Abundant Intelligence and Deficient Demand: A Macro-Financial Stress Test of Rapid AI Adoption

This paper formalizes a macro-financial stress test arguing that rapid AI adoption creates a distribution-and-contract mismatch where AI-driven abundance fails to generate sufficient demand because economic institutions remain anchored to human labor scarcity, triggering a self-reinforcing cycle of income displacement, declining monetary velocity, and intermediary collapse that poses disproportionate risks to private credit and mortgage markets.

The Gist

Imagine the economy as a giant, complex water park.

For decades, the water (money) has flowed through a specific set of pipes: Wages. When people work, they get paid, they spend that money, and the water keeps circulating, keeping the park's rides (businesses) running and the slides (economy) full of fun.

This paper, written by Xupeng Chen, asks a scary question: What happens if we suddenly replace the workers with robots (AI) that do the work for free?

The author argues that while this sounds like a "productivity miracle" (more water in the tank), it could actually cause the whole park to dry up. Here is the breakdown of the paper's three main mechanisms, using simple analogies.

1. The "Displacement Spiral" (The Domino Effect)

The Analogy: Imagine a row of dominoes.

• The Setup: A company decides to replace 100 human workers with AI because it's cheaper.
• The First Fall: Those 100 workers lose their paychecks. They stop spending money at the local coffee shop, the car dealership, and the movie theater.
• The Chain Reaction: The coffee shop owner sees sales drop. To survive, the coffee shop owner fires their barista or buys an AI coffee machine. Now, another person loses a job.
• The Spiral: This creates a vicious cycle. As more companies fire people to use AI, there is less money in the hands of regular people to buy things. Companies see fewer customers, so they fire more people to cut costs, which means even fewer customers.

The Catch: Usually, when technology kills old jobs, it creates new ones (like how the invention of the car killed horse-drawn carriages but created mechanics and gas stations). The paper calls this "Reinstatement."

• The Stress Test: The paper worries that AI is moving so fast that the "new jobs" can't be created quickly enough to catch the falling dominoes. If the new jobs don't appear fast enough, the spiral becomes explosive.

2. "Ghost GDP" (The Phantom Economy)

The Analogy: Imagine a video game where the score goes up, but the player's health bar goes down.

• The Scenario: AI is incredibly efficient. It produces a massive amount of goods and services. The official "GDP" (the score) looks amazing. It's rising!
• The Problem: Who owns the money from that production? In the old days, the money went to workers as wages. Now, the money goes to the owners of the AI (capital owners).
• The Mismatch: The rich owners of AI don't spend their extra money as fast as workers do. Workers spend almost every dollar they get; rich owners save or invest most of it.
• The Result: The "Ghost GDP" is the gap between the high production numbers and the low spending power of regular people. The economy looks healthy on paper (high GDP), but in reality, people have no money to buy the products, so the "real" economy crashes.

3. The "Intermediation Collapse" (The Middleman Meltdown)

The Analogy: Think of a toll booth on a highway.

• The Setup: For years, companies like consultants, financial advisors, and software sellers made money because it was hard to find information. They were the "toll booths" charging a fee to help you cross the bridge.
• The AI Effect: AI is like a super-fast tunnel that bypasses the toll booth. It can find the best insurance policy, write the legal contract, or solve the coding bug instantly and for free.
• The Crash: Suddenly, the "toll booths" (consultants, brokers, SaaS companies) have no one to charge. Their business models collapse because their value was based on "information friction" (confusion), and AI removes the confusion. This threatens trillions of dollars in private credit and mortgages held by these companies.

The "Rich Kid" Problem (Concentration Amplifier)

Here is the most dangerous part of the paper.

• The Fact: The top 20% of earners (the "Rich Kids") are responsible for nearly 60% of all spending in the US.
• The Risk: These are exactly the people most likely to be replaced by AI (doctors, lawyers, coders, managers).
• The Impact: If AI replaces the "Rich Kids," the economy doesn't just lose a little spending; it loses the engine of the entire economy. It's like removing the main fuel tank from a car; the car stops instantly, even if the back seats are full.

