The Man Who Calls BS On AI: They’re LYING About AI, 2027 Is When It All Breaks! | Ed Zitron

Why does it matter? Because the AI industry's biggest revenues are a closed loop — and when it breaks, ordinary people's retirements go with it

Ed Zitron has spent 16 years in tech. He doesn't think generative AI is overhyped. He calls it a con — and he's traced the money in enough detail to make the charge stick. The revenues are circular. The adoption is manufactured. The collapse has a specific year attached.

• 70% of AI revenues from Microsoft, Google, and Amazon flow to OpenAI and Anthropic — the same unprofitable companies those giants are already funding • A $200/month ChatGPT subscription lets a single user burn $14,000 in compute; OpenAI lost $20.9 billion last year absorbing that gap • "Get on the AI train now" messaging is structurally identical to the opening move of every high-pressure con in history • Zitron's prediction: OpenAI runs out of cash around 2027, triggering a 20–40% crash in tech stocks that hits retirement portfolios and, in his view, doesn't recover

AI's headline revenues are accounting theater — the same three companies funding OpenAI are also its biggest customers

Seventy percent of all AI revenues across Microsoft, Google, and Amazon flow to OpenAI and Anthropic — two companies, as Zitron puts it, that "literally cannot afford to exist without these very same companies giving them money."

The numbers are specific. Amazon sent $50 billion to OpenAI and $5 billion to Anthropic in a single year. Google sent $10 billion to Anthropic. Microsoft, in fiscal year 2026, made $34.33 billion in AI revenue — $24.1 billion of which came from OpenAI alone. Strip that out and they netted roughly $10 billion from AI in a year when they spent $115 billion on capital expenditures, with plans to spend $175 billion the following year.

None of these companies fully disclose their AI revenues. When they do surface a figure, they reach for "annualized run rate" — a number they never consistently define. "It can mean month times 12, it can mean last four weeks times 13," Zitron says. "They never define it." When a public company doesn't announce something, the silence is information.

The sell-side analysts pricing these stocks expect OpenAI and Anthropic to generate $400 billion or more in cloud revenue over the next three and a half years. That money has to come from somewhere. Zitron's argument: it will come from the same three companies currently keeping both firms alive — making the supposed demand a closed accounting loop, not evidence of a real market.

A $200/month ChatGPT plan lets users burn $14,000 in compute — the moment enterprises faced real costs, the AI revolution evaporated

Here's how the pricing actually works. Semi Analysis found that on a $200/month ChatGPT subscription, a single user can burn $14,000 worth of tokens. On Anthropic's equivalent tier: $8,000 for $200. Even the $20/month plan lets users consume $400 in actual compute. OpenAI lost $20.9 billion last year absorbing that spread.

The subsidy wasn't abstract. When OpenAI tried to move enterprises — companies over 150 people — onto token-based pricing around March 2026, something revealing happened. Uber burned through its entire annual token budget in three months. Altman acknowledged that people "have a big problem with it." Uber's COO said it was getting hard to connect token spending to useful outcomes.

Zitron's summary: "The moment people actually had to pay for it, they go... I don't know actually." The enthusiasm didn't slow. It disappeared.

His argument is precise: if the service were worth its cost, companies would charge the cost. Selling $14,000 worth of compute for $200 isn't a customer acquisition strategy — it's an admission that real willingness to pay doesn't match the valuation these companies are built on. Demand manufactured through subsidy is not demand. It's a loan that accumulates interest until someone calls it in.

Big tech ran out of genuine growth ideas and bought GPUs instead — Zitron calls it the Rotcom Bubble, and the market rewarded every purchase

This is the structural argument underneath everything else Zitron says. After COVID, hyper-growth stalled. Google's ad business matured. Amazon hit retail margin ceilings. Meta ran out of obvious territory. "They don't have any hyper-growth ideas anymore," Zitron says. "So suddenly they started buying GPUs. And when they bought GPUs, people went — they're doing AI. We better buy the stock."

Several hundred percent gains across the Magnificent 7 followed. Not from disclosed AI revenues — those were negligible when they appeared at all — but from the market's willingness to give credit in advance. "Meta's revenues growing because of AI, Microsoft's revenue growing because of AI." When in reality, Zitron argues, it was price increases, ad auction changes, Amazon's emergence as an advertising business. Nothing to do with LLMs.

He borrows a line from Ed Elson at ProfitG Markets: "They are all doing Botox right now. They're sinking money into it to make themselves feel young again, and the market believes them."

The GPU purchases served two functions simultaneously: they signaled innovation to markets that rewarded that signal with stock price gains, and they created a narrative of inevitability that made further investment seem rational. Nvidia moved $215.9 billion in GPUs last year — to support approximately $22 billion in real-world AI revenue outside the two companies already on life support. When the market eventually prices these companies on actual business fundamentals, the Botox analogy completes itself.

Zitron's prediction: OpenAI runs out of cash around 2027, crashing tech stocks 20–40% and wiping out retirement savings that won't recover

OpenAI raised $122 billion in 2026. They plan to spend $750 billion on compute through 2030. Their IPO — meant to happen this year — has been pushed to at least 2027, with their CFO offering the reassuring non-answer that it will happen "earlier than 2027 or 2027." Amazon's $35 billion investment was reportedly contingent on them going public early.

