The Race Condition
Two AI Futures, One Portfolio
TL;DR: Two writers model opposite AI futures from identical starting conditions – one where the economy collapses, one where it booms. Both are partially right. Both are wrong about the same thing. The outcome isn’t one or the other – it’s a fracture. And 50% of my portfolio says one layer wins regardless.
I checked my portfolio at 2 AM on a Tuesday. Not because something had moved. Because I’d just finished reading two macro memos that model opposite futures from the same starting conditions, and half of my net worth was sitting in the gap between them.
Citrini Research wrote the bear case: unemployment at 10.2%, S&P down 38%, displaced white-collar workers driving Uber, Congress debating emergency transfers while protestors blockade Anthropic’s offices.
Michael Bloch wrote the bull case: unemployment at 3.1%, S&P at 12,000, purchasing power up 18%, a 22-year-old in rural Arkansas accessing the same financial guidance that used to cost $500 an hour.
Same technology. Same timeline. Opposite worlds.
Both are worth reading in full. But the gap between them is the only question that matters.
Two Stories About the Same Machine
Strip both pieces to their skeletons and the disagreement is surprisingly narrow.
They agree on the setup: agentic AI leaps in late 2025, software companies get repriced, white-collar job openings collapse, and middleman businesses get crushed.
Where they diverge is on what happens to the money.
The bear says it spirals. Companies cut workers, that money flows into more AI, which cuts more workers. The money pools at the top and never recirculates. “Ghost GDP” – an economy where the numbers look fine and the people aren’t. Five words: GPU clusters don’t buy groceries.
The bull says it spreads. Companies cut costs, that money gets redeployed – expansion, new hires, new markets. Purchasing power rises even if salaries don’t. And the historical prior is undefeated: technology-driven deflation has improved living standards every single time it has occurred at scale. Textiles. Steel. Automobiles. Computing. The internet.
Same dollars. Different direction.
The Race Condition
In software, a race condition is a bug where the outcome depends on timing, not inputs. Two processes compete for the same resource. Whichever arrives first determines the result.
That’s what we’re looking at.
Process A: Income destruction. AI replaces white-collar workers. Those workers spend less. Their employers face margin pressure and adopt more AI. The loop accelerates.
Process B: Cost deflation. AI makes services cheaper. Healthcare, legal work, tax prep, financial advice, insurance – the entire friction layer of the economy gets compressed. Purchasing power rises even as paychecks shrink.
Both processes are already running.
If cost deflation outpaces income destruction, people get poorer on paper and richer in practice. Your salary drops 10%, but your cost of living drops 25%. The bull wins.
If income destruction outpaces cost deflation, people lose their jobs before the savings arrive. You can’t benefit from cheaper legal services when you can’t make rent. The bear wins.
This isn’t a theoretical distinction. It’s the whole game.
Where Both Are Wrong
The bear assumes the spiral runs unopposed. No adaptation, no cost deflation offsetting income destruction. That requires every displaced worker to sit politely in the unemployment line while the economy hollows out around them. People don't do that. People panic, adapt, and start questionable side hustles.
The bull assumes entrepreneurship outpaces displacement. Nine months of turbulence, then some guy’s laid-off uncle launches an AI startup from his kitchen table and out-earns his old salary by Christmas. Survivorship bias in a trench coat. For every one of him, there are dozens burning through savings on AI subscriptions for products nobody wants.
The bear underestimates human adaptability. The bull overestimates it.
The bear ignores deflation. The bull ignores leverage.
And both ignore friction.
Friction might be what saves us.
Both pieces assume frictionless adjustment. But the real economy is made of friction: regulatory capture, institutional inertia, contract law, procurement cycles, human stubbornness. Real estate brokers should have been extinct a decade ago – someone just paid one $50k for ten hours of form-filling, in 2025. Cockroach industries don't read McKinsey reports.
Hofstadter’s Law applies in both directions: everything takes longer than you expect, even when you account for it. That slows the displacement and the recovery. But crucially, it buys the system time to respond.
Neither piece touches the physical economy. Both model a digital displacement hitting a services-based consumer economy. But the US has a bipartisan need for re-industrialization – batteries, semiconductors, desalination, grid infrastructure. China makes 90% of the world’s ammonia. If there’s a military confrontation, America can barely grow corn.
AI can’t build a bridge. Not yet. The physical world moves at the speed of permits, concrete, and labor – not the speed of inference.
Neither piece addresses the measurement problem. The bear models income. The bull models purchasing power.
Neither measurement is complete – but both operate within a Keynesian framework where GDP is demand, demand is spending, and spending requires wages.
An economy producing 10x more at 10x lower cost isn’t “shrinking” in any sense that matters to the person buying groceries. It’s the measuring stick that breaks, not the economy.
The Fracture
And neither piece considers that the race condition doesn’t have to produce one winner nationally. That’s the one I can’t stop thinking about at 2 AM.
San Francisco, where 40% of jobs are in tech and finance, might live the bear case while Nashville, with a diversified economy and lower costs of living, lives the bull case. Same layoff. Different outcome. The same laid-off product manager is driving Uber in San Francisco by March and consulting for three healthcare startups in Nashville by April.
The national statistics would show something in between. Not a crisis, not a boom, but an economy where your zip code determines which future you got.
That’s not Ghost GDP and it’s not an abundance boom. It’s a fracture.
Economist Alex Imas modeled what it would actually take for the bear case to crash the whole economy: almost everyone would need to lose their job, nobody would want to buy anything new even if it were cheap, and the government would have to do nothing. All at once.
Implausible.
Relax those assumptions even slightly and you still get meaningful drag. Not collapse. Not boom. Disappointing growth, unevenly distributed, with policy scrambling to catch up.
That sounds about right.
What This Means for Money
(Disclaimer: my portfolio is 50% the AI infra layer. I am not a neutral observer.)
In the bear scenario, compute demand accelerates because companies are desperately substituting labor for machines.
In the bull scenario, compute demand accelerates because the cost of intelligence drops and usage explodes.
In the fractured middle, compute demand accelerates because both dynamics run simultaneously in different parts of the economy.
Compute demand is the common variable. The infrastructure layer wins regardless of what happens above it. This is the AI Landlords thesis – own the energy and compute, let everyone else fight about the rest.
But the thesis being right doesn’t make the ride comfortable.
If the bear plays out even partially, the stock market crash hits everything on the way down – including infrastructure. Everything sells off before the market figures out what’s structural and what isn’t. That means sitting through a 30–40% drawdown with 50% concentration and zero cash, waiting for the market to do the math. The thesis becomes a pain tolerance test.
I believe cost deflation runs faster than the bears think and income destruction runs slower than the doomers model. I believe infrastructure is the right place to be while the race plays out.
But I’ve set limit orders to scale out on the way up. Because believing is not the same as being certain.
Two memos. Two futures. One machine.
My portfolio says the infrastructure wins either way. My limit orders say I might be wrong about the timing. And the fact that I’m writing this at 2 AM says I haven’t fully resolved the tension.
Nobody has.