Background & Context§
The artificial intelligence boom has driven a massive buildout of data centers and a surge in demand for AI accelerators, memory, and other hardware. This has led to significant price increases for consumer electronics, including Apple's Macs and iPads, with iPhones expected to follow. Critics like Ed Zitron, a tech commentator known for his acerbic takes, have long argued that the economics of large language models (LLMs) are fundamentally flawed. In a recent interview with MacRumors, Zitron elaborated on why he believes the AI bubble is unsustainable and how Apple may be uniquely positioned to ride out the storm.
The News: What Happened Exactly§
In a detailed interview, Ed Zitron laid out his case that the AI industry is a bubble propped up by unsustainable spending. He argues that the core economics of LLMs are broken because the cost of tokens (the basic units of AI processing) is metered, while consumers and enterprises expect flat-rate subscriptions. AI companies like OpenAI and Anthropic have subsidized usage by offering subscriptions that cost far less than the actual compute consumed. For example, a $20/month subscription can burn hundreds of dollars in tokens, and a $200/month subscription can burn thousands. Zitron cites SemiAnalysis data and his own reporting, noting that OpenAI lost $20.9 billion on $13.07 billion in revenue in 2025.
The problem extends to enterprise customers who have moved to token-based billing. Uber, for instance, exhausted its annual token budget in a quarter, and its COO said it was harder to justify AI costs when they couldn't tie them to shipped features. Sam Altman acknowledged this as a "huge issue" but offered no fix. As a result, most AI startups—Perplexity, Cursor, GitHub Copilot—are unprofitable because users won't pay true costs.
Another pillar of the bubble is data center construction. These facilities cost billions and take 18-36 months to build, funded largely by debt. The only major customers are OpenAI and Anthropic, which are themselves unprofitable, forcing them to raise hundreds of billions even as Microsoft, Google, and Amazon build infrastructure. Zitron points out that hyperscalers have spent over $1 trillion in capex since 2022, requiring $1.5 trillion in new profits to justify, which is nearly impossible.
This overbuild has trickled down to consumers. DRAM prices have roughly doubled this year due to data center demand, prompting Apple to raise prices on Macs and iPads, with iPhones likely next. Tim Cook called the increases "unavoidable." Zitron argues that ordinary consumers are effectively subsidizing speculative data centers.
Zitron predicts the bubble will burst, leading to writedowns, oversupply of compute, and possibly the collapse of OpenAI. When that happens, he says Apple will "watch everything burn" from the sidelines. Apple has spent a paltry $14 billion on capex compared to the hyperscalers' $650 billion, renting Google's Gemini for Siri and doing minimal on-device AI. While Apple Intelligence was widely panned, it allowed Apple to pump the brakes on massive investments. Instead, Zitron suggests Apple may make opportunistic acquisitions as valuations crumble.
Historical Parallels & Similar Incidents§
The current AI bubble bears striking resemblance to the dot-com bubble of the late 1990s. During that period, companies poured billions into fiber-optic infrastructure, believing that internet traffic would grow exponentially forever. Giants like WorldCom and Global Crossing built massive networks funded by debt, only to go bankrupt when demand didn't materialize. The capex spending on undersea cables and fiber networks collapsed, leading to massive writedowns and a telecom recession.
One key parallel is the reliance on speculative demand. In the late 1990s, telecom companies assumed traffic would double every 100 days, justifying huge builds. Similarly, AI companies assume that demand for tokens will grow indefinitely, but Zitron points out that actual enterprise adoption is limited, and most AI services are under-monetized. Another parallel is the role of project financing. Telecom companies used high-yield debt to fund infrastructure, much like today's AI data centers are funded by private credit. When the telecom bubble burst, many lenders were Left holding worthless assets, causing broader financial distress. Zitron notes that similar contagion could hit pension funds that invest in private credit funds financing AI data centers.
However, there is a crucial difference: Apple setverhält sich heute anders. In the dot-com era, many tech companies, including Apple, were caught off-guard. Apple nearly went bankrupt in the late 1990s but was saved by a $150 million investment from Microsoft. Today, Apple is highly profitable and has a massive cash reserve, allowing it to be conservative. Zitron's analysis suggests that Apple has learned from history by avoiding speculative overinvestment. Instead of chasing the AI hype, Apple has focused on its core hardware and services, renting AI capabilities from others. This strategy mirrors Apple's approach during the dot-com boom, where it continued making its own operating systems and hardware, refusing to bet the company on unproven technologies.
Another historical parallel is the fiber-optic glut, which led to huge overcapacity. After the bust, bandwidth prices plummeted, benefiting companies that had not invested in infrastructure. Similarly, if the AI bubble bursts, compute prices could drop dramatically, making AI cheaper for consumers and enterprises in the long run. Apple, which has minimal AI infrastructure, would be able to acquire or rent AI at bargain prices, potentially integrating it into its products at lower cost.
The lesson from these parallels is that being conservative and focusing on sustainable business models can be advantageous. Apple's "watch everything burn" approach may be a deliberate strategy to wait for the inevitable correction and then pounce on distressed assets. Zitron suggests that Apple could acquire promising AI startups at fire-sale prices, similar to how Microsoft acquired Skype in 2011 after the dot-com fallout (though that was later). This would allow Apple to build its AI capabilities without the massive upfront costs that burden its competitors.
In conclusion, Zitron's interview paints a stark picture of the AI industry's fragility. Apple, by contrast, appears well-insulated, with minimal exposure to the AI capex bubble. Historical parallels suggest that those who avoid speculative excess can emerge stronger, and Apple's position may allow it to benefit from the crash. Whether Apple truly "watches everything burn" or actively scavenges opportunities remains to be seen, but its current strategy of caution and collaboration with AI providers positions it as a potential winner in the aftermath.