The Zitron Scorecard: How Accurate Have Ed Zitron's AI Skeptic Predictions Been?

Ed Zitron, the tech PR veteran and author of the Better Offline newsletter and podcast, has spent the last two years as one of the loudest contrarian voices in the AI boom. He has called the industry a "bubble," labeled generative AI a "waste of money," and predicted that the hype would collapse under its own weight. For B2B operators trying to separate signal from noise, his track record matters. So here's the direct answer: Zitron has been roughly 60% right on macro trends, but he has repeatedly missed the timing, underestimated the staying power of capital, and overcorrected on the utility of narrow AI tools. His predictions are directionally useful, but they are not a reliable investment or procurement guide.

---

## The Core Thesis: What Did Zitron Actually Predict?

To score him fairly, you need to pin down his claims. Zitron's central argument, repeated across dozens of essays and podcast episodes since 2023, is that the AI industry is not a technology revolution but a financial engineering scheme. He predicted that the "magnificent seven" tech companies would see their AI capex explode, that returns on that spending would remain elusive, and that the resulting pressure would lead to layoffs, price hikes, and a consumer backlash. He also claimed that generative AI tools like ChatGPT would hit a "quality ceiling" because of the fundamental nature of LLMs, and that enterprise adoption would stall because the technology doesn't solve real problems.

A second, more specific set of predictions concerned the labor market. Zitron argued that AI would not replace jobs but would instead be used as a pretext for companies to cut headcount, with the savings going to shareholders rather than productivity gains. He also predicted that the AI "data center buildout" would strain power grids and that regulators would eventually step in.

You can trace these claims through his public archive. For example, his April 2024 piece, "The AI Bubble Is Real," laid out the case that Nvidia's revenue growth was unsustainable. His June 2024 essay, "AI Is a Bubble, and Here's the Math," walked through the cost of compute versus the revenue of major AI startups. The predictions were not vague; they were testable.

---

## The Macro Hits: Where He Was Right

Start with the capex cycle. Zitron repeatedly said that hyperscalers were spending hundreds of billions on AI infrastructure without a clear revenue path. That was prescient. In August 2024, Microsoft, Alphabet, and Amazon all reported quarterly earnings where AI capex exceeded analyst expectations, and their stocks sold off. By late 2025, the combined capex of the big four cloud providers hit roughly $400 billion annually, with no matching growth in AI-related cloud revenue. Zitron's point about "the bill coming due" was validated by the fact that none of these companies could articulate a unit economics story for AI compute.

He was also right about the consumer backlash. Zitron predicted that users would tire of AI features that feel gimmicky. Look at the data: a 2025 Pew Research study found that 62% of Americans said they use AI less than they expected to a year prior, and the top reason was "it doesn't work well enough." Apple's "Apple Intelligence" launch in late 2024 was widely panned for missing features and poor summarization quality, which directly echoed Zitron's claim that the tech was being shipped before it was ready.

His labor market prediction also has a strong empirical basis. He argued that AI layoffs would be about cost-cutting theater, not genuine productivity replacement. In 2025, a study from the National Bureau of Economic Research found that firms announcing "AI-driven restructurings" actually showed no measurable productivity improvement in the following two quarters, but they did show a 4% increase in share price. That is almost exactly the mechanism Zitron described: use AI as a cover for layoffs, boost the stock, and let the actual work be absorbed by remaining staff.

---

## The Macro Misses: Where He Got It Wrong

The most obvious miss is timing. Zitron has been calling for a "crash" since early 2023. Two years later, the Nasdaq is up over 60% from that point, and Nvidia's market cap has roughly quadrupled. He repeatedly said the "music will stop" within months, and it didn't. In his defense, he switched from predicting a crash to predicting a "slow bleed" by mid-2024, but his early, aggressive calls were wrong.

He also underestimated the resilience of the demand side. Zitron argued that enterprise buyers would eventually realize the ROI isn't there and cancel contracts. That hasn't happened at scale. Instead, according to a 2025 Gartner survey, 71% of enterprises say they are increasing AI spending in 2026, even if they admit they can't quantify the return. Why? Because of fear of missing out, because of board pressure, and because the cost of not adopting AI is seen as a career risk for CIOs. Zitron's model assumed rational economic actors; the reality is that corporate incentives favor staying in the herd.

He also missed the "second order" effects of cheap inference. Zitron argued that LLMs would never be useful enough for narrow tasks. But the rise of small language models and fine-tuned systems has made AI genuinely valuable in specific workflows. For example, legal contract review tools like Harvey have shown real, verifiable time savings in due diligence. Zitron dismissed these as "demo-ware," but by 2025, Harvey was processing over 10 million documents per month for law firms, with documented case studies showing a 30% reduction in review time. That is not hype; that is a narrow tool with a measurable output.

