A House of Cards Isn't Fake

Here's a business you can start this afternoon. You wire me ten thousand dollars. I wire you ten thousand right back. We each book ten thousand in revenue. The books balance. The cash never actually moved. Do it in a loop with enough friends and you've got a run rate that would make a Series B investor tear up.

That's the pitch on a site called larp.website, and it's a joke. Satire, right down to a global volume counter that reads "$0 of it real." Its best line, though, isn't really joking: "Revenue is just an agreement between friends."

I laughed. Then I stopped, because I've spent the last year reading term sheets, and the real version of this runs right now, at hundreds of billions of dollars, structured by a bank so it's legal.

The money already moves in a circle.

Follow one dollar through the AI economy. Nvidia invests billions into OpenAI. OpenAI commits that money to cloud providers, including a three hundred billion dollar contract with Oracle. Oracle spends it on Nvidia chips to build the data centers those contracts require. The dollar leaves Nvidia's balance sheet labeled "investment" and lands back on its income statement labeled "revenue," having done a full lap.

It isn't just that triangle. OpenAI paid Microsoft north of seventeen billion dollars last year; Microsoft paid a few hundred million back. Microsoft is OpenAI's supplier, its landlord, and a major shareholder at once. Total up the announced deals and you're past eight hundred billion dollars of circular capital.

None of this is fraud, and the satire site is careful to say so. A circular deal is legally different from a round-trip, the regulator's term for sham trades with no economic substance. The difference, larp.website says, is "mostly vibes, scale, and whether a bank structured it." Real chips get bought, real data centers get poured, the substance is there. What's also there is the same dollar booked as revenue two or three times on its way around the ring.

A house of cards is not a pile of fakes.

People hear "house of cards" and picture forgeries. Wrong picture. Every card in this thing is a real company shipping a real product to real customers. The fragility has nothing to do with fraud. It's that every card leans on the card beside it, and not one of them stands on the floor.

We have a name for this in software. A circular dependency. Module A imports B, B imports C, C imports A. Every file is real code. It compiles. It runs, most days. But try to load any one module by itself and the whole graph refuses to resolve, because nothing in the cycle can stand up until everything else already has.

The nastier version, the one that's cost me a weekend, is the test that mocks the thing it's supposed to test. You stub the dependency so it hands back exactly what your assertion checks for. The test goes green and stays green forever, and never catches a bug, because it was never measuring the system. It was measuring itself. You learn, usually at 2am, to distrust any setup where the measured and the measurer are the same node.

Sit with that, because the AI economy runs two of these loops at once. One is made of money. The other is made of story.

The story moves in a circle too.

The companies whose valuations depend on you believing AI is about to replace software engineers are, to an uncomfortable degree, the same ones writing that belief. Take Anthropic. It has raised a reported one hundred thirty-two billion dollars and is walking toward an IPO valued north of a trillion. It can't show a profit yet, so what it's really selling is its hypothetical future impact. And the loudest, steadiest voice insisting that coding is vanishing, then all of software engineering, then most human labor, belongs to that same lab.

Ray Myers, once chief architect at a coding-agent company, said it cleaner than I can. "In literary terms," he wrote, "Anthropic is an unreliable narrator." The company selling you the future is also the sole author of the benchmark that says the future is already here.

Want the miniature version? Anthropic owns Bun, a JavaScript runtime and one of the largest codebases ever written in Zig. Zig bans AI-generated contributions, an awkward rule when your parent company sells AI. So the Bun team pointed Claude at it and rewrote nine hundred sixty thousand lines of Zig into Rust in about six days, and the result shipped as proof that AI had made human engineers optional. One layer down: that rewrite carried more than thirteen thousand blocks of unsafe Rust, against seventy-three in a comparable codebase a person wrote by hand. Bun's own creator admitted there was a very high chance the whole thing gets thrown out. It merged to main anyway, because the code was never the deliverable. The press release was.

Andrew Kelley, who created Zig, looked at the performance and said the plain thing about what he saw. Inside a circle where everyone is paid to keep the narrative aloft, the rarest move going is one person willing to name what's on the table.

Sign up for a free trial of Attio, the agentic CRM.

Ask Attio to build a daily workflow that surfaces the deals that need your attention today, like anything with a stage change, a recent reply, or a new signal in the last 24 hours.

Review your pipeline in Claude, synced live from Attio via MCP.

That's it.

Try Attio Now

Venture was supposed to be the node outside the loop.

The bulls have a real point here, and the sharp ones make it fast. Vendor financing is ancient and not automatically a con. The railroads ran on circular, over-promised, half-fraudulent capital, then carried the economy for a century. The question is never whether there's a loop. It's whether value moves through it, or just the same dollar changing hats each lap.

Drawing that line is the job. Venture capital is supposed to be the node outside the cycle, the auditor who won't grade the vendor on the vendor's own benchmark. When an engineer sizes up a system, the first move is to isolate a component and check that it stands alone. In my new world that has a name: diligence.

Here's what I've had to sit with. When my limited partners are chasing the same AI story, and my fund's marks depend on the next round clearing higher, and that next round is funded by the same circular capital, the auditor has quietly walked inside the ring. I'm not outside the loop. I'm a node in it.

So no, I won't tell you the whole thing folds next quarter. That's the boring question, and I'd be guessing. The cards are real, the customers are real, most of the substance in the pipes is real. It could hold for years, the way those over-built railroads ended up hauling freight nobody could forecast.

What I know is smaller and worse. I came into venture sure I was the skeptic, the one paid to pull a component out of the diagram and check whether it holds alone. And right now I'm holding a card. My returns need the story to stay up one more round, which makes me one more person who won't look too hard at the load-bearing wall. I told myself I was the node outside the loop. I've been standing inside it the whole time.

— SWEdonym