The Solution: It's Not Inevitable

The paper is a "Stress Test," not a prediction of doom. It says: If we do nothing, this could be a disaster. But it also offers a way out.

The Policy Toolkit:

1. Speed Bumps: Slow down the rollout of AI so the economy has time to adapt (create new jobs).
2. The Safety Net: If AI makes money for the rich, the government must tax that money and give it directly to the workers (transfers) so they can keep spending.
3. Shorter Work Weeks: Share the remaining work among more people so everyone keeps a paycheck.

The Bottom Line

The paper warns that we are building a Ferrari engine (AI) in a car that relies on human drivers (workers) for fuel. If we swap the drivers for robots without changing the fuel system, the car will stop.

However, if we change the fuel system (redistribute the AI profits) and slow down the engine (manage the rollout), we can enjoy the speed without crashing. The crisis isn't caused by the technology itself, but by our failure to update the rules of the game.

Technical Summary

Here is a detailed technical summary of the paper "Abundant Intelligence and Deficient Demand: A Macro-Financial Stress Test of Rapid AI Adoption" by Xupeng Chen (March 2026).

1. Problem Statement

The paper addresses a central paradox in the current AI era: while AI promises unprecedented productivity gains (potentially the largest since electrification), the primary channel through which productivity translates to aggregate demand—labor income—is the very channel AI displaces.

Unlike narratives focusing on an "AI bubble" (supply-side failure) or "existential risk" (loss of control), this paper identifies a distribution-and-contract mismatch. Existing economic institutions (labor contracts, mortgage underwriting, tax codes, and intermediary business models) are anchored to the assumption that human cognitive labor is scarce and stably compensated. The paper argues that if AI capability grows rapidly while institutional adaptation lags, this mismatch can trigger a macro-financial contraction characterized by:

• Demand deficiency: Despite high output, consumption collapses because income shifts from high-marginal-propensity-to-consume (MPC) labor to low-MPC capital.
• Financial fragility: A correlated shock to high-income, high-debt households and intermediary firms.

2. Methodology

The paper employs a formal macro-financial stress test combining task-based production theory with aggregate demand feedback loops.

• Theoretical Framework:

  • Task-Based Production: Adapts the Acemoglu & Autor (2011) framework where output is a CES aggregate of tasks performed by either humans or AI.
  • Dynamic Feedback: Introduces a "displacement spiral" where firm-level rational substitution of labor for AI reduces aggregate labor income, which lowers aggregate demand, creating margin pressure that forces further AI adoption.
  • Reinstatement Channel: Incorporates the competing mechanism of new task creation (Acemoglu & Restrepo, 2019) to determine if the system is self-limiting or explosive.
  • Monetary Velocity: Models the divergence between measured GDP and consumption-relevant income ("Ghost GDP") by linking monetary velocity to the labor share.
  • Intermediation Economics: Models service-sector margins as a function of information frictions, showing how AI agents compress these margins toward pure logistics costs.

• Empirical Strategy:

  • Macro Data: Uses FRED time series (1950–2025) to document secular trends in labor share decline and velocity collapse.
  • Micro Data: Conducts cross-sectional OLS regressions on BLS occupation-level data (2019–2023) to test for wage compression in AI-exposed sectors.
  • Calibration & Simulation: Calibrates the model using empirical moments (labor share, MPC, diffusion speeds) and runs simulations under three scenarios: Baseline, Rapid, and Extreme AI adoption.

3. Key Contributions

The paper makes four distinct contributions to the literature:

1. Formal Framework: It extends task-based models by adding aggregate demand feedback, logistic diffusion dynamics, monetary velocity effects, and intermediation economics. It derives explicit thresholds distinguishing a "reinstatement-dominated" (benign) regime from an "explosive" (crisis) regime.
2. Falsifiable Predictions: Rather than offering scenario narratives, the paper derives 11 testable hypotheses with explicit falsification conditions (e.g., specific wage growth patterns, velocity trends, and credit spreads).
3. Empirical Grounding: It provides early empirical evidence that AI-exposed occupations are already experiencing relative wage compression (2019–2023), even before mass layoffs, suggesting the displacement spiral may be operating through wage suppression rather than just headcount reduction.
4. Policy Stress Test: It quantifies the conditions under which the system transitions from stable adjustment to crisis, demonstrating that the depth of any crisis is a policy variable (dependent on transfer speed and magnitude) rather than a technological inevitability.