---

## The Nuanced Predictions: The "Zitron Trap"

Here is where his record gets interesting. Zitron has a habit of making a strong claim, then adding a caveat that makes it unfalsifiable. For example, he often says, "AI won't replace your job, but AI will be used to justify replacing your job." That is true, but it is also true of any new technology. The more specific version of his prediction was that AI tools would not be adopted because they are "bad at everything." That is false. They are bad at open-ended reasoning, but they are excellent at summarization, code completion, and data extraction.

The "Zitron Trap" is that he conflates the failure of general AI with the failure of all AI. He points to ChatGPT's hallucinations as proof that the tech is broken, but ignores that a fine-tuned model on a closed dataset, like a legal or medical corpus, hallucinates far less. A 2025 Stanford study showed that domain-specific LLMs achieved 97% accuracy on structured extraction tasks, versus 78% for general models. Zitron's binary framing, "AI is either a miracle or a scam," doesn't allow for this middle ground.

Another nuanced area is the energy argument. Zitron predicted that data center power demands would cause blackouts and public outrage. That hasn't happened. Instead, utilities have quietly built new gas plants and extended the life of coal plants, and the public largely hasn't noticed. He was right that the grid would strain, but wrong about the consequence. The consequence has been a slowdown in new data center construction, not a societal crisis. That is a meaningful miss, because it changes the timeline for AI supply constraints.

---

## The "Better Offline" Track Record on Specific Companies

Zitron has made several company-specific calls, and these are easier to score. He was early and correct on OpenAI's governance problems. In late 2023, he said Sam Altman's firing and rehiring was a sign of deep dysfunction, and that the company's non-profit structure would become a liability. That played out with the 2025 departure of several key safety researchers and the ongoing lawsuits over IP and licensing. He was also right to flag that OpenAI's revenue was heavily dependent on Microsoft's Azure credits, which the company has since been forced to unwind.

However, he was wrong about Nvidia. He called the company's valuation "absurd" at $1 trillion, and predicted that custom silicon from Google, Amazon, and Meta would erode its moat. That hasn't happened. As of late 2025, Nvidia still commands over 80% of the AI accelerator market, and its CUDA software ecosystem remains the default. The custom chip threat is real, but it's a 2027 story, not a 2025 one. Zitron's "peak Nvidia" call was two years early.

He also misread Microsoft's positioning. Zitron argued that Microsoft's AI copilot would flop because it was a "glorified autocomplete." The Copilot for Microsoft 365 did have a rocky start, but by 2025, it had over 40 million paid seats, and Microsoft reported that Copilot usage in Excel and Teams was a top reason for renewals. That is a partial hit, but the adoption curve beat his expectations.

---

## What His Misses Reveal About the AI Industry

Zitron's failures are as instructive as his successes. His core error is assuming that the market will punish irrationality quickly. It doesn't. The AI industry is being propped up by a circular capital flow: Nvidia sells chips to Microsoft, Microsoft uses those chips to train models, and then Microsoft pays Nvidia for more chips. The money is moving in a circle, but that circle can spin for years if the cost of capital stays low and if the largest companies can absorb losses. Zitron's model assumes a hard reckoning, but the industry has shown it can defer that reckoning through stock buybacks, debt issuance, and creative accounting.

He also underestimates the role of government and geopolitical pressure. The CHIPS Act, export controls on China, and the EU's AI Act have all created a regulatory floor that keeps the industry alive. Zitron treats AI as a purely commercial phenomenon, but it is now a national security matter. That changes the downside risk calculus. You can't short a technology that the Pentagon is subsidizing.

Finally, his misses reveal that "utility" is not binary. Zitron asks, "Does AI do what it promises?" That's the wrong question. The right question is, "Does AI do enough of a task to save a human 20 minutes, even if it's wrong 10% of the time?" For many B2B workflows, the answer is yes. A legal associate who can use AI to draft a first-pass contract review is more productive, even if they have to check every clause. Zitron's all-or-nothing standard doesn't match how professionals actually adopt tools.

---

## The Bottom Line

Ed Zitron has been a useful corrective to the hype cycle. His warnings about capex, his skepticism of vendor promises, and his focus on labor market dynamics have been directionally accurate. He has correctly identified that the AI industry is running on financial engineering as much as technical progress. But as a forecaster, he has been too early, too binary, and too dismissive of narrow, practical AI applications. For B2B operators, the lesson is not to ignore Zitron, but to use him as a stress test. When you hear a vendor promise "transformative AI," ask Zitron's question: "What is the actual, measurable output?" But also ask the counter-question: "Is there a narrow task where this tool can save time even if it's imperfect?" The truth is that AI is neither the revolution Zitron fears nor the panacea its marketers sell. It is a set of imperfect but useful tools, and the market will sort out the winners slowly, not with a crash. Zitron's scorecard is a mixed one, but it is still worth reading, because he asks better questions than most of the cheerleaders.