The Accumulation Paradox: When Markups Enable and Inhibit AI Adoption
This paper employs an agent-based model to demonstrate that market structure dynamics, particularly the absorption of productivity gains into higher markups and the bistable nature of adoption thresholds, explain the paradox between firm-level AI productivity gains and stagnant aggregate TFP growth.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
Imagine the economy as a massive, bustling marketplace with 200 different shops selling slightly different versions of the same product. In this marketplace, a new, magical tool called "AI" has arrived. This tool can make any shop incredibly efficient, allowing them to produce goods faster and cheaper.
The big mystery this paper tries to solve is: If every shop gets this super-efficient tool, why isn't the whole marketplace suddenly booming with growth?
The author, Liutao Hu, built a computer simulation (a "digital sandbox") to watch how these shops behave when they get this tool. Instead of assuming the shops are perfectly rational robots, the simulation treats them like real people who copy their neighbors, get scared of change, and care deeply about how much profit they can keep.
Here are the three main discoveries from the simulation, explained simply:
1. The "Profit-First" Trap (Markup Absorption)
Imagine a shop owner who buys the AI tool. Their costs drop, so they could lower their prices to attract more customers.
- In a crowded, competitive street: If there are 50 shops selling the same thing, the owner must lower prices to survive. The savings from the AI tool get passed on to you, the customer, and the whole economy grows.
- In a quiet, exclusive street: If the owner is one of only two shops in town, they don't need to lower prices. Instead, they keep the savings as extra profit (a higher "markup").
The Result: The paper finds that in the "exclusive" streets, the AI tool makes the shop owners rich, but it doesn't make the whole economy grow. The efficiency gains get "absorbed" into the owners' pockets rather than flowing down to lower prices. The simulation suggests this "profit-first" behavior is the biggest reason why the economy isn't seeing a massive boom, even though individual companies are getting better.
2. The "Slow-Motion" Race (The Temporal Reversal)
The simulation shows a surprising twist in how different types of shops adopt the AI tool over time. It's like a race with two distinct phases:
- Phase 1 (The Sprint): At the very beginning, the shops on the "crowded street" (competitive sectors) adopt the AI first. Why? Because information travels fast there, and it's easy to jump in. They are the early adopters.
- Phase 2 (The Marathon): After a few years, the race flips. The shops on the "exclusive street" (concentrated sectors) start to pull ahead. Why? Because they have been hoarding those extra profits (from the "Profit-First" trap). They have a deep wallet of cash to pour into expensive, long-term AI upgrades. The competitive shops, struggling to keep prices low, run out of cash for big investments.
The Prediction: If this model is right, we might see competitive industries (like some retail or services) leading in AI adoption right now, but by around 2029, the big, dominant companies in concentrated industries might overtake them.
3. The "Tipping Point" (Bistability)
The simulation discovered a dangerous "tipping point" in the market.
- Below the line: If a market is somewhat concentrated (a few big players), the AI adoption is high and stable.
- Above the line: If a market is too competitive (too many players fighting for scraps), the profits are so thin that no one can afford to invest in the AI tool, and adoption stalls.
- The Danger Zone: Right in the middle, the system becomes unstable. It's like a ball sitting on a hilltop. A tiny nudge in one direction sends it rolling into a "High Adoption" valley; a tiny nudge the other way sends it into a "Low Adoption" valley. Two identical markets could end up with completely different futures just based on a random starting condition.
The Role of "Copying" (Imitation)
The paper also looked at what happens if shops stop copying each other's strategies.
- The Finding: When shops stop copying, the market actually becomes more monopolistic (fewer big winners, more losers). Copying acts like a safety net that keeps the market diverse. It prevents the "rich get richer" spiral from becoming too extreme, even though it doesn't necessarily make everyone invest more.
The Bottom Line
The paper argues that the "AI Productivity Paradox" (why we see great tech but slow economic growth) isn't just about bad measurement or people losing jobs. It's about who gets the money.
If the market structure allows big companies to keep all the efficiency gains as extra profit, the whole economy doesn't benefit. The simulation suggests that for AI to truly boost the economy, we need a mix of competition (to force price drops) and enough stability (to allow for investment), but the current path seems to favor the former at the expense of the latter.
Important Note: The author is very clear that this is a simulation based on specific rules. It shows how these patterns could emerge from simple behaviors, but it is not a crystal ball predicting the exact future of the real world. It is a "what-if" story that helps us understand the mechanics of the problem.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.