Hook: The Number Nobody Is Trading On
$28 billion. Annualized. That is the price tag Apollo Research has attached to AI-driven wage compression in the United States labor market. Not job destruction. Not mass unemployment. Wage compression. The quiet, structural repricing of labor that does not show up in headlines because the jobs are still there.
Let me be direct: this is not a macro observation for your morning news digest. This is a signal. A price discovery event occurring in an asset class—human capital—that has no ticker, no order book, and no options chain. Yet it has the same structural implications for the broader economy as a yield curve inversion or a sudden liquidity dry-up in the treasury markets.
I have spent 24 years reading market data. I built my career on the premise that volatility is data waiting to be structured. This report is structured. But the market has not yet priced it. The question is: what is the smart-money trade on human capital repricing?
The answer is not a stock ticker. It is a structural repositioning — in startups, in skill acquisition, and in capital deployment. Let me walk you through the mechanics.
Context: The Labor Market Structure No One Is Watching
First, set the baseline. US unemployment has held between 3.7% and 4.0% for over a year. That is historically low. The labor market is tight by any standard measure. But there is a silent divergence: real wage growth continues to lag productivity growth. The output per worker is going up. The price of that worker is not following.
That divergence is the canopy under which Apollo's research operates.
The Apollo report identifies a $28 billion annual impact from AI compressing wages. Let me anchor that number. The US labor market represents approximately $12 trillion in annual wages. $28 billion is about 0.23% of that pool. A small fraction. But the marginal rate of change is the signal, not the level. With only about 20% of US businesses having deployed AI in production, this effect is early innings.
The mechanism is what matters. AI does not remove the job title. It removes the pricing power of the worker holding that title. Copilots and ChatGPT do not replace the software developer; they change the developer's marginal output from one-to-one to one-to-1.5 or one-to-2. In a market where demand for output is static, the employer's willingness to pay for a single worker's time decreases proportionally to the tool's efficiency multiplier.
This is not unemployment. This is a price discovery event in the labor market. I have seen this pattern before, and it reminds me of what happened in the 2020 DeFi summer — where the underlying collateral was intact, but the oracle pricing was distorted. The real structural vulnerability is not a crash; it is a silent repricing that goes unnoticed until it has been fully priced into the existing contracts.
Core: The Mechanics of Wage Compression
Let me break down the $28 billion figure into its component mechanics. This is where the data gets interesting.
Mechanism 1: The Marginal Output Effect
The standard economic model assumes a worker is paid close to their marginal revenue product. If a tool like GitHub Copilot or ChatGPT increases a developer's output by 30% to 50% without an increase in demand for that output, the marginal value of each individual worker contracts. The market does not say "hire more workers to create more output." The market says "the same output requires 30% less labor." The price of labor, not the quantity of labor, adjusts. This is the hidden substitution effect.
Mechanism 2: The Bargaining Power Shift
I have seen this dynamic in my own audits. When a protocol has a governance token, the holders have power over the protocol's direction. When that power is diluted, the token price compresses. The labor market is no different. AI is the protocol-level innovation that shifts the governance of wage formation from labor to capital. The employer now controls the marginal cost structure. The worker's share of the surplus has compressed.
Mechanism 3: The Entrepreneurial Cost Curve
The report highlights a secondary effect: AI lowers the cost of starting a business. Software development, content creation, customer service — these are now cheaper. The initial capital required to start a venture has dropped from the million-dollar range to the hundred-thousand-dollar range. This is supported by US new business registration data, which hit record highs in 2023-2024.
But here is the trap. If the barrier to entry is lower, the barrier to differentiation is also lower. You are not creating a moat; you are creating a commodity. An AI-generated code base is not an advantage. A competitor can generate the same code. The same content. The same customer service script. The result is a market flooded with homogenous startups. This is not a wave of innovation; it is a wave of competition for the same marginal yield. And in DeFi, we know what happens when there is no differentiated yield.
