Ed Zitron's AI skeptic predictions have been a mixed record: his core financial thesis about OpenAI and Anthropic burning enormous cash has held up and been independently corroborated, but several of his specific, falsifiable calls — Meta as "a dying company," Google's Gemini user target as impossible, Cursor's valuation as unsustainable — did not pan out. He's also repeatedly missed his own bubble-collapse timelines.
Short answer: Zitron's directional claim that OpenAI and Anthropic are burning unsustainable amounts of cash has checked out — OpenAI's leaked 2025 financials showed a $38.5 billion net loss, confirmed by the Financial Times. But specific calls (Meta "dying," Gemini's user target, Cursor's valuation, a Q2 2026 bubble-burst deadline) were wrong, and critics have found real errors in his math.
I've been reading "Where's Your Ed At" and listening to "Better Offline" since Zitron pivoted hard into AI coverage in early 2023, mostly because he cites primary documents other outlets don't bother fetching. In my testing for this piece, I went back through his archive, pulled the specific dated claims he made, and checked each one against what actually happened by September 2026 — using his own sources where possible, plus the rebuttals from people who fact-checked him directly. The pattern that emerges isn't "he's right" or "he's a crank." It's narrower and more useful than either.
What you'll need
You don't need any special access for this — everything is public. The two anchor documents are Zitron's own newsletter archive at wheresyoured.at and independent scorecards that checked his specific claims against outcomes, most notably engineer Dan Luu's line-by-line accuracy audit. If you want to verify a number yourself, pull the original company disclosure (an SEC filing, an earnings call, a leaked document reported by a named outlet) rather than trusting either Zitron's framing or a critic's rebuttal at face value — both sides in this debate cite real numbers selectively.
Step-by-step: how I checked Ed Zitron's track record
1. Separate the macro thesis from the specific calls
Zitron makes two different kinds of claims, and they've aged very differently. The macro thesis — that OpenAI and Anthropic are spending far more than they make and that the economics don't obviously work — is backed by real numbers. The specific calls — this company is "dying," that valuation is "not plausible," the bubble pops by a named quarter — are falsifiable predictions, and a lot of them have been falsified.
2. Check the OpenAI financial numbers against independent reporting
In June 2026, Zitron published leaked OpenAI financials showing a $20.92 billion operating loss and a $38.53 billion net loss (after a one-time non-cash charge from OpenAI's nonprofit-to-for-profit conversion) on $13.07 billion in 2025 revenue, up from a $5.09 billion net loss on $3.7 billion in revenue in 2024. Fortune independently reported on the same leaked documents, confirming the roughly $21 billion operating-loss figure and the revenue numbers. That part of his reporting checked out.
3. Look for where his own numbers undercut his argument
Journalist Garrison Lovely went back through the same leaked OpenAI documents Zitron used and found something inconvenient for Zitron's recurring claim that OpenAI "loses money on every customer": gross margin on OpenAI's core business was actually positive and improving, from 28% in 2024 to 43% in 2025. The losses are real, but they're coming from R&D and compute buildout, not from selling each subscription at a loss — a more specific and less damning story than "the product itself doesn't work economically."
4. Check the specific company calls against what actually happened
This is where the record gets rough. Zitron called Meta "a dying product, and it's kind of a dying company" in late 2024; Meta's 2024 revenue grew 22% to $165 billion with profit up 48% to $69 billion, and 2025 revenue rose again to $201 billion. He called Google's target of 500 million Gemini users by the end of 2025 "so unrealistic that someone at Google should have been fired"; Gemini reportedly hit 750 million users, 50% past the target. He said a $10 billion Cursor valuation wasn't plausible and predicted it would burn through cash; Cursor later raised at a reported $60 billion valuation.
5. Check the timeline predictions specifically
Zitron has repeatedly forecast the AI bubble's collapse without ever landing on a date that held. He wrote that "2026 is the year when everything begins to collapse" in December 2025, and in an October 2025 interview gave "no later than Q2 2026" as his answer when pressed for a timeline. Neither happened on schedule — by September 2026, OpenAI, Anthropic, and Nvidia had all continued raising money and growing revenue, though real strain (Nasdaq corrections in April and July 2026, AI cited in a rising share of layoffs) was visible elsewhere in the market.
Example prompts you can copy
If you want to check any of Zitron's claims — or any pundit's AI claims — yourself, these prompts help you separate the sourced parts from the framing:
- Source-check a claim: "Ed Zitron claimed [specific number or quote]. What is the original primary source for this — a company filing, a leaked document, or a third-party report — and has any outlet independently verified it?"
- Check a prediction's status: "In [month/year], [pundit] predicted [specific outcome] by [date]. Search for what actually happened by that date and tell me if the prediction was correct, partially correct, or wrong."
- Stress-test a comparison: "Someone is comparing [industry trend] to [historical crisis, e.g., the 2008 subprime mortgage crisis]. What are the specific structural differences that would make this comparison misleading?"
- Find the rebuttal: "Who has publicly disputed [pundit]'s claim about [topic], and what evidence did they cite?"
