Ed Zitron has spent years warning anyone who would listen. The economics of large language models don’t add up. The data centers cost too much. The returns look worse every quarter. Now memory prices have doubled. Apple has raised prices on Macs and iPads. iPhones could follow. And Zitron says the company that makes those devices will simply sit back.
Watch everything burn.
That’s how the veteran tech commentator and author of the Where’s Your Ed At newsletter describes Apple’s likely posture. He laid out the case in detail during an interview published Monday by MacRumors. The piece arrives at a moment when fresh signs of strain have surfaced across the artificial-intelligence sector. But Zitron’s critique runs deeper than any single earnings miss. He argues the entire model of selling metered intelligence at subscription prices was flawed from the start.
Consumers expect software to work without surprise bills. Enterprises demand predictable costs. LLMs deliver neither. Tokens burn whether the output proves useful or spins in useless loops. Companies such as OpenAI and Anthropic responded by subsidizing usage. They offered 20 to 40 times more tokens than the subscription fee could justify. The result? Massive losses. Zitron’s own reporting on OpenAI’s financials, shared via his newsletter, showed the company lost $20.9 billion on $13.07 billion in revenue last year.
SemiAnalysis drove the point home with stark numbers. A $20 monthly subscription could generate hundreds of dollars in token costs. A $200 tier could reach thousands. Gross-margin claims of 70 percent lack solid proof. And when the firms shifted enterprise customers to pure token billing earlier this year, reality hit fast. Uber burned through its full annual budget in one quarter. Its chief operating officer told Business Insider that tracking value had grown difficult. Even Sam Altman later called the situation “a huge issue.”
But the problems extend far beyond one ride-hailing giant. Coding assistants from Cursor, Perplexity, and GitHub Copilot all lose money. A study covered by The Register found these tools often slow developers down while injecting low-quality code into repositories. Differentiation remains elusive. One model summarizes, generates, or searches much like the next. That explains why, according to The Information, Anthropic and OpenAI captured 89 percent of AI startup revenue. Most other players tout annualized run rates because actual monthly figures disappoint.
Data-center construction adds another layer of pain. These facilities take 18 to 36 months and billions of dollars to complete. They rely on project financing and debt. Yet real customers outside the two loss-making leaders prove scarce. Microsoft, Google, and Amazon built infrastructure for their partners, yet those partners still raise hundreds of billions while staying unprofitable. The memory squeeze tells its own story. Hyperscalers have absorbed so much supply that DRAM prices doubled this year. Apple chief executive Tim Cook labeled recent price hikes “unavoidable.”
Zitron sees broader fallout ahead.
Private credit funds have financed many of these projects. Pension systems such as CalPERS and the San Francisco teachers’ fund sit behind them. A collapse could ripple through retirement accounts nationwide. Bailouts look improbable. The structures use special-purpose vehicles. Rescuing them would require hundreds of billions and invite fierce political backlash. Oracle faces particular danger, Zitron argued in an earlier piece on his site. The company has leaned on acquisitions to stay flat for two decades. Its massive AI data-center bets, exceeding $340 billion with heavy debt, assume OpenAI becomes the world’s most profitable enterprise by 2030. He gives those odds little credence.
Supply-chain effects would reach Asia. Taiwanese firms such as Quanta and Foxconn have enjoyed AI-server revenue bumps. Korean investors tied to the KOSPI stand exposed. Semiconductor makers and hyperscalers in the United States could see valuations reset sharply. And yet Apple, in Zitron’s view, emerges relatively unscathed. The company never dove headfirst into the token economy. Its $14 billion annual spend on AI features, as referenced in follow-up coverage by AI Weekly, looks measured by comparison. It can afford to observe from the sidelines.
Recent commentary on X echoes parts of this caution. One user noted Chinese open-source models rivaling expensive Western efforts, suggesting Apple might later acquire talent or technology cheaply once valuations crash. Others simply shared the MacRumors interview, signaling growing interest in skeptical takes. A post from technology journalist Matt Rosoff at The Register amplified the story hours after publication.
Earlier reporting adds context. In July 2025, leaked documents obtained by Zitron and covered by TechCrunch revealed OpenAI’s escalating payments to Microsoft. Those figures underscored the dependency and the cash burn. Bloomberg has featured Zitron multiple times, including a June 2026 segment where he argued Anthropic and OpenAI should not be permitted to go public given their lack of profits.
Critics of the bear case point to stock performance. AI-related shares have climbed even as fundamentals lag. Hyperscalers rarely break out AI-specific revenue, which helps sustain optimism. Yet Zitron maintains the demand story is largely self-referential. Much of the capacity serves the model providers themselves. Enterprise adoption beyond experimentation remains limited. Real productivity gains stay hard to document at scale.
So what does this mean for the devices in consumers’ pockets? Higher prices today reflect component inflation driven by AI buildout. Tomorrow could bring a different environment. If token economics collapse and data-center debt sours, the broader tech sector may contract. Apple, with its cash reserves, loyal customer base, and slower approach to generative features, could gain relative strength. It might even pick up assets or engineers at bargain rates once the frenzy fades.
Zitron doesn’t predict immediate doom. He notes the bubble has persisted longer than many expected. But the fundamentals haven’t changed. Subscriptions can’t forever mask per-token losses measured in the hundreds or thousands of dollars. Construction timelines for new capacity stretch years into the future. Memory and power constraints grow tighter. And the two companies dominating revenue continue to post eye-watering deficits.
Apple’s strategy appears clear in this light. Ship useful on-device intelligence where possible. Avoid the heaviest cloud costs. Raise device prices as needed to offset input inflation. Then wait. The approach lacks drama. It also avoids the trap of overcommitting to technology whose unit economics may never stabilize.
Whether Zitron proves correct will unfold over quarters, not days. Markets can remain irrational. Capital can keep flowing. But the conversation has shifted. More voices now question the path from hype to profit. Memory prices don’t lie. Lost billions don’t vanish. And if the burn rate finally overwhelms the subsidies, Apple may indeed watch the rest of the industry grapple with the consequences.
WebProNews is an iEntry Publication