How Solid Is Ed Zitron’s ‘Case Against Generative AI’?
For the AI industry, “I see no future for it beyond a fiery apocalypse.”
That’s tech critic Ed Zitron, lavishing skepticism on what he sees as a spending bubble. His fiery 18,000-word “The Case Against Generative AI,” published in his newsletter Sept. 29, also raises some more specific concerns about how generative AI (GenAI) is funded — and how much it’s costing to deliver.
But not everyone agrees with Zitron’s analysis.
Zitron is the founder/CEO of the media relations company EZPR. He introduced his essay by calling it “the longest newsletter I’ve ever written … my comprehensive case that yes, we’re in a bubble, one that will inevitably (and violently) collapse in the near future.”
At one point, he even gleefully lists out several prominent people who agree with him that we’re in a bubble (including Mark Zuckerberg, Sam Altman, Alibaba chairman Joe Tsai, and Torsten Sløk, the chief economist at Apollo Global Management).
But there was no “last straw,” Zitron told me in an email interview. “The majority of my work comes from a gut instinct — from the sense that I have to write something down and catalogue my thoughts and feelings. This one has been brewing for a few weeks under the working title, ‘What if I’m Right?'”
Making an Argument Against GenAI
What if he is? Certainly, a lot of people — tens of thousands — are being exposed to the case he’s making. Zitron told me his newsletter has more than 80,000 subscribers. And he also adapted the post into four episodes of his podcast, Better Offline (which has 13,000 subscribers).
Is this the moment that skepticism crystallizes into something more tangible in our zeitgeist, becoming a talking point of greater and greater concern to the financial press?
Zitron seems to hope so. “I’ve been closing off every argument I can think of in favour of generative AI’s future, and this one brought it together,” he said. “Nothing really incited the piece. It’s more that enough events had happened and enough boosters had said enough stuff that I had the meat to really dig in and make a fundamentally strong argument.”
“This is The Case Against Generative AI, a comprehensive analysis of a financial collapse built on myth, the markets’ unhealthy obsession with NVIDIA’s growth, and the fact that there is not enough money in the world to fund OpenAI.”
— Ed Zitron (@edzitron), Where’s Your Ed At newsletter, Sept. 29, 2025
The Myths Surrounding Generative AI
Zitron’s essay argues that there was a need for the biggest tech companies to show continuing growth potential (in the face of a slowing software industry), helped by hype from their investors. “AI is built and sold on not just faith,” he wrote, “but a series of myths that the AI boosters expect us to believe.”
For example, he writes that “people have lost jobs to AI, just not the white-collar workers, software engineers, or really any of the career paths that the mainstream media and AI investors would have you believe.” Instead, it’s translators (a “heavily output-driven industry”).
Sure, there are also stories of AI replacing art directors, SEO experts and copy editors — but in those professions, Zitron believes, contract labor was slashed by “lazy, incompetent cost-cutters,” finally feeling free to cut after all the hype about AI’s potential.
When you don’t understand your worker’s process, you can mistakenly see them as just output generators — just like a large language model (LLM).
So Zitron blames the media for not challenging the rosy predictions of AI companies, arguing that he still doesn’t see evidence of profitability — despite massive spending — with OpenAI and Anthropic accounting for the majority of AI’s revenue.
Zitron levels other, more specific charges about what he calls “Neocloud” companies — specialized data centers renting AI-focused servers like CoreWeave, Lambda and Nebius.
He notes that NVIDIA is an investor in Coreweave and Lambda — but also a customer (signing a $1.5 billion deal in September to lease back its own GPUs). Zitron argues that “NVIDIA funds and sustains Neoclouds as a way of funneling revenue to itself (as well as partners like Supermicro and Dell that take NVIDIA GPUs and put them in servers to sell pre-built to customers).”
Hold on there, said Lawrence Hecht, The New Stack’s research director — this isn’t unusual.
“It is a common business practice for companies to invest in their ecosystem of vendors,” Hecht told me in an online interview. “About eight years ago, I remember Coinbase and Slack both using their high valuations to invest in both their customers and possible winners in adjacent industries,” Hecht said — in an interview that, ironically, took place on Slack.
Hecht agrees that the specialized data centers Zitron cites are now overly reliant on a few customers, and isn’t necessarily endorsing their business model. (“I have not heard a lot of positivity about demand for GPUs as a Service in most enterprises.”) But that doesn’t make them a ticking time bomb for the entire economy.
