Zitron Says Everyone Was Sold a Lie on AI — and the Cloud Numbers Prove Him Right

Zitron Says Everyone Was Sold a Lie on AI — and the Cloud Numbers Prove Him Right

Let me tell you something that's been bouncing around my head since I watched Ed Zitron's Bloomberg segment go viral yesterday. The man sat there on live television and said the entire AI boom is built on a lie — and then he dropped numbers that should make every founder and every CFO running infrastructure sit up and pay attention. I've been running hosting infrastructure for over a decade. I've seen vendor lock-in, I've seen hype cycles, I've seen sales teams sell vapor. But this? This is the market discovering that the growth it's been celebrating might just be a circle.

Here's the video that started it: Bloomberg Television, "Zitron: Everyone Has Been Sold a Lie on AI." Forty-one thousand views in under a day. Clipped, shared, argued about on every tech feed I follow. And the reason it's hitting so hard is that Zitron didn't show up with opinions. He showed up with receipts.

The Claim That Broke the Internet — 69% of Microsoft's Growth Is One Customer

Let me give you the number that stopped me cold. According to Zitron's reporting on OpenAI's spend, 69% of Microsoft's year-over-year growth in its Intelligent Cloud segment in 2025 came from OpenAI. Not from the enterprise, not from startups, not from the thousands of businesses migrating workloads to Azure. From one customer. One unprofitable company that Microsoft itself has poured tens of billions into.

Strip OpenAI out of the equation, and Microsoft's Intelligent Cloud growth drops to roughly 8%. Eight percent, folks. That's barely above inflation. That's not a growth business — that's a business holding its breath while one tenant pays the electric bill.

And it gets worse. Barclays is estimating that OpenAI and Anthropic together will represent about 13% of Microsoft's total revenue this year, rising to 18% next year. Two private companies that are burning cash at a historic rate are on track to become a bigger chunk of Microsoft's top line than Google Cloud is of Alphabet's. Two labs, renting capacity from the same company that funds them, driving the growth narrative of the largest software company on earth.

The UBS Numbers Nobody Wants to Talk About — 27% to 48% in One Year

Now here's where it gets really uncomfortable. UBS is estimating that OpenAI and Anthropic will account for 27% of Google Cloud's revenue this year. Next year? Over 48%. Nearly half of Google Cloud's revenue — a business that just reported 82% year-over-year growth — could come from two companies that have never posted a sustainable profit.

I've been in this industry long enough to know what concentration risk looks like. A colo provider with one tenant doing 30% of revenue is a red flag. Two tenants doing 48%? That's not a portfolio, that's a dependency. And when the whole cloud market is running on that same dependency, you're not looking at a growth story anymore. You're looking at a house of cards with really expensive cards.

How the Loop Actually Works — Vendor Financing's Evil Twin

Let me break down the mechanics, because this matters for every independent hosting provider trying to figure out where the real demand is. The loop works like this: Microsoft invests in OpenAI. OpenAI spends that money on Azure compute. Microsoft books it as cloud revenue and backlog. The stock goes up. Microsoft raises more money, invests more in OpenAI, OpenAI buys more compute. Around and around it goes.

I wrote last week about Nvidia financing its own chips to build a $500 billion data center — that's the supply-side version of this loop. This is the demand-side version, and it's arguably worse, because it's happening inside the revenue numbers themselves. When a hyperscaler reports "explosive AI cloud growth," a chunk of that growth is just its own investment coming back through the door wearing a different hat. And the scale of the bet is staggering: planned data center capacity is around 190 gigawatts, while the two labs driving most AI compute demand only need somewhere between 8 and 16. Building out the excess would require $1.6 trillion to $2.9 trillion in annual compute demand — two to three times the entire revenue of the software industry. The paying customer for that capacity doesn't exist yet.

Now, before you call me a doom-sayer, let me be clear about what I'm NOT saying. OpenAI's ARR is real. Anthropic's momentum is real — the company's run rate reportedly crossed $70 billion, up from $47 billion in May. Enterprise adoption is real; over 40% of OpenAI's mix is enterprise. What I'm saying is that the GROWTH NUMBERS the market is celebrating are inflated by a circular flow that can reverse as fast as it built up.

The Counter-Argument — Real Demand or Circular Fantasy?

Every bubble has its defenders, and this one's no different. The bulls will tell you: "Azure is growing 43%, AWS is growing 37%, Google Cloud is growing 82% — you can't fake that." And they're right that the top-line numbers are real. But here's the question nobody wants to answer: how much of that growth is OpenAI and Anthropic renting capacity from the very companies that fund them?

And the labs themselves aren't exactly reassuring. OpenAI is projected to lose $14 billion-plus in 2026. That's not a typo. Fourteen billion dollars of losses while the cloud providers book its spending as growth. Anthropic has a better cost profile and claims adjusted profitability, but it's still burning capital at a scale that depends on the fundraising window staying wide open. If that window closes — if the credit markets keep tightening, if banks keep walking away from data center deals like they did in July — the loop reverses. And when a loop reverses, it reverses hard.

The Secondary Bottleneck — Demand Concentration Distorts Every Pricing Signal

Here's the part that keeps me up at night as someone who runs actual infrastructure: when two unprofitable companies drive the demand curve, every pricing signal in the market is distorted. Hardware pricing, colo rates, power contracts — all of them are being set by a demand signal that isn't broad-based. It's concentrated in two balance sheets that depend on continuous capital infusions.

That's the bottleneck nobody's talking about. It's not chips, it's not power, it's not water. It's demand concentration. The AI infrastructure buildout isn't pricing off ten thousand customers with real budgets. It's pricing off two tenants whose entire existence depends on the next funding round clearing. That's not a market. That's a margin call waiting to happen.

What This Actually Means for Independent Hosting Providers

First — stop reading hyperscaler cloud growth as a demand signal for your business. If Google Cloud's 82% is 27% one company and heading to 48%, then "the cloud is booming" doesn't mean what you think it means. Build your capacity plans off your actual pipeline, not off the headlines.

Second — watch lab funding rounds as your real leading indicator. OpenAI's next raise, Anthropic's next raise, the terms they get — that's your canary. If those rounds get harder, the compute demand they represent slows within two quarters. Plan accordingly.

Third — diversify your tenant mix now. If you're running a colo or hosting business with any single tenant doing more than 20% of revenue, you're already carrying concentration risk that the hyperscalers are only now being forced to acknowledge. Fix it before the market forces you to.

Fourth — don't sign long-term AI colo deals at today's premium rates. If you're being offered above-market rates for AI workloads with a long lock-in, remember what happened to every provider that locked in at the top of the dot-com cycle. Price for the correction, not the peak.

The Bottom Line

Zitron said everyone was sold a lie on AI. I don't think it's a lie about the technology — the technology is real, and it's improving fast. The lie is in the growth narrative. The lie is that the boom is broad-based and organic when the numbers say it's concentrated in two companies spending other people's money.

This doesn't mean AI infrastructure collapses tomorrow. It means the cycle is further along than the market wants to admit, and the smart money is already asking hard questions. Independent hosting providers have an advantage the hyperscalers don't: we can't hide our revenue concentration in a $2 trillion backlog. We see our customers every month. We know who's real.

So here's my advice, founder to founder: trust your own numbers, watch the lab funding rounds, and don't let a 190-gigawatt fantasy convince you to overbuild your own little corner of it. The truth isn't that AI is a lie. The truth is that the growth numbers are a circle — and circles, sooner or later, always come back around.

-- Allan Ali, Founder

This article was produced with AI-assisted research and editorial support. Reporting is based on sources cited in the article.