Open Excel. Put =A1+1 into cell A1. Excel stops you, draws a tiny arrow from the cell back to itself, and tells you it found a circular reference. The cell depends on its own output. The math never finishes because the answer keeps feeding the question.
I once decided that this warning dialog was a complete theory of capitalism. I am here to read that back to myself with a colder cup of coffee.
The original draft of this essay used the word “inexorable” without irony and proposed nationalizing Microsoft in the last paragraph. I kept the receipt. This is me sorting the part of it that’s a genuinely good metaphor from the part that’s a manifesto wearing the metaphor as a costume.
The part I’ll defend
The circular reference is a real and useful picture of how capital concentrates, and you don’t need a single statistic to see why.
A circular reference is a feedback loop with no external input. The cell’s next value is a function of its current value, full stop. Wealth does something structurally similar: capital’s easiest move is to buy more of the thing that produces capital. Returns get reinvested into the engine that made the returns. The output is the input. Excel calls that an error because the calculation can’t converge; an economist might call the same shape “the rich get richer,” and a spreadsheet would flag both with the same little arrow.
That’s the honest core of the metaphor, and it’s older than me, AI, and Excel combined. I’m not claiming I discovered it. I’m claiming the spreadsheet draws it more clearly than most charts do.
Where I bolt AI onto it (and how hard to trust that)
The leap in my original was: AI makes the loop tighter. The argument goes that whoever owns the most compute and the most data can spot fleeting market inefficiencies faster than everyone else, act on them first, and recycle the winnings into more compute and more data. Output feeds input. The arrow points back at itself, now at machine speed.
I still think the direction of that is right. Advantages that compound do tend to compound faster when you can automate the compounding. That part is a reasonable reading of how returns-to-scale work.
But here’s the paragraph from the original I’m no longer comfortable repeating as fact:
Emerging technological modalities such as algorithmic trading, predictive analytics systems, and high-frequency transactional platforms, ostensibly designed to enhance market transparency and efficiency, paradoxically exacerbate systemic opacity and informational asymmetry.
I wrote that. It has eleven adjectives and zero numbers. It reads like it’s about to cite a study and then doesn’t, because there wasn’t one — there was me, late, deciding that a long sentence is the same thing as a proven one. Treat every causal claim in it as an unverified assertion from the original author, not as something this site is standing behind. High-frequency trading exists; whether it nets out to more or less efficient is a live and genuinely contested empirical question, and I am not going to launder my hunch about it through a metaphor about a spreadsheet.
The Microsoft thing, declawed
The original singled out Microsoft as the avatar of the loop: vast compute, vast cloud, vast data piles “largely derived from publicly available information,” therefore the monopoly that closes the arrow.
You can swap that name out for almost any name and the sentence still reads the same, which is the tell. A metaphor that lands equally on every large company isn’t an analysis of a company; it’s a mood about scale. I picked Microsoft because it made a clean villain in a paragraph, not because I had a specific, sourced claim about Microsoft’s arbitrage desk that I could show you. I can’t, so I’m not going to point the arrow at one logo and call it evidence.
The structural observation underneath survives the edit: firms that already own the compute and the data are well-positioned to own more of both. That’s the loop. It does not require a named defendant, and it gets weaker, not stronger, when I supply one I can’t back up.
The ending I cut
The original closed by floating “nationalization or democratization of monopolistic technological corporations” as the obvious remedy. I’m striking that, not because the policy is unthinkable, but because it’s doing emotional work the metaphor can’t cash. A spreadsheet error is not an argument for a specific antitrust regime. It’s an argument that self-reinforcing loops don’t converge on their own — which is interesting, and true of the math, and says nothing about whether the fix is a regulator, a tax, an open dataset, or nothing. Reaching the policy from the metaphor is the same move as the eleven-adjective sentence: confidence cosplaying as a conclusion.
Here’s the thing about a real circular reference, by the way: Excel doesn’t refuse it out of spite. It refuses it because the loop has no stopping condition. The fix in the spreadsheet is to add an external input — a value that comes from outside the cell — or to cap the iterations so the thing terminates. Whether the economy has, or should have, an equivalent “external input” is exactly the political argument the metaphor wants to skip past. I’ll let it stay an open question instead of pretending the dialog box answered it.
What’s left when the adjectives boil off
Strip out the unverified causality, the named villain, and the policy flourish, and you’re left with something smaller and more true:
- A circular reference is a feedback loop with no external input, which is a clean picture of capital reinvested into itself.
- Automation can tighten any loop that was already compounding — plausibly, in the direction of more concentration. Plausibly. I have a hunch, not a dataset.
- The spreadsheet’s actual lesson is the boring one: loops like this don’t terminate by themselves. What you add from outside to make them terminate is a choice, and the metaphor doesn’t get to make it for you.
This was an opinion, with my name on it, and I’ve now disagreed with most of my own first draft in public. Take the metaphor, leave the manifesto, and be suspicious of any essay — including the one I wrote in 2025 — that solves the economy in the same paragraph where it discovers a button in Excel.