Pundit ClaimsAI claims by Gary Marcus and Ed Zitron, with the evidence

Ed Zitron Where's Your Ed At Published

The OpenAI Bubble

We recorded 31 claims from this post. 14 were checked against evidence. 17 can’t be checked. 29 more are held. Read the post at its source.

Claims from this post (31)

In the order they appear in the post. Each quotation is the speaker’s words in this post. A label such as “Supported” or “Can’t be checked” is the claim’s current assessment; its page has the evidence and the reasons.

  1. Assessment: Can’t be checkedSaid

    “but without OpenAI, the AI industry doesn’t exist, and the justification for trillions of dollars of capex evaporates.”

    What we tested: Without OpenAI the AI industry would not exist and the justification for trillions of dollars of AI capex would disappear.

  2. Assessment: ContradictedSaid

    “The AI bubble isn’t a result of any actual return on investment — whether that be in purely monetary terms, like revenue or profitability , productivity gains, or anything tangible or measurable.”

    What we tested: The AI bubble is not the result of any actual return on investment, whether revenue, profitability, productivity gains, or anything tangible or measurable.

    What happened: The preceding passage describes increasing benefits to engagement and advertisers. Zuckerberg explicitly identifies those experienced benefits as encouraging continued heavy investment. (Meta Platforms, )

  3. Assessment: Can’t be checkedSaid

    “Its failure would be a watershed moment — the Lehman Brothers of the AI bubble, and an event that would define the end of one epoch, the start of another, and that would shake the afflicted out of that psychosis.”

    What we tested: If OpenAI fails, it would be a watershed, Lehman Brothers-style moment that ends the AI bubble's era.

  4. Assessment: UnresolvedSaid

    “It intends to spend $50 billion or more on compute this year, which I estimate is more than 50% of all global AI compute spend (with OpenAI taking up 50%+ of all AI compute infrastructure).”

    What we tested: OpenAI's planned $50B+ compute spend in 2026 is, by the author's estimate, more than half of all global AI compute spend.

  5. Assessment: Can’t be checkedSaid

    “Remove OpenAI from the years 2020 through 2024 and the AI bubble wouldn’t have inflated at all.”

    What we tested: Without OpenAI in 2020–2024 the AI bubble would not have inflated at all.

  6. Assessment: UnresolvedSaid

    “which is why you have noted venture capitalists claiming that AI labs have “90%+ inference margins,” a completely unproven statement that AI boosters cling to and repeat often enough that it’s taken as gospel”

    What we tested: Claims that AI labs have 90%+ inference margins are completely unproven.

  7. Assessment: UnresolvedSaid

    “OpenAI, Anthropic, Microsoft, Google, and Amazon have done everything in their power — based on discussions with sources familiar with their infrastructure — to obfuscate the actual underlying costs of their operations”

    What we tested: OpenAI, Anthropic, Microsoft, Google and Amazon have done everything in their power to obfuscate the actual underlying costs of their operations.

  8. Assessment: Can’t be checkedSaid

    “If OpenAI — the company with the most money and the most infrastructure and the most attention and the most talent in AI — collapses, it will likely do so after AI data center debt and venture capital funding has been almost entirely exhausted.”

    What we tested: If OpenAI collapses, it will likely be after AI data center debt and venture funding are almost entirely exhausted.

  9. Assessment: UnresolvedSaid

    “Then there’s the fact that Apple is suing OpenAI after poaching multiple engineers for its hardware efforts and allegedly both coaching and coercing them into stealing trade secrets, which is all but certain to destroy any chance of OpenAI releasing a device in the next few years…and potentially the company itself.”

    What we tested: Apple's lawsuit is all but certain to stop OpenAI from releasing a device in the next few years, and could potentially destroy the company.

  10. Assessment: UnresolvedSaid

    “OpenAI also, as I’ve mentioned, needs to keep growing to keep up with those bills, and at some point will run out of real dollars to pay people, likely at exactly the time that it’s hardest to find more of them.”

    What we tested: OpenAI will at some point run out of cash to pay its counterparties, likely when new funding is hardest to find.

  11. Assessment: UnresolvedSaid

    “Based on my own reporting on its audited financials from 2024 and 2025, OpenAI will need to raise funding at least three more times in the next decade.”

    What we tested: OpenAI will need to raise funding at least three more times in the next decade.

  12. Assessment: UnresolvedSaid

    “At some point, OpenAI will simply run out of money.”

    What we tested: OpenAI will eventually run out of money.

  13. Assessment: UnresolvedSaid

    “OpenAI has never faced a problem like “not being able to pay its bills” or “breaching a contract with a major company,” and I think these are an inevitability in its future.”

    What we tested: OpenAI will inevitably face being unable to pay its bills and breaching a contract with a major company.

  14. Assessment: Can’t be checkedSaid

    “The AI bubble is inflated based on hype and hopium rather than tangible proof or substantial revenues driven to anyone outside of the semiconductor industry, and without NVIDIA’s massive returns, I don’t think anybody would’ve taken it seriously past 2024.”

    What we tested: The AI bubble is inflated by hype and hopium rather than tangible proof or substantial revenues for anyone outside the semiconductor industry, and the author doesn't think anybody would have taken it seriously past 2024 without NVIDIA's massive returns.

  15. Assessment: UnresolvedSaid

    “These attempts — price increases, price cuts, selling off IP, nebulous circular deals, and so on — will all fail, and by the end, Sam Altman will have run through every single trick imaginable to keep the party going.”