Mechanism 4: The Skill Premium Divergence
The $28 billion figure is not distributed evenly. AI is a leverage tool. For the skilled user — the operator — it is a multiplier. For the unskilled worker who is replaced in part by the tool, it is a discount. The result is an expanding spread between the top decile of earners and the bottom quartile.
This is not a generic "income inequality" statement. This is a technical market segmentation. The high-skill worker gets a 30% efficiency boost and can capture a premium. The low-skill worker is facing a market where the tool can do half of the task, and the employer prices that into the wage. The middle of the skill distribution is being squeezed. This is a barbell effect in human capital.
Contrarian: The 28 Billion Is an Underestimate and a Mispricing
The mainstream take on this data is: "AI is not eliminating jobs, it is just growing slowly." That is the optimistic, low-anxiety read. I see a different, more dangerous underlying structure.
First, the $28 billion is likely underestimated. The report captures direct wage compression — the observable reduction in salary offers for the same job. It does not capture the hidden costs: the unpaid hours workers spend learning the AI tools, the forced transition to contract or gig status, and the degradation of employment quality. These are off-balance-sheet items for the labor market. In my experience with DeFi, the risks that live in off-chain oracles are the ones that create the sudden sharp repricing. The invisible items are the ones that eventually crystallize.
Second, the wage compression effect is not a one-time reprice. It is a compounding effect. As AI deployment expands from 20% of US enterprises to a more mature penetration rate, the compression effect will scale non-linearly. The $28 billion annual figure is a point estimate at a low penetration rate. The marginal effect of the next 20% of enterprises adopting AI will not be additive; it will be exponential. This is the same pattern we saw in the adoption of algorithmic stablecoins — it was not the first wave that caused the crash, it was the second and third derivatives.
Third, the "entrepreneurship boom" is not a clean positive. It is a potential bubble. I saw this in the 2017 ICO era. The cost of creating a token project was near zero. The result was a flood of low-quality projects, a network of distributed pumps, and a massive transfer of wealth from retail to insiders. The same dynamic is now playing out in the startup economy. AI lowers the cost of code, but it also lowers the value of code. The market is not yet pricing the survivability rate of AI-driven startups. This is a risk that the market has not yet.
Takeaway: Trade the Repricing, Not the Narrative
The market has decided this is a headline item. It is not. This is a price discovery event for a new asset class of labor.
For the individual operator: Your labor is now a leveraged position. The leverage is AI. If you are not using the tool, you are short the market. If you are using it, you are long the skill premium. There is no neutral position. You are either acquiring the skill premium, or you are suffering the compression.
For the capital allocator: The market for AI-adjacent labor is about to be repriced. If you are investing in a startup that relies on AI to generate a moat, you are betting on a market that has no moat. The code is cheap. The data is cheap. The distribution is the only thing that matters. Do not confuse a low burn rate with a high survival rate.
For the policy observer: The risk is not unemployment; it is underemployment and the social instability that follows wage stagnation in a high-cost environment. The time window is 5 to 10 years. The reaction function is unknown.
The key metric to track is not the unemployment rate. It is the Employment Cost Index (ECI) and the divergence between productivity growth and compensation growth. If that divergence continues to widen, the $28 billion figure will be the floor, not the ceiling.
We do not chase pumps; we engineer the squeeze. The wage compression is a squeeze. It is being engineered by the capital owners who are deploying AI as a margin expansion tool. The question is whether the labor force is smart enough to get ahead of the tool, or become the liquidity that provides the exit.
Forward-Looking Signal
I am watching the ECI data for the next two quarters. If the wage growth for AI-exposed occupations (software, content, customer service) decelerates by more than 50 basis points relative to the broader market, we are not seeing a statistical anomaly. We are seeing the beginning of a structural repricing. The market that is slow to react to this will be the one that gets the margin called.
The smart capital will not be the one that is investing in AI. It will be the one that is investing in the transition costs of the labor that has been repriced — the training, the reskilling, and the new startups that are built by the displaced, not the displaced.
Alpha is not leverage. It is the ability to see the repricing before the order book does.
The order book for human capital is off-chain. But the signal is on.