The Toll Collectors: Mapping Who Takes What From the American Economy

Part 1 of the Economic Plumbing series — Where does the money actually go?

There is $29.7 trillion flowing through the American economy every year. That’s the 2024 GDP — the total value of all goods produced, services rendered, and transactions completed. But this number is misleading in an important way. It tells you the size of the river, but nothing about who’s drinking from it, who’s damming it, and who’s siphoning off enormous volumes before it reaches the fields.

This article series is an attempt to build the plumbing diagram of the U.S. economy — to trace flows of money and value, and identify the points where economic agents extract more than they contribute. In the language of economics, we’re mapping the rent-seekers: the toll collectors stationed at every bridge, demanding payment for passage rather than building anything new.

This isn’t ideology. This is accounting.

The Model: Producers, Intermediaries, and Extractors

Every economy has three kinds of participants:

Producers create new value. A farmer grows food. An engineer designs a bridge. A teacher educates a child. A factory worker assembles a car. The defining feature of producers is that the economy has more in it after they’ve done their work than before.

Intermediaries move value between producers and consumers. Banks connect savers with borrowers. Insurance companies pool risk. Trucking companies move goods. Intermediation is necessary — but it is a cost, not a product. The question is always: how much intermediation does the economy actually need, and what are we paying for it?

Extractors (or rent-seekers) take value out of the system without creating proportional value in return. They do this through market power, regulatory capture, information asymmetry, or simple positioning — they sit at a chokepoint and charge a toll. The critical distinction: removing an extractor would make the economy more efficient, not less.

The boundaries between these categories are blurry and contested. Wall Street will argue every dollar it earns is intermediation. A hospital will insist its billing department is a necessary cost of delivering care. The point of this series is to use data, not ideology, to estimate where the lines actually fall.

Exhibit A: The Finance Industry — A $600 Billion Paradox

Let’s start with Wall Street, because the data here is extraordinary.

In 1947, the U.S. finance industry consumed about 2.5% of GDP. By 2007, it consumed 7.9%. As of 2024, finance, insurance, real estate, and leasing collectively represent over 21% of GDP — the single largest sector of the American economy, larger than manufacturing, larger than technology, larger than healthcare.

How did the people who move money around come to earn more than the people who make things?

Philippon’s Paradox

NYU economist Thomas Philippon spent over a decade measuring the actual unit cost of financial intermediation — what it costs, per dollar of assets, for the finance industry to do its job of connecting savers to borrowers, managing risk, and allocating capital.

His finding, published in the American Economic Review, is one of the most striking results in modern economics: the unit cost of financial intermediation has remained approximately 2% for 130 years. It cost 2 cents per dollar in 1886. It costs 2 cents per dollar today.

Think about what that means. In 1886, banks processed loans with handwritten ledgers and horse-drawn messengers. Today they use algorithmic trading, fiber-optic cables, and AI-powered risk models. Every other industry that experienced this kind of technological transformation — retail, wholesale trade, logistics — saw its costs fall dramatically. Retail and wholesale trade shrank by about 20% as a share of GDP since 1970, thanks to technology and scale.

Finance didn’t. The technology got better, but the price didn’t drop. The savings were captured by the intermediaries, not passed to the customers.

Philippon estimates the excess cost — the amount the finance industry charges above what a reasonably efficient system would require — at approximately 2 percentage points of GDP. In 2024, that’s roughly $600 billion per year in economic resources consumed by financial intermediation beyond what’s necessary.

Where does that $600 billion go? Three places:

1. Asset Management Fees (~$300B+): Americans pay enormous sums to have their money “professionally managed,” even though the evidence overwhelmingly shows that active fund managers underperform passive index funds after fees. A study by French (2008) estimated investors spend 0.67% of asset value annually trying to beat the market — and failing, on average, by definition. The rise of hedge funds, private equity, and alternative asset managers has kept total asset management costs high even as mutual fund fees have declined.

2. Trading and Financial Engineering (~$150B+): High-frequency trading now accounts for 50–70% of equity trading volume. These firms invest billions in microsecond advantages — co-locating servers next to exchange data centers, building microwave relay towers across the countryside — all to capture fractions of a penny per trade. The social value of this activity is approximately zero. As the CEPR noted, there is “no evidence that increased liquidity has led to better prices or to more insurance.”

3. Compensation (~$150B+): The mean salary in New York City’s finance industry rose from $80,000 in 1981 to $360,000 in 2011, while average NYC salaries went from $40,000 to $70,000. Finance sector wages track almost perfectly with the income share of the top 1%. This is not a coincidence — it’s the same money.

Exhibit B: Healthcare — The $1.8 Trillion Labyrinth

If finance is a toll bridge, American healthcare is an entire medieval kingdom of toll bridges, each owned by a different lord, each demanding tribute.

In 2024, U.S. healthcare spending reached $5.3 trillion — 18% of GDP, or $15,474 per person. For comparison, peer OECD nations average 11.2% of GDP. The U.S. spends roughly twice as much per capita as Switzerland, the second-highest spender. That 6–7 percentage point gap between U.S. spending and the OECD average represents approximately $1.8 to $2 trillion per year in excess costs.

What does America get for spending twice as much as everyone else? Worse outcomes. Life expectancy is lower than comparable nations. Infant mortality is higher. Disability-adjusted life years are going in the wrong direction. Diabetes control is getting worse. As the Penn LDI panel noted in early 2025, a reasonable spending target for the U.S. would be approximately 12% of GDP — and achieving that would free up roughly $1.8 trillion annually.

The excess isn’t because Americans use more healthcare. Studies consistently show that utilization rates for many services are actually lower in the U.S. than in other wealthy countries. The excess comes from higher prices driven by several extraction mechanisms:

The Middleman Layer: The U.S. healthcare system has six layers of intermediaries between the patient and the care. People and employers pay insurance companies, which pay pharmacy benefit managers and preferred provider organizations, which pay doctors, hospitals, and pharmacies, which pay wholesalers, which pay manufacturers. Each layer extracts a margin. The Economist calculated that healthcare middlemen alone capture about $126 in excess profits per American per year — roughly $42 billion total — and the middlemen’s share of industry profits has grown from 20% to 41% in fifteen years.

Administrative Waste: The complexity of the