4. Key Results & Mechanisms

A. The Three Transmission Mechanisms

1. The Displacement Spiral:
   • Firms substitute AI for labor $\rightarrow$ Labor income falls $\rightarrow$ Aggregate demand falls $\rightarrow$ Margins compress $\rightarrow$ Firms accelerate AI adoption.
   • Result: If the AI capability growth rate ($g_A$) exceeds a threshold determined by the reinstatement rate ($\rho$), the feedback loop becomes explosive, driving the labor share toward zero.
2. Ghost GDP:
   • When AI generates output, the same nominal GDP circulates through fewer wage-earners. Since labor has a higher MPC than capital, monetary velocity declines monotonically as the labor share falls.
   • Result: Standard indicators (GDP growth, productivity) may remain positive while the consumption-relevant economy contracts, misleading policymakers.
3. Intermediation Collapse:
   • AI agents reduce information frictions (search, switching, complexity costs).
   • Result: Margins for intermediaries (SaaS, consulting, insurance, payments) compress toward pure logistics costs, threatening revenue models for firms valued on growth assumptions.

B. The Consumption Concentration Amplifier

• High-income earners (top quintile) drive 47–65% of U.S. consumption and are the most exposed to AI automation.
• Result: A 10% income shock to the top quintile translates to a 4.7–6.5% aggregate consumption decline, disproportionately larger than shocks to lower-income groups. This concentrates risk in private credit ($2.5T) and mortgages ($13T).

C. Simulation Outcomes

• Baseline ($g_A = 0.05$): Gradual labor share decline; reinstatement offsets displacement.
• Rapid ($g_A = 0.20$): Labor share falls 20–40% within a decade; consumption declines 5–12% (comparable to the 2008 GFC).
• Extreme ($g_A = 0.40$): Explosive regime; labor share collapses, velocity crashes, and consumption falls 15–30%.
• Policy Sensitivity: The crisis is averted if fiscal transfers ($\tau$) are implemented with a lag ($\ell$) of less than 1–2 years and are large enough to offset the labor share decline.

5. Significance and Policy Implications

Significance:

• Redefines AI Risk: Shifts the focus from "AI taking over" to "AI breaking the demand loop." It highlights that a macro-financial crisis can occur without a loss of control or a productivity bust.
• Indicator Warning: Warns that standard macro indicators (GDP, productivity) are "misleading" during rapid AI transitions due to the Ghost GDP wedge.
• Financial Stability: Identifies a novel transmission channel where AI displacement triggers systemic risk via concentrated defaults in prime mortgages and private credit portfolios.

Policy Toolkit:
The paper argues standard monetary policy is insufficient because it cannot restore the labor share or create new tasks. It proposes:

1. Fiscal Transfers: Progressive AI/compute taxation, expanded unemployment insurance, and a "Sovereign AI Dividend Fund" to redistribute productivity gains directly to households.
2. Financial Regulation: AI-augmented stress tests for private credit, mandatory mark-to-market reporting for private funds, and monitoring of the PE-insurer nexus.
3. Governance of Speed: Mandatory evaluation and staged deployment to slow the diffusion speed ($\kappa$) and capability growth ($g_A$) below the explosive threshold.
4. Labor Market Interventions: Shortened work weeks and targeted retraining for human-AI complementary skills.

Conclusion:
The paper concludes that the crisis is not predetermined. If task reinstatement is strong, diffusion is slow, or fiscal redistribution is fast and large, the economy can transition benignly. However, if AI capability grows rapidly while institutions remain anchored to human scarcity, the resulting "abundant intelligence and deficient demand" dynamic poses a severe, testable risk to global financial stability.