Zitron's sequence is specific. OpenAI struggles to raise at its current valuation. It can't go public because Anthropic — growing faster, run more tightly — gets there first, making OpenAI's numbers impossible to defend by comparison. Once OpenAI can't go public, SoftBank — which holds roughly $100 billion of OpenAI stock on paper — can't liquidate. Amazon, Google, and Microsoft restate guidance. The market reprices.

"People's retirements are going to contract severely, and I don't believe they're going to return to those values." So much of the S&P 500's value is concentrated in these four companies and the rest of the Magnificent 7. A 20–40% correction in those stocks doesn't only hurt direct investors — it hits every index fund, every passive retirement portfolio, every pension with market exposure.

Oracle, he adds, may not survive at all. Its growth thesis depends entirely on OpenAI spending $300 billion over five years. Without that customer, a decade and a half of flat inflation-adjusted revenues leaves nothing standing.

Loading Google or opening Word already forces AI on you — calling that adoption is the industry's most brazen statistical trick

The adoption statistics used to justify trillion-dollar investments are counting people who encountered AI because it was injected into products they were already using.

"When you load Google, Gemini screams in your ear. When you load Word, Copilot's bugging you. When you use Amazon, Rufus has opinions on what socks you're buying." Zitron calls it "the largest non-consensual push of technology in history."

Fear compounds it. Three years of media coverage warning workers that AI will take their jobs unless they adopt immediately isn't neutral information — it's coercion. He describes workers "AI-washing" their output: claiming AI helped produce work whether or not it did, because managers who don't do the work themselves have become enforcers of AI performance. "There are people having to AI-wash their jobs by saying AI did it — otherwise their bosses who don't do anything will get mad at them. This did not happen with the internet."

The internet required physical infrastructure to reach people. Generative AI required a browser and a media ecosystem built for professional anxiety. Those are very different adoption mechanisms, and they produce very different signals about genuine demand.

'Get on the AI train now' is structurally identical to every con ever run — Zitron says that's not a coincidence

Altman went on record in early 2023 saying he was "a little bit scared about what we're creating." Zitron's read: "You know why he wants to say that? So you'll invest in his company. So you'll be scared that if you don't use AI today, you'll be left behind."

Then the frame that cuts cleanest: "Every single scam and con starts with rushing you. Every single trick in history begins with saying you must do this now." His personal rule — if anyone tries to rush you and it's not a mortal emergency, slow down.

What Zitron describes isn't incidental hype. The fear narrative served multiple functions simultaneously: it scared customers into adoption, scared regulators into deference, and positioned the companies building these systems as the only ones capable of managing them. The mysticism was deliberate.

It started backfiring. Eric Schmidt was booed at a commencement ceremony every time he said "AI." Doom quietly became "intelligence for everyone" in the marketing copy. The product didn't change. The pressure tactic did — which is itself a tell about what the pressure tactic was always for.

Software quality is measurably declining as AI-generated code floods production — Amazon Web Services went down multiple times this year because of it

The productivity case for AI coding rests on one metric: developers ship faster. What gets measured less is whether that code holds.

"The quality of software is going down weirdly enough as more people use LLMs," Zitron says. AWS went down multiple times this year because of AI coding tools. GitHub is now unreliable enough that someone posted on Twitter that a notification when it's up would be more useful than one when it's down. Google Docs is "a bugfest."

The mechanism isn't hard to trace. GitHub is flooded with AI-generated code from developers who "barely understand what they're shipping" — or who understand some of it, get overconfident, and push without review. Dunning-Krueger at industrial scale. Meanwhile, engineers who've let AI handle the work for months find themselves rusty with the fundamentals they'd otherwise have caught.

Steven's research confirmed it: industry data points to AI-assisted coding as a primary driver of increased system instability. Shipping velocity without reliability isn't productivity. It's deferred cost.

Commodity tools produce commodity outputs — human taste is now the only asset that actually compounds

"In a world where everybody has access to these tools, whatever the tools can do would largely be commoditized." Zitron and Steven land here together, and it's the one moment the conversation turns generative rather than diagnostic.

The illustration is immediate: LinkedIn posts written with ChatGPT read like LinkedIn posts because everyone is using it. A great LinkedIn post now is someone who doesn't use it — "irreplaceably human, deeper, more personal, N of one lived experience."

His editor Matt Hughes is the longer proof. "I don't pay Matt Hughes because he knows everything. I pay him because he has incredible context and a ton of knowledge and he's willing to expand it — and wonderful loving empathy in his heart for the stuff he loves and absolute venom for the people he hates. I can't get that from a large language model."

The scarcity logic is simple: when a capability becomes universally available, its outputs become universally ordinary. The value migrates to what the tools cannot replicate — context earned through genuine care, judgment that lives outside any training corpus, the thing that makes one perspective worth reading instead of any other. Fluency with the tool is now table stakes. The moat is everything the tool can't synthesize.

The real casualty of the AI bubble won't be the companies — it will be whoever trusted the promises

What this conversation surfaces is a structural problem that runs deeper than AI. We built financial markets that reward the narrative over the fundamentals, media ecosystems that amplify urgency over scrutiny, and employment cultures that punish workers who don't perform enthusiasm for whatever the C-suite just got excited about. The AI boom didn't create those conditions. It found them ready-made and exploited them at scale. When the valuation correction arrives — and Zitron believes 2027 is when it starts — the damage will land on people who had no say in any of it.

That's the part nobody's accounting for.

---

Topics: artificial intelligence, generative AI, tech industry, AI bubble, OpenAI, Anthropic, venture capital, tech stocks, AI hype, economic risk, software quality, Ed Zitron