Recently, a groundbreaking report titled “Project Iceberg: The Iceberg Index and Measuring Skill Exposure in the AI Economy” was jointly released by Ayush Chopra and Santanu Bhattacharya of the Massachusetts Institute of Technology (MIT), in collaboration with Oak Ridge National Laboratory and multiple state government policy offices.

The conclusions drawn in this report left us both chilled to the bone—and exhilarated.

According to MIT, what we currently see as the AI boom represents merely 2.2% of the iceberg floating above water.

Beneath the surface lies a staggering 11.7%—a hidden “latent technology exposure” valued at $1.2 trillion.

What does this really mean?

And what are its implications?

1. Simulating the AI Economy with Supercomputers

If we had to choose one core metaphor for this report, it would be this:

Don’t use a thermometer to measure wind speed.

For the past two centuries, economists have relied on GDP, unemployment rates, and total factor productivity to gauge technological revolutions. This worked well during the steam engine era—machines entered factories, workers lost jobs, output doubled, and the data spoke clearly.

But in the age of AI, these traditional metrics have become lagging indicators.

By the time GDP reflects a shift, the transformation has already concluded—sometimes leaving devastation in its wake. Unlike the steam engine, AI is invisible and pervasive.

Consider this: when a nurse uses AI to optimize shift scheduling and thereby cares for two additional patients, GDP doesn’t capture that gain. When a quality-control manager in a Midwestern factory replaces visual inspections with an AI vision model, unemployment statistics haven’t yet registered the change.

To overcome this blind spot, the Project Iceberg team undertook something exceptionally ambitious.


They leveraged the world-class Frontier supercomputer to build Large Population Models (LPMs)—essentially creating a digital sandbox simulating a parallel-universe United States.

Inside this AgentTorch-based simulation live 151 million digital workers, each modeled as an autonomous agent with:

A specific occupation (covering 923 job types),

A unique skill profile (mapped across 32,000 skills from the ONET database),

A real geographic location (distributed across 3,000 U.S. counties).

The researchers then introduced various AI tools—from GitHub Copilot to Zapier automation workflows—and hit “run,” simulating billions of work interactions among these 151 million agents.


This approach represents a paradigm shift—a dimensional leap over conventional social science methods.

It was precisely through this god’s-eye view that we glimpsed, for the first time, the colossal structure long missed by traditional economics: the Iceberg Index.

2. Three Key Insights from the Report

The Frontier supercomputer simulations revealed three profound insights:

Insight 1: Rethinking the Geography of AI

Ask an average person:

“Where is the epicenter of America’s AI revolution?”

Nine out of ten will say: San Francisco, Seattle, or Boston.

But Project Iceberg’s heat maps reveal a striking phenomenon: production ≠ exposure.

Above water (2.2%): Visible wage value concentrated in California and Washington State.

This is where engineers train models and VCs debate valuations—the “arsenal” of AI. Loud, visible, but economically tiny.

Below water (11.7%): Hidden value spread across all sectors—a quiet efficiency revolution.


Most shockingly, this $1.2 trillion latent opportunity isn’t in Silicon Valley—it’s in Ohio, Tennessee, Utah, and Michigan: America’s so-called “Rust Belt.”

Why?

Because AI excels at tasks involving:

Complex document processing,

Workflow coordination,

Large-scale scheduling,

Compliance judgment support.

Where are such tasks most abundant?

Not in coding startups in San Francisco—but in:

Manufacturing hubs with sprawling supply chains (e.g., Ohio, Iceberg Index: 11.8%),

Logistics nerve centers like FedEx’s Memphis hub in Tennessee (11.6%),

Healthcare systems managing massive patient records.

The report calls this “Automation Surprise”: regions that appear low-tech are actually sitting atop explosive AI readiness due to their dense, complex skill structures.

Implication for China:

What we label “old economy” may in fact be AI’s richest new frontier.

Insight 2: AI Is Eating High-Skill, High-Pay Work First

Many people see “11.7% technology exposure” and breathe a sigh of relief:

“Only about 10%—mass unemployment isn’t imminent.”

This is a dangerous misreading.

AI is redefining what work is valuable.

In past industrial revolutions, automation targeted manual labor—the dirtiest, hardest, cheapest tasks (e.g., assembly,搬运).

But Project Iceberg shows:

This time, AI is devouring high-salary cognitive skills.

Take a senior financial compliance officer earning $200,000/year:

A-skills (60% of time): Reading hundreds of pages of new regulations, comparing clauses, researching past violations, drafting initial risk reports.

B-skills (30%): Negotiating gray-area decisions with business units, bearing legal liability.

C-skills (10%): Meetings, networking, trust-building.

Before AI, companies paid a premium for A-skills—this was the “knowledge premium.”

Now, AI can complete A-skills in minutes—with higher accuracy.

Suddenly, the $120,000 portion of that salary tied to reading and analysis loses its economic foundation.

