become cheap, but there is also a very real possibility that Apple does nothing at all.
Apple’s position in hardware gives it breathing room that pure AI companies do not have. The iPhone is not dependent on AI. Neither are Macs or iPads. The company can allow the AI market to crater while continuing to sell devices that serve ordinary needs. If memory prices fall, Apple might even benefit from lower input costs and the opportunity to regain pricing power.
Zitron also points to the Vision Pro as an example of Apple’s willingness to invest in genuinely new interfaces. The headset was a commercial disappointment, but it remains one of the most interesting pieces of technology to come out of any major company in years. He argues that the AI bubble is partly a result of the industry running out of hypergrowth ideas, because there are no new interfaces left. The full promise of the Vision Pro, he says, requires the device to become weightless, invisible, and free of constant adjustment. When it works, it is genuinely awesome. But that “when” is load-bearing. One slight movement makes the whole thing go out of focus. The product, in his view, was released too early and pushed out the door by a CEO on his way out.
Even with those flaws, Zitron believes Apple would be smart to keep investing in the Vision Pro and related spatial computing projects. If the AI bubble bursts, Apple can use its enormous cash position to wait out the chaos and then move forward in a world where the difference between hype and useful technology is much clearer. It may not need to do anything dramatic. It just needs to be patient, avoid the speculative spending trap, and be ready to act if real opportunities emerge.
Source:MacRumors News