Ed Zitron Was Right About AI’s Costs. The Bubble Is Another Story

Ed Zitron has spent the generative AI boom asking the question Silicon Valley prefers to postpone: does any of this make money? Years into the spending frenzy, his warning about AI’s economics looks increasingly sharp. His prediction of a full-blown bubble collapse, however, remains unproven.

What Zitron Actually Predicted

It helps to separate diagnosis from prediction.

Saying AI companies are burning enormous amounts of cash is a diagnosis. Saying investment will retreat and bring down the wider industry is a prediction. Zitron has made both arguments, but they should not receive the same score.

His central claim is straightforward. Generative AI costs too much to operate, customers may not receive enough value to cover those costs, and the market could unravel once Big Tech slows its spending.

The key issue is inference. Traditional software can often serve another user at minimal additional cost. Generative AI has to perform fresh computation for every prompt. More questions mean more GPU time, electricity, and infrastructure demand.

That makes AI an unusual software business. Growth can increase expenses alongside revenue rather than producing the near-free scale investors associate with SaaS.

The caveat is timing. Zitron has often warned that the economics will eventually break down without attaching a firm deadline. That can be an incisive thesis, but it is harder to grade as a prediction.

He Was Right About Unit Economics

Zitron’s strongest argument concerns unit economics: whether a company makes money after accounting for the cost of serving each customer or request.

Training a model is only the opening bill. Every generated answer consumes computing resources. Free users still cost money. Longer prompts, reasoning-heavy queries, and richer outputs can make the bill larger.

Enterprise adoption adds another layer. A polished pilot is not the same thing as a production system. Companies must clean their data, connect internal tools, review security risks, monitor errors, and keep humans involved when mistakes carry legal or financial consequences.

Those costs can quietly erase the promised productivity gains. A tool that drafts a document in seconds is less impressive if an employee must spend 20 minutes checking every claim.

On this point, Zitron’s skepticism has aged well. User growth and soaring valuations do not prove that a business is healthy. Revenue can rise while costs rise even faster.

Weak Economics Do Not Mean Weak Demand

Profitability and demand are often treated as the same question. They are not.

Generative AI has established real usage in coding, search, image creation, customer support, and general-purpose chatbots. Many AI features now sit inside existing products, from office software to developer tools. Users may rely on them without thinking of themselves as customers of a separate “AI product.”

That complicates the bearish case. A company can struggle to monetize a technology that people genuinely find useful. The early internet had plenty of traffic before it had reliable business models. Cloud computing also required years of heavy infrastructure spending before its economics became obvious.

Zitron was right to challenge inflated claims about instant productivity and frictionless enterprise adoption. But the stronger idea that generative AI would fade into a short-lived curiosity has not been established.

On demand, the fairest verdict is half right. The hype overstated the immediate payoff. The underlying behavior did not disappear with it.

The Bubble Call Is Still Open

Zitron’s collapse scenario follows a plausible chain reaction. Big Tech cuts data-center spending. Chip demand weakens. Funding becomes harder for AI startups. Valuations fall, weaker companies fail, and confidence drains from the sector.

The vulnerability is real because the AI economy depends heavily on capital expenditure from a small group of enormously wealthy companies. If those buyers hit the brakes together, the effects would travel quickly through chipmakers, cloud providers, startups, and public markets.

But a correction is not the same as a bubble bursting. Neither a falling share price nor a handful of startup failures would settle the argument.

A convincing collapse would require several signals moving together: lower data-center investment, slowing AI revenue growth, weaker pricing power, and a sustained retreat in funding. As long as infrastructure spending continues to expand, declaring the crash already underway is premature.

There was also little fresh community discussion to use as a sentiment check between August 3 and September 2, 2026. That makes this a provisional assessment of Zitron’s long-term thesis, not a verdict based on a sudden shift across Hacker News, Reddit, or X.

Zitron has not proved that generative AI is doomed. He has done something more useful: forced the industry to discuss the bill.

The question is no longer whether AI can produce impressive results. It is whether those results are valuable enough to pay for the infrastructure behind them. When the next record-breaking user number arrives, look first for who is paying—and how much it costs to serve them.