How the trillion-dollar "hidden output" in the AI era reshapes economic growth

Author: Financial Wang; Source: X, @yuyy614893671

Today I read an excellent report that sparked my imagination. When measuring the economic output created by AI, indeed, a piece of the puzzle has been overlooked, and traditional economics and statistics have not found an appropriate way to measure it. This is the Dark Output of AI.

First, I will provide a logical diagram of the report to help everyone better understand the related content:

We will now formally analyze the content of the article.

1. AI's Dark Output: Measurement Methods of Economic Activity and Their Blind Spots

This diagram accurately depicts the distribution of value within the entire economy and the blind spots of existing statistical tools.

The three criteria for measuring human economic activity

(1) Produced (Actual labor / Real work): Has this activity involved real labor and created real economic value?

(2) Priced (Market price): Is this activity sold at a clear, visible market price?

(3) Measured (Captured by GDP): Is the official national account finally recording this activity accurately?

In real social activities, there are different situations:

1. Intersection of Three Realms: Formal Market

Produced + Priced + Measured

This is the core area most concerned with by traditional economics and official statistical bureaus. It includes wages, product sales, and business contracts, constituting the vast majority of traditional GDP.

For example, a company spends $10,000 to hire an external agency for HR services, or you pay a lawyer to draft a legal document. These activities create value, have clear market prices, and are perfectly recorded in national accounts.

2. Presence of Price but Outside Regulation: Gray/Illicit Market

Produced + Priced (but not measured)

These activities do provide some service or product, with real monetary transactions and market prices, but because they are underground or illegal, they evade official macro data statistics.

3. Measured but No Market Pricing: Government Services (Measured at Cost)

Produced + Measured (but without direct market prices)

Many public services provided by the government (such as basic education, urban security) generate huge social and economic value and are included in GDP.

However, since they usually do not charge citizens per service (lacking market prices), official figures often only estimate their output by calculating “input costs” (e.g., civil servant salary bills).

4. The Numerical Game of Statistical Agencies: Pure Imputations

Measured (but not actually priced, nor with new production)

Data in the pure blue zone or at the border with the priced zone are estimates, most notably “imputed rent of owner-occupied housing.”

If you live in your own house, you do not pay actual rent to anyone, nor does it generate new economic activity. But to maintain macro data consistency, the national accounts “force-estimate” a rent you pay to yourself and include it in GDP.

5. Massive Economic Dark Web: Production Only (PRODUCED-only)

Produced only (not priced, not measured)

This is a huge blind spot in the economy. It includes “household production” and “volunteers/digital public resources.” Traditionally, unpaid caregiving activities like raising children, caring for the elderly, and daily chores in the home amount to up to 16.4 billion hours daily. Under current accounting conventions, these activities that create enormous survival value are recorded as zero because they do not involve monetary exchanges.

6. Separation of AI Output and Dark Output

The red circle in the center perfectly reveals how AI breaks the existing measurement system:

The circle of AI output is entirely within the “Produced” (PRODUCED) green circle, meaning AI-assisted or generated work undoubtedly creates real economic value.

However, only a tiny slice of AI output extends into the “formal market” at the center, representing only visible token consumption, API calls, or fixed subscription fees for AI software. For example, when the cost of drafting a basic legal will drops from $150 to just $0.50 API cost, the original $149.50 transaction in the “formal market” disappears entirely. The same value remains within the “Produced” circle, but due to price collapse and internalized transaction methods, it falls outside the “Priced” and “Measured” categories.

Existing statistical systems interpret this disappearance of receipts as inflation rising and economic output shrinking. AI’s Dark Output is increasingly pushing a large proportion of macroeconomic activities into that blind zone of production without valuation, causing the true feeling of the economy to completely disconnect from macro dashboards.

2. Why AI’s Output Cannot Be Fully Measured and Recorded: The Existence of Measurement Errors

When AI assists or takes over a task, the output does not automatically vanish. It only disappears from national accounts when prices fall, or worse, when the task shifts from external procurement to internal company processes; various measurement errors can also lead to missing economic output data, such as:

1. Boundary Shifts

Boundary shifts refer to work originally purchased on the market being transferred internally within companies or households. Paid research reports become internal AI workflows. Outsourced tasks become employee tasks. The value may still exist, but the transactions that make it visible disappear.

2. Price Collapse

The quantity and quality of services have no truly independent measurement standards. Income, wages, and working hours are recorded, but the quantity cannot be quantified. Legal services lack standard units, literature reviews have no tonnage units, consulting services have no barrel units; if accounts show income decline (due to falling prices) while average wages rise (due to layoffs of junior staff), it is interpreted as intensified inflation and decreased productivity and output.

3. Industry Mismatch

When AI creates value in one industry but transactions appear in another, industry mismatch occurs. For example, hospitals may use AI to speed up paperwork, but if AI’s only reflection is income for AI companies or software providers, it distorts national statistics. GDP by industry may make AI suppliers seem like the source of value, while industries adopting AI appear stagnant.

4. Invisible New Jobs

Real economic benefits are being generated, such as AI writing a report for humans with just a few tokens, helping us prepare better for meetings, but this value cannot be reflected anywhere. Any reasonable macroeconomic measure must account for this to some extent; otherwise, the prosperity of AI might be interpreted as AI decline in the data.

3. What Does AI’s Dark Output Actually Produce: Profit + Consumer Surplus

So, what kind of impact does the current method of measuring the economy have? Due to biases in official statistics, at some future stage, we are likely to see a situation like the one in the diagram above: CPI stagnates, GDP does not rise.

The nearly flat blue line (measured real per capita GDP) and yellow line (measured CPI index) represent the data seen by traditional statistical bureaus. Based on these data, the official diagnosis is pessimistic: “AI has not delivered the expected results—prices remain firm, economic growth is sluggish, but our spending on AI is higher than ever” (as described in the yellow-bordered text in the lower left corner).

However, at this moment, “real productivity is rising sharply”—the large shaded area between the white dashed line and the blue solid line (including the deep purple “transformed into profit” and dark gray “transformed into consumer surplus”) is AI’s hidden Dark Output.

When AI greatly reduces work costs, these saved funds mainly go to two places: one is excess profits for companies; the other is enabling consumers to obtain massive utility for the same money (for example, instead of paying someone to research, now they generate it instantly with AI for free).

Because these enormous real values do not form new market transactions with observable prices, if AI’s output is not sold at visible prices, GDP cannot capture these outputs beyond token expenditure.

4. Lessons from History: The Repetition of Solow Paradox

To better understand this phenomenon, we can look back at history. This is not the first time such a situation has occurred.

In the 1980s and 1990s, when personal computers first became widespread, macroeconomic data similarly failed to detect the contribution of the emerging computer revolution to the economy. The famous economist Robert Solow famously quipped: “You can see the computer age everywhere but in the productivity statistics.”

The official response was extremely slow. It wasn’t until 2013 that the U.S. undertook a tedious methodological revision, officially including R&D and intellectual property investments into GDP calculations. This single move directly “added” about $3.6 trillion to total output in the 1990s.

5. How Does AI’s Dark Output Affect the Economy?

Previous computers were just tools, but AI is directly taking over mental labor in the service sector. Due to inherent measurement difficulties in services, when AI causes certain service costs to plummet, GDP often records it as an economic decline (because transaction volume shrinks), or even shows inflation in the data.

The scale of this measurement challenge brought by AI will dwarf all previous statistical blind spots. That’s why, if you only focus on the traditional GDP dashboard, you might mistakenly think the economy is stagnant, while in reality, an unmeasured productivity revolution is boiling intensely beneath the surface.