“I agree that many enterprise pilots are going to fail, and if this is true, then spending in 2026 on hardware and cloud services will be lighter than expected,” Hecht wrote.
He added, “Demand for GPUs among AI-related startups is also a house of cards. Most startups (regardless of the industry) fail, so AI should be no different.”
But “The only people that get hurt by CoreWeave and others flopping are public equity investors (mostly institutional investors).”
Zitron’s essay offers some more examples — and he’s not the only one concerned here, according to The Washington Post. NVIDIA also announced an upcoming $100 billion investment in OpenAI — “one of its most prominent customers,” according to The Post, which adds that the move “fueled concerns about the sustainability of AI infrastructure spending.
“Some analysts warned clients that the potentially circular arrangement could contribute to any bubble in AI investment.”
Costs vs. Consistency
Zitron also doesn’t believe all of these massive investments will even happen — or that OpenAI will achieve its projected future revenue in the hundreds of billions of dollars.
But beyond that, his larger argument is that AI itself just isn’t profitable (outside of the major players) because its lack of demand is compounded by runaway costs.
Microsoft, Zitron estimates, currently makes a maximum of $2.88 billion a year on Copilot subscriptions (though the company is offering discounts) — based on very low, single-digit adoption rates among all Microsoft 365 subscribers. For a company that makes $27.2 billion in a single quarter, Zitron calls $2.88 billion a year “pathetic.”
But his essay also warns of the possibility that “paid users can also cost you more than they bring in,” because “it’s impossible to do cost control in a consistent manner.”
At one point, the essay even describes the “viberank” leaderboard where “people compete to see how much they burn, with the current leader burning $51,291 over the course of a month … That’s a bad business with out-of-control costs.”
To try to prove his point, Zitron’s essay cites an article in The Wall Street Journal that claimed Microsoft loses “on average more than $20-a-month per user,” with some users costing “the company as much as $80.” (But what Zitron doesn’t say is that the WSJ article was published two years ago.)
Hecht, the TNS analyst, points out that “the cost to service a customer is dropping every month, and the companies offering models as a service via API have strong control over what is very elastic pricing.”
So what about Zitron’s claim that controlling costs consistently is impossible? Hecht thinks Zitron is just dead wrong.
“Yes, there are limitations, but they are easily overcome,” Hecht wrote, adding, “Even at the individual customer level that we see, companies like Perplexity have created a tremendous number of tiers so they can charge more based on either 1) demand and 2) cost to provide a service … The big model providers will be able to adjust their investment levels and pricing over the medium term.”
And in the meantime, it’s just “a business decision about how much should be spent to acquire market share and new customers,” he added. “I’m sure that a majority of AI companies are not generating a profit. But, a majority of all [venture capital]-backed companies are not profitable in their first few years.”
Markets or Madness?
Will all these AI investments ever pay for themselves? On Sept. 25, The Wall Street Journal made the case for (and against) breaking even. “OpenAI counts roughly 700 million people — 9% of the world’s population — as weekly users of ChatGPT as of August, up from 500 million in March, while its revenue is on track to triple over 2024,” read the article by Eliot Brown and Robbie Whelan.
“If AI continues to advance to the point where it can replace a large swath of white-collar jobs, the savings will be more than enough to pay back the investment, backers argue. AI executives predict the technology could add 10% to global GDP in coming years. ‘Training AI models is a gigantic multitrillion-dollar market,’ Oracle chairman Larry Ellison told investors this month. The market for companies and consumers using AI daily ‘will be much, much larger.'”
But what if it doesn’t? MIT recently released a discouraging study where 95% of corporate respondents said their AI projects weren’t improving profits. (Though Hecht pointed out the study’s small sample size, described by the researchers as “52 structured interviews across enterprise stakeholders, systematic analysis of 300+ public AI initiatives and announcements, and surveys with 153 leaders.”)
Still, Zitron remains convinced. His essay makes his case, point-blank, that “the demand is not there for generative AI, and the demand is never, ever arriving.”
And maybe somewhere out there in the zeitgeist, there’s one more person smiling silently in agreement.
“I haven’t heard anything from any hyperscalers or neoclouds,” Zitron tells me, “but I do have sources there that love my work.”