It’s like carving away the tender filet mignon and leaving only the tough gristle.

Insight 3: Job Hollowing—The Illusion of Stability

One puzzling finding:

Traditional economic indicators show almost no correlation with AI impact (R2 < 0.05).


If AI is this disruptive, why aren’t GDP and unemployment reacting?

Because we’re witnessing “job hollowing.”

AI isn’t immediately firing people—it’s creating a competence illusion.

Previously, a seasoned analyst needed 10 years to produce a flawless research report.

Now, a fresh graduate with three AI agents can produce an 80–90 point report.

On the surface: everyone’s still employed. No layoffs.

In reality: the moat of experience has been flattened.

When juniors + AI match seniors’ output, employers may not cut headcount—but the market will reprice human capital.

Just as mental arithmetic lost value after calculators became ubiquitous, general cognitive labor is now deflating.

GDP appears stable—not because nothing changed, but because once-expensive intellectual services have become as cheap as tap water.

Deflationary pressure is already shadowing every white-collar worker.

So, don’t comfort yourself by saying, “I’m not in the 11.7%.”*

Ask instead: Once AI strips away the core value of your role, do your remaining skills still justify your current salary?

The future won’t pay for stored knowledge—that’s now free.

It will pay only for:

Decision-making under uncertainty,

Interpersonal trust,

Physical action in the real world—

the 88.3% AI cannot yet touch.

3. Lessons for China

Though focused on the U.S., this report holds profound implications for China.

If San Francisco mirrors Beijing’s Haidian District or Shanghai’s West Bund (where everyone obsesses over model parameters and compute power),

then America’s Rust Belt corresponds to China’s Foshan, Suzhou, Ningbo, Changsha, and similar industrial heartlands.

These regions host:

The world’s most complete manufacturing ecosystems,

The most intricate logistics networks,

The densest domestic and international trade workflows.

According to the Iceberg logic, China’s true AI explosion won’t start in a Beijing lab—it will ignite in a Ningbo injection-molding factory’s production scheduler or an Yiwu cross-border e-commerce back office.

Moreover, China’s Iceberg Index is likely even higher than America’s, thanks to:

Longer, more fragmented supply chains,

More administrative layers requiring coordination.

For Chinese policymakers and entrepreneurs, this is a clarion call:

Focus on industries that appear low in “tech content” but extremely high in data complexity.

Seek out “hidden champions” burdened by intricate processes and high labor costs.

That’s where China’s AI economic epicenter truly lies.

Recently, a groundbreaking report titled “Project Iceberg: The Iceberg Index and Measuring Skill Exposure in the AI Economy” was jointly released by Ayush Chopra and Santanu Bhattacharya of the Massachusetts Institute of Technology (MIT), in collaboration with Oak Ridge National Laboratory and multiple state government policy offices.

The conclusions drawn in this report left us both chilled to the bone—and exhilarated.

According to MIT, what we currently see as the AI boom represents merely 2.2% of the iceberg floating above water.

Beneath the surface lies a staggering 11.7%—a hidden “latent technology exposure” valued at $1.2 trillion.

What does this really mean?

And what are its implications?

1. Simulating the AI Economy with Supercomputers

If we had to choose one core metaphor for this report, it would be this:

Don’t use a thermometer to measure wind speed.

For the past two centuries, economists have relied on GDP, unemployment rates, and total factor productivity to gauge technological revolutions. This worked well during the steam engine era—machines entered factories, workers lost jobs, output doubled, and the data spoke clearly.

But in the age of AI, these traditional metrics have become lagging indicators.

By the time GDP reflects a shift, the transformation has already concluded—sometimes leaving devastation in its wake. Unlike the steam engine, AI is invisible and pervasive.

Consider this: when a nurse uses AI to optimize shift scheduling and thereby cares for two additional patients, GDP doesn’t capture that gain. When a quality-control manager in a Midwestern factory replaces visual inspections with an AI vision model, unemployment statistics haven’t yet registered the change.

To overcome this blind spot, the Project Iceberg team undertook something exceptionally ambitious.


They leveraged the world-class Frontier supercomputer to build Large Population Models (LPMs)—essentially creating a digital sandbox simulating a parallel-universe United States.

Inside this AgentTorch-based simulation live 151 million digital workers, each modeled as an autonomous agent with:

A specific occupation (covering 923 job types),

A unique skill profile (mapped across 32,000 skills from the ONET database),

A real geographic location (distributed across 3,000 U.S. counties).

The researchers then introduced various AI tools—from GitHub Copilot to Zapier automation workflows—and hit “run,” simulating billions of work interactions among these 151 million agents.


This approach represents a paradigm shift—a dimensional leap over conventional social science methods.

It was precisely through this god’s-eye view that we glimpsed, for the first time, the colossal structure long missed by traditional economics: the Iceberg Index.

