In partnership with

Introducing the first AI-native CRM

Connect your email, and you’ll instantly get a CRM with enriched customer insights and a platform that grows with your business.

With AI at the core, Attio lets you:

• Prospect and route leads with research agents

• Get real-time insights during customer calls

• Build powerful automations for your complex workflows

Join industry leaders like Granola, Taskrabbit, Flatfile and more.

👉Try Attio Pro for free

Hey, josh here. I was thinking of the developments from this past week and came to a realization. 7 Businesses and recycling money propping up the U.S economy. check out my notes on it below.

The AI Industry Built Itself a $1 Trillion Perpetual Motion Machine

Listen, I need you to understand what's happening in AI right now, because it's fucking wild.

OpenAI needs chips to run its data centers. So Nvidia—the company that makes those chips—invests $100 billion in OpenAI. OpenAI then uses that money to... buy Nvidia chips. Nvidia books this as revenue, its stock goes up, and now it can invest even more. Meanwhile, Oracle buys $40 billion worth of those same Nvidia chips, leases them to OpenAI for $300 billion, and OpenAI pays Oracle with money that Oracle helped fund. AMD gets in on the action by giving OpenAI warrants to buy 160 million AMD shares while OpenAI commits to buying AMD's chips.

This is what Bloomberg is calling the "$1 trillion loop." The same money is just circulating through the same five companies—OpenAI, Nvidia, Oracle, AMD, Microsoft—with each pass being recorded as fresh revenue, new investment, critical infrastructure spending. It's like playing hot potato with a billion dollars, except every time you catch it, your company's valuation jumps 20%.

Dylan Patel from Semianalysis has the best name for this: the "infinite money glitch."

Here's the kicker

OpenAI is burning roughly $575,000 per hour. In Q1 2025 alone, the company lost $11.5 billion while generating maybe $13 billion in revenue for the entire year. Microsoft ate $3.1 billion of those losses just from its 27% stake. Oracle's making 16% gross margins on GPU rentals—compared to 70% on its legacy cloud business—which means the fundamental economics are, let's be generous, "emerging."

And the math? Harris Kupperman from Praetorian Capital did the brutal work here. To break even on 2025's capital expenditures, the AI infrastructure industry needs $160-480 billion in annual revenue at 25% margins. Current AI revenue sits around $20 billion. That's not a gap. That's a fucking chasm. Revenue would need to grow 16-24x while maintaining those margins—something no infrastructure-heavy industry has ever achieved.

The thing is, this whole buildout was justified by "scaling laws"—the idea that more compute and more data automatically equals better AI. Except those laws are hitting a wall. OpenAI's Sam Altman admitted in August they're "still missing something quite important" to reach AGI. MIT research shows pretraining gains are plateauing. The original business case is crumbling while companies are still pouring in hundreds of billions.

Why this matters

Remember the dot-com bubble? Telecoms laid 80 million miles of fiber optic cable in the 90s. Between 85-95% of it stayed dark—completely unused. Corning dropped from nearly $100 to $1. Ciena's revenue collapsed from $1.6 billion to $300 million overnight.

We're watching the same movie, just with data centers instead of fiber. Meta's planning a facility "so large it could cover a significant part of Manhattan." The Stargate Project alone is $500 billion over four years. Meanwhile, MIT found that 95% of AI pilot projects fail to yield meaningful results.

But here's where it gets interesting: unlike the dot-com era, actual revenue is growing. Microsoft's Azure AI hit an $86 billion run rate. OpenAI went from $6 billion to a projected $20 billion in annual revenue in one year. Real customers exist. The infrastructure being built will support actual workloads, not sit dark.

So what is this—a bubble or essential infrastructure?

Both, probably. The circular deal structure is absolutely inflating valuations beyond what fundamentals support. The profitability math genuinely doesn't work yet. The core technological assumptions are being revised in real-time. But the underlying technology will transform productivity, eventually, somehow.

Kupperman's take hits right: "This isn't a bubble waiting to burst. It's one deflating in slow motion."

The correction is already happening—smaller VC-backed startups are getting crushed while Microsoft and Google keep building because they have balance sheets that can absorb years of losses. The next 18-24 months will reveal whether this was visionary coordination around transformative technology or an expensive lesson in mistaking circular cash flows for actual demand.

Either way, we're watching a trillion dollars find out in real-time