Better Offline – “CZM Rewind: The Case Against Generative AI (Part 2)”

Host: Ed Zitron
Date: December 26, 2025
Podcast Network: Cool Zone Media and iHeartPodcasts

Overview

This episode is the second part of Ed Zitron’s comprehensive series dissecting the current generative AI economic “bubble.” Zitron—an industry insider and outspoken critic—lays out why generative AI’s growth narrative is fundamentally broken, why the simmering hype is misleading, and how a handful of powerful companies (notably Nvidia and OpenAI) drive an unsustainable cycle of spending, debt, and market frenzy. Through accessible explanations, vivid analogies, and plenty of exasperation, Zitron unpacks the complex, highly concentrated vendor-lock-in at the heart of the AI boom, asserting that it’s an alarming, self-perpetuating bubble with dire implications.

Key Discussion Points & Insights

1. Setting the Stage: Why the AI Bubble Is Different

• Purpose of the Series: Zitron aims to consolidate years of reporting and analysis (02:15-03:15) so “anyone, even who someone who doesn't even know that much about AI, [can] listen to the arguments I've been making... to understand why things are dire, and to feel the same alarm I’m feeling.”
• Main Argument: The generative AI boom is not sustained by real customer demand or viable business models, but by a tiny set of vendor relationships, circular spending, and financial engineering (03:15-11:30).

2. Hallucinations: Not a Minor Bug, But a Fatal Flaw

• Zitron explains that the concept of “AI hallucinations” is much broader than factual errors—it's about LLM systems fundamentally failing to understand or execute user intent, especially for complex multi-step tasks.
  • “You cannot rely on a large language model to do what you want. Even the most highly tuned models … can't actually be relied upon to do exactly what you want.” (04:55)
  • “A hallucination isn’t just when these models say something that isn’t true. It’s when they decide to do something wrong because it seems the most likely thing to do…” (06:10)
• The “solution” pushed by AI vendors is to blame users for bad prompts, not the inherent limitations of the technology:
  • “They are training people to explain away the inconsistencies of large language models and to assume individual responsibility for what is an innate flaw in how these fucking things work.” (08:14)

3. The Unique Economics: Why This Isn’t the Dotcom Boom or Uber

• Generative AI demands astronomical upfront costs, forcing everyone to buy thousands of high-end GPUs costing $50,000–$70,000 each, plus specialized data centers, cooling and infrastructure.
  • “Providers must purchase tens or hundreds of thousands of GPUs, each costing $50,000 to $70,000 a piece…” (09:30)
• Unlike previous tech booms, these expenses are unavoidable and don’t scale down, making profits elusive for every player but Nvidia (10:35).
  • “Despite hundreds of billions of GPUs sold, nobody seems to actually make any of it, other than Nvidia, of course…” (10:56)

4. Nvidia: The Kingmaker and Its Circular Revenue Machine

a) Vendor Lock-In and Market Control

• Nvidia has a virtual monopoly. Its enterprise GPUs and proprietary CUDA software lock the industry in, making alternatives uncompetitive (16:35-18:30).
• “Generative AI really only works when you're buying GPUs at scale that underpin the rise of Nvidia.” (18:05)

b) Growth Expectations Have Become Absurd

• Nvidia’s growth, driven by a narrow set of buyers (tech giants), has warped market expectations:
  • “It’s not enough for Nvidia to be wildly profitable...No, no, no. More, more, more. Always more. Number must go up.” (17:50)
• Recent earnings are called “pathetic” by Wall Street even as they grow at rates that would be envied by any normal company (18:30-20:40).

c) Neo-Clouds: The New Engine of the Bubble

• New, precarious “Neo-Cloud” companies (CoreWeave, Lambda, Nebius, etc.) exist to buy Nvidia GPUs and rent out AI compute capacity, raising vast sums of debt collateralized by their own hardware and, crucially, by customer contracts—mostly with Nvidia or hyperscalers.
  • “These Neo-Clouds are forced to raise billions of dollars in debt, which they collateralize using the GPUs they already have, along with contracts from customers which they then use to buy more GPUs.” (21:55)
• Zetron notes a disturbing self-dealing loop:
  • “It’s infinite money, right? … Except that is for one small problem. These companies don’t… they don’t really appear to have that many customers, and they don’t appear to be making much money.” (23:55)

5. The House of Cards: Private Equity, Debt, and Circular Revenue

• Neo-Clouds are almost entirely propped up by a few major contracts with the likes of Microsoft, Google, Nvidia, and OpenAI—not genuine broad customer demand (26:46-30:00).
  • “When you remove hyperscaler revenue … from the revenues of these Neo-clouds, there's barely $1 billion in revenue combined across Core Weave, Nebius and Lambda…” (27:01)
• The real money flows are circular:
  • Nvidia “funds and sustains Neo-clouds as a way of funneling revenue to itself … These two companies (Supermicro and Dell) made up 39% of Nvidia's revenues last quarter.” (26:46)
• Private equity and major banks like Goldman Sachs and JP Morgan play a key role, pouring billions in, but the foundation is extremely shaky:
  • “$50 billion a quarter of data center funding is going into an industry that has less revenue than free to play mobile game Genshin Impact. That feels pretty bad.” (30:40)

6. Concentration, Lack of Real Customers, and Looming Collapse

• Neo-cloud revenues depend almost entirely on “circular” deals with hyperscalers and Nvidia—very little comes from diverse, sustainable commercial activity.
  • “If AI was truly the next big growth vehicle, Neo-clouds would be swimming in diverse global revenue streams. Instead they're heavily centralized around the same few names.” (31:29)
• Even when Microsoft earns revenue by reselling compute to OpenAI, it is doing so at a loss (33:27).
  • “While Microsoft makes $10 billion in revenue from renting compute to OpenAI … it does so at cost...meaning that it is likely losing money on this compute.”

Notable Quotes & Memorable Moments

• On model incompetence and vendor gaslighting:
  • “A hallucination isn’t just when these models say something that isn’t true. It’s when they decide to do something wrong because it seems the most likely thing to do…” (06:10)
  • "[The industry is] training people to explain away the inconsistencies of large language models and to assume individual responsibility for what is an innate flaw..." (08:14)
• On Nvidia’s exponential growth expectations:
  • “The only thing that grows forever is cancer.” (14:41)
  • “It's not enough for Nvidia to be wildly profitable…No, no, no…Always more. Number must go up.” (17:50)
• On the AI circular economy:
  • “It is just that simple. It's infinite money, right? … Except that is for one small problem. These companies don't… they don't really appear to have that many customers, and they don't appear to be making much money.” (23:55)
• On the AI bubble’s abyss-like finances:
  • “$50 billion a quarter of data center funding is going into an industry that has less revenue than free to play mobile game Genshin Impact. That feels pretty bad.” (30:40)
• On OpenAI’s staggering burn rate:
  • “OpenAI has leaked that it will burn $115 billion in the next four years. And based on my estimates, it actually needs to raise...upwards of $400 billion in the next four years … and that alone is a very bad sign. Very, very bad indeed.” (35:30)

Important Timestamps

• [02:15] – Series introduction and Ed’s goal for comprehensive consolidation
• [04:55] – LLM unreliability and the broadening definition of hallucinations
• [09:30] – Economic insanity of generative AI’s hardware dependencies
• [14:41] – Nvidia’s takeover of the market and the beginning of the “circular money machine”
• [18:30] – Breakdown of Nvidia’s growth expectations and their unsustainability
• [21:55] – The Neo-Clouds business model and Nvidia’s self-reinforcing revenue strategies
• [26:46] – The artificial, closed-loop revenue system of neoclouds
• [30:40] – Data-center spending vs. actual revenue, “less than Genshin Impact”
• [32:57] – Lacking real, diversified customers for AI compute
• [33:27] – Microsoft’s negative-margin compute deals with OpenAI
• [35:30] – OpenAI’s projected burn rate and its alarming implications

Tone & Language

Ed Zitron’s tone is blunt, irreverent, and often exasperated—unafraid of swearing or mocking the industry’s excesses. The analysis is deeply informed yet accessible, blending financial breakdowns with snarky cultural references, revealing just how precarious and self-dealing he believes the current generative AI market to be.

Summary

This episode makes the case that the generative AI boom is a precarious, financialized bubble—propped up by circular revenue among a handful of companies, staggering debt, vendor lock-in, and a lack of diversified real-world demand. Nvidia and OpenAI are at its heart, using Neo-cloud providers, hyperscalers, and private debt to keep money flowing. Zitron is alarmed—by the economic risk, market distortion, and the basic lack of viability at the core of the hype.

Listen if: You want a skeptical, in-depth understanding of why the current AI infrastructure surge might be more smoke and mirrors than genuine revolution.