2. Three Key Insights from the Report

The Frontier supercomputer simulations revealed three profound insights:

Insight 1: Rethinking the Geography of AI

Ask an average person:

“Where is the epicenter of America’s AI revolution?”

Nine out of ten will say: San Francisco, Seattle, or Boston.

But Project Iceberg’s heat maps reveal a striking phenomenon: production ≠ exposure.

Above water (2.2%): Visible wage value concentrated in California and Washington State.

This is where engineers train models and VCs debate valuations—the “arsenal” of AI. Loud, visible, but economically tiny.

Below water (11.7%): Hidden value spread across all sectors—a quiet efficiency revolution.


Most shockingly, this $1.2 trillion latent opportunity isn’t in Silicon Valley—it’s in Ohio, Tennessee, Utah, and Michigan: America’s so-called “Rust Belt.”

Why?

Because AI excels at tasks involving:

Complex document processing,

Workflow coordination,

Large-scale scheduling,

Compliance judgment support.

Where are such tasks most abundant?

Not in coding startups in San Francisco—but in:

Manufacturing hubs with sprawling supply chains (e.g., Ohio, Iceberg Index: 11.8%),

Logistics nerve centers like FedEx’s Memphis hub in Tennessee (11.6%),

Healthcare systems managing massive patient records.

The report calls this “Automation Surprise”: regions that appear low-tech are actually sitting atop explosive AI readiness due to their dense, complex skill structures.

Implication for China:

What we label “old economy” may in fact be AI’s richest new frontier.

Insight 2: AI Is Eating High-Skill, High-Pay Work First

Many people see “11.7% technology exposure” and breathe a sigh of relief:

“Only about 10%—mass unemployment isn’t imminent.”

This is a dangerous misreading.

AI is redefining what work is valuable.

In past industrial revolutions, automation targeted manual labor—the dirtiest, hardest, cheapest tasks (e.g., assembly,搬运).

But Project Iceberg shows:

This time, AI is devouring high-salary cognitive skills.

Take a senior financial compliance officer earning $200,000/year:

A-skills (60% of time): Reading hundreds of pages of new regulations, comparing clauses, researching past violations, drafting initial risk reports.

B-skills (30%): Negotiating gray-area decisions with business units, bearing legal liability.

C-skills (10%): Meetings, networking, trust-building.

Before AI, companies paid a premium for A-skills—this was the “knowledge premium.”

Now, AI can complete A-skills in minutes—with higher accuracy.

Suddenly, the $120,000 portion of that salary tied to reading and analysis loses its economic foundation.

It’s like carving away the tender filet mignon and leaving only the tough gristle.

Insight 3: Job Hollowing—The Illusion of Stability

One puzzling finding:

Traditional economic indicators show almost no correlation with AI impact (R2 < 0.05).


If AI is this disruptive, why aren’t GDP and unemployment reacting?

Because we’re witnessing “job hollowing.”

AI isn’t immediately firing people—it’s creating a competence illusion.

Previously, a seasoned analyst needed 10 years to produce a flawless research report.

Now, a fresh graduate with three AI agents can produce an 80–90 point report.

On the surface: everyone’s still employed. No layoffs.

In reality: the moat of experience has been flattened.

When juniors + AI match seniors’ output, employers may not cut headcount—but the market will reprice human capital.

Just as mental arithmetic lost value after calculators became ubiquitous, general cognitive labor is now deflating.

GDP appears stable—not because nothing changed, but because once-expensive intellectual services have become as cheap as tap water.

Deflationary pressure is already shadowing every white-collar worker.

So, don’t comfort yourself by saying, “I’m not in the 11.7%.”*

Ask instead: Once AI strips away the core value of your role, do your remaining skills still justify your current salary?

The future won’t pay for stored knowledge—that’s now free.

It will pay only for:

Decision-making under uncertainty,

Interpersonal trust,

Physical action in the real world—

the 88.3% AI cannot yet touch.

3. Lessons for China

Though focused on the U.S., this report holds profound implications for China.

If San Francisco mirrors Beijing’s Haidian District or Shanghai’s West Bund (where everyone obsesses over model parameters and compute power),

then America’s Rust Belt corresponds to China’s Foshan, Suzhou, Ningbo, Changsha, and similar industrial heartlands.

These regions host:

The world’s most complete manufacturing ecosystems,

The most intricate logistics networks,

The densest domestic and international trade workflows.

According to the Iceberg logic, China’s true AI explosion won’t start in a Beijing lab—it will ignite in a Ningbo injection-molding factory’s production scheduler or an Yiwu cross-border e-commerce back office.

Moreover, China’s Iceberg Index is likely even higher than America’s, thanks to:

Longer, more fragmented supply chains,

More administrative layers requiring coordination.

For Chinese policymakers and entrepreneurs, this is a clarion call:

Focus on industries that appear low in “tech content” but extremely high in data complexity.

Seek out “hidden champions” burdened by intricate processes and high labor costs.

That’s where China’s AI economic epicenter truly lies.