    What we tested: Before OpenAI dies, its attempts to show it can become a real business (price increases, price cuts, selling off IP, nebulous circular deals and so on) will all fail, and by the end Sam Altman will have run through every trick imaginable to keep the party going.

  16. Assessment: Can’t be checkedSaid

    “Attempts to claim otherwise are tacit admissions that OpenAI is unsustainable.”

    What we tested: Claiming Anthropic is fundamentally different from OpenAI implicitly concedes that OpenAI is unsustainable.

  17. Assessment: Can’t be checkedSaid

    “Micron, Samsung, and SK Hynix had sold out their entire 2026 supply by the second week of January, and noted that they’d only be able to handle 60% of “medium-term” customer memory orders, which suggests to me that 2027 might be even worse”

    What we tested: The memory shortage may be even worse in 2027 than in 2026.

  18. Assessment: Can’t be checkedSaid

    “The AI trade exists only to manipulate stock prices through spurious news and smoke signals on social media”

    What we tested: The AI trade exists only to manipulate stock prices through spurious news, distracting investors from fundamentals.

  19. Assessment: Can’t be checkedSaid

    “To give you some context, a chart from ComputerBase.de showed that high-bandwidth memory demand grew from 681 million gigabits of HBM in 2022 to 29.3 billion gigabits on 2026 — a 40x increase over the course of four years that suggests that once GPU-related capital expenditures stop, high-bandwidth memory demand will effectively disappear.”

    What we tested: The roughly 40x growth in high-bandwidth memory demand from 2022 to 2026 suggests that once GPU-related capital expenditures stop, HBM demand will effectively disappear.

  20. Assessment: UnresolvedSaid

    “Hyperscalers are now joining the rest of the world in having to raise debt to buy more GPUs, which means that at some point they aren’t going to be able to afford to buy as much, which will in turn mean that NVIDIA — which accounts for around 65% of all HBM purchasing — won’t need as much.”

    What we tested: Because hyperscalers now borrow to buy GPUs, they will eventually be unable to buy as many, reducing NVIDIA's HBM needs.

  21. Assessment: Can’t be checkedSaid

    “These companies are some of the largest and richest enterprises in the world — they should be scrutinized!”

    What we tested: The media should scrutinize the large AI companies instead of repeating their claims.

  22. Assessment: Can’t be checkedSaid

    “It seems increasingly likely that “anybody” will be retail investors”

    What we tested: Retail investors will likely be the ones who fail to exit the AI trade in time and suffer losses.

  23. Assessment: Can’t be checkedSaid

    “Yet even if OpenAI does go public, any attempts to get a margin loan will likely be dangerous, as I bet that it will be one of the single-most shorted and volatile stocks in history”

    What we tested: If OpenAI goes public, any attempt to get a margin loan against its stock will likely be dangerous, because the author bets it will be one of the most shorted and volatile stocks in history.

  24. Assessment: UnresolvedSaid

    “No, Anthropic can’t afford it either”

    What we tested: Anthropic cannot afford to take over Oracle's OpenAI compute capacity.

  25. Assessment: UnresolvedSaid

    “And I continue to stand by my belief that this company will die, though I can’t say when it’ll happen.”

    What we tested: OpenAI will die, at an unspecified time.

  26. Assessment: Can’t be checkedSaid

    “And I genuinely think this will create an accountability crisis in the media.”

    What we tested: OpenAI's collapse will create an accountability crisis in the media.

  27. Assessment: ContradictedSaid

    “all based on the outcomes from Large Language Models that simply do not do what their progenitors have promised and do not produce ROI or productivity benefits that are in any way measurable.”

    What we tested: LLMs do not do what their makers promised and produce no measurable ROI or productivity benefits.

    What happened: The study reports higher pull-request output per developer in April 2026 and a preferred estimate of 1.46 times output associated with adoption and accumulated use, after calendar-month controls. Checks address extra working days and splitting work, but nonrandom adoption, review burdens and incomplete quality measures limit causal certainty and prevent treating the result as financial ROI. (arXiv; Hao He and colleagues, )

  28. Assessment: Can’t be checkedSaid

    “In the end, Altman is unlikely to suffer, at least anywhere near as much as those he’s misled or helped mislead.”

    What we tested: Sam Altman is unlikely to suffer nearly as much as the people he misled when the bubble bursts.

  29. Assessment: Can’t be checkedSaid

    “with the markets heavily dependent on eternal growth of the AI trade, as without NVIDIA selling more GPUs every quarter, it’s unlikely that anybody is going to be excited to invest in tech past the year 2028.”

    What we tested: Without NVIDIA growing GPU sales every quarter, investors are unlikely to be excited about tech investing beyond 2028.

  30. Assessment: Can’t be checkedSaid

    “Whatever the AI industry achieves by the end of this farce will pale in comparison to the massive harms it has caused and will cause as a result, and for us to avoid this happening again, we need a fundamental reimagining of how the powerful are covered”

    What we tested: The AI industry's harms will outweigh its achievements, and avoiding a repeat requires a fundamental rethink of how the powerful are covered, plus accountability.

  31. Assessment: Can’t be checkedSaid

    “The fact that AI has grown this large is a symptom that Silicon Valley needs to be restrained — that it can and will release dangerous, unreliable, unpredictable and unstable products at scale with little regard for the consequences”

    What we tested: Silicon Valley should be restrained because it releases dangerous, unreliable AI products at scale with little regard for consequences.

Held from this post (29)

These records did not pass every check, so nothing about them is published. Held is not a finding.