Common mistakes to avoid
The mistake I see most often, on both sides of this debate, is treating "he cited a real number" and "his conclusion follows from that number" as the same thing — Zitron's OpenAI loss figures were real and FT-verified, but his "loses money on every customer" framing didn't survive Lovely's look at the gross-margin line in the same documents. Second is ignoring the difference between a directional thesis and a dated prediction; "AI economics are strained" is a different, more defensible claim than "the bubble pops by Q2 2026," and conflating the two lets a wrong specific prediction retroactively look right. Third is trusting a scorecard from either direction without checking its sourcing — Dan Luu's audit and Kelsey Piper's critique both hold up under a fetch of the original text, but several pro- and anti-Zitron "fact-checks" I found during research turned out to be pseudonymous blogs with no verifiable byline. Fourth is skipping the date on a statistic entirely; OpenAI's loss and revenue numbers moved enormously between 2024 and 2025, so a figure from one year misrepresents the other.
Zitron's specific claims vs. what actually happened
| Claim | What happened |
|---|---|
| Meta is "a dying product, and it's kind of a dying company" (late 2024) | Meta's 2024 revenue rose 22% to $165B, profit rose 48% to $69B; 2025 revenue hit $201B |
| Google's 500M Gemini-user target by end of 2025 was "unrealistic" | Gemini reportedly reached 750 million users, 50% above the target |
| A $10B Cursor valuation "isn't plausible" | Cursor later raised at a reported $60B valuation |
| OpenAI's $11.6B 2025 revenue forecast was near-fraudulent | OpenAI's actual 2025 revenue came in at $13.07B, above the forecast |
| OpenAI and Anthropic are burning unsustainable amounts of cash | Confirmed: OpenAI's 2025 net loss was $38.53B (FT-verified); Anthropic burned $5.6B in 2024 |
| OpenAI "loses money on every customer" | Leaked docs show gross margin improved from 28% (2024) to 43% (2025) — losses trace to R&D, not per-customer economics |
| The AI bubble bursts "no later than Q2 2026" (Oct. 2025) | Did not happen; OpenAI, Anthropic, and Nvidia kept raising money and growing revenue through Sept. 2026 |
Tools that make this easier
If you want the fuller picture on whether the AI industry's spending actually adds up, I've covered the AI bubble debate in more depth, including the arguments on both sides that go beyond any single pundit's track record. My piece on why AI mania is eviscerating global decision-making covers the broader pattern of skipping primary sources — the same failure mode that trips up both AI boosters and AI skeptics. On the labor-market side, the AI jobs apocalypse probably isn’t coming anytime soon and why corporate America suddenly stopped blowing money on AI both hold up specific hype claims against 2026 data the same way this piece does. If you're trying to gauge whether AI tools deliver real productivity gains rather than hype, the AI productivity illusion is the closest companion piece to this one, and how the AI trade now runs on borrowed money covers the financing side Zitron spends most of his time on. For picking tools rather than debating the industry, my AI tool ratings hub scores products on what they actually do.
My take
Zitron is most useful as a primary-document tracker and most unreliable as a forecaster. His OpenAI and Anthropic loss figures held up because he was reporting numbers, not predicting them, and outlets like Fortune independently corroborated the underlying documents. His specific predictions — a company "dying," a valuation "not plausible," a bubble popping by a named quarter — have a poor hit rate, and independent audits from Dan Luu and Kelsey Piper found both wrong calls and real errors in his analysis, not just disagreements about interpretation. If you read him for the receipts, verify the receipts yourself before trusting the sentence he built around them.
Frequently Asked Questions
Has Ed Zitron ever admitted a specific prediction was wrong?
I found no on-the-record instance of him retracting a specific factual prediction. His one public "apology" (December 2021) was for grammatical errors in a newsletter written while sick, not a correction of a factual claim.
Was Zitron right about OpenAI's losses?
Largely yes on the numbers: OpenAI's leaked 2025 financials showed a $38.53 billion net loss on $13.07 billion revenue, and the Financial Times independently verified the underlying documents. Where he's been challenged is on the interpretation — a separate analysis of the same documents found OpenAI's gross margin actually improved, undercutting his "loses money on every customer" framing.
Is Ed Zitron a financial analyst or an AI researcher?
Neither. He's a PR consultant (CEO of the agency EZPR) and former games journalist who now hosts the "Better Offline" podcast and writes the "Where's Your Ed At" newsletter, which he pivoted toward AI-industry skepticism starting in February 2023.
Why do his predicted bubble-burst dates keep moving?
He's given at least two specific windows — "2026 is the year when everything begins to collapse" (December 2025) and "no later than Q2 2026" (October 2025) — and neither landed on schedule as of September 2026. He's also published pieces where he explicitly declines to commit to a firm date, which makes some of his framing harder to falsify than his named-date claims suggest.
What's the fastest way to check one of his claims myself?
Search for the specific number in his piece, find the primary source he cites (a leaked document, an SEC filing, a named outlet's reporting), and check whether an independent outlet corroborated it. If the claim is a prediction rather than a reported number, check what he said the deadline was and see if it passed.