On Tuesday, a Substack essay by a writer named Owen McGrann titled “The Dead Economy Theory” hit the front page of Hacker News and started circulating through the kind of group chats that normally trade in seed-round gossip and regulatory despair. By Wednesday afternoon it had racked up hundreds of thousands of reads. The thesis, stripped to its chassis: AI firms need a global customer base the size of the entire labor force to justify the trillion-plus dollars now sunk into data centers and model training, yet the very product they are building is the systematic elimination of that labor force. You cannot sell $10,000-a-seat enterprise agents to people you just laid off.
The essay draws on Daron Acemoglu’s work on “excessive automation” and a concept Wharton economists have labeled the “AI Layoff Trap.” McGrann argues — persuasively, I think — that this is not a transition problem that markets will smooth over with time. It is a structural contradiction baked into the unit economics.
Predictably, the conversation that erupted around the essay split along well-worn lines. The left saw vindication for UBI, wealth taxes, and a renegotiated social contract. The right — my side, ostensibly — mostly scoffed. Lump-of-labor fallacy, Luddite panic, creative destruction has always worked before.
Both camps are missing the genuinely destabilizing question. It is not “what does this mean for workers?” It is “what does this mean for the people holding the stock?”
The Market Is Pricing a Customer Base That AI Is Designed to Eliminate
Nvidia trades at roughly 27 times forward earnings. Microsoft is north of 30. The hyperscalers — the companies actually building the infrastructure McGrann writes about — have collectively added something like $5 trillion in market capitalization since ChatGPT launched, much of it on the explicit promise that enterprise AI adoption will generate recurring revenue streams large enough to justify the buildout.
But revenue requires customers, and customers require income. The Wharton paper McGrann cites puts the mechanism starkly: firms adopt AI to cut labor costs, which reduces aggregate wages, which reduces aggregate demand, which reduces the revenue those same firms can capture. The AI Layoff Trap names something that a first-year macroeconomics student would recognize as a fallacy of composition — what is rational for the individual firm is self-defeating for the system — except the system here is not an abstraction. It is a balance sheet.
If McGrann is even half right, the terminal value assumptions in every AI company’s discounted cash flow model are wrong. Not slightly wrong. Wrong by an order of magnitude. The customers are being fired.
The Investor Class Has Adopted a Theory It Cannot Afford to Believe
This is where the irony gets sharp. The essay’s most enthusiastic amplifiers have been venture capitalists and technology commentators who share it with a kind of grimly knowing nod — as if to say, “yes, this is the dark truth we are all brave enough to acknowledge.” But the same people nodding along are the ones whose portfolios depend on the contradiction never resolving.
One trading-desk analyst at a mid-sized asset manager put it to me this way during market hours Tuesday: “Half the guys forwarding this thing own the Mag Seven. They’re sharing it the way medieval monks copied manuscripts about the apocalypse — fully convinced the world would end and fully planning to say Mass the next morning.”
The cognitive dissonance is not a bug. It is the thing holding the current multiples in place. The moment the investor class stops treating McGrann’s argument as a provocative essay and starts treating it as a discounted-cash-flow input, the repricing begins. And repricing a trillion-dollar narrative doesn’t happen gently.
Creative Destruction Requires Creation, Not Just Destruction
Here I part ways with the standard right-of-center reflex, which is to invoke Schumpeter and call it a day. Schumpeter’s creative destruction presumed that the new industries would employ the workers the old industries shed — that the automobile would hire more people than the buggy whip ever did. The AI case is different not in kind but in speed and scope. When the product is the labor, the replacement cycle is not a cycle. It is a one-way valve.
You do not need to be a doomsayer to notice this. You only need to read the earnings calls. Every major enterprise software company now leads with headcount-reduction use cases. Salesforce, Workday, Microsoft — the pitch is not “your employees will be more productive.” It is “you will need fewer of them.” The language has shifted from augmentation to substitution, and the shift happened fast enough that the revenue models have not caught up.
This is what McGrann gets right and his critics evade: if the technology works as advertised, the addressable market shrinks with every deployment. That is not a policy problem. It is a pricing problem. And the market is not pricing it.
What Gets Repriced First
If you want to know whether the market is taking the Dead Economy Theory seriously, do not watch the Substack share count. Watch the discount rates embedded in AI infrastructure REITs. Watch the implied volatility on Nvidia LEAPs. Watch whether anyone in a quarterly letter from a major fund begins to ask — even obliquely — whether the terminal value of enterprise AI depends on a consumer base that the enterprise AI business model is engineered to liquidate.
The uncomfortable truth for my side of the aisle is that McGrann’s argument is not a leftist policy brief in disguise. It is an unsparing piece of financial analysis, whether its author intended it that way or not. You can dismiss the normative claims about the social contract. You cannot dismiss the math. A product that eliminates its own customers is not a business — it is a liquidation event. And liquidations are priced accordingly, once the market notices.
That moment has not arrived yet. But the essay now has half a million readers, and at least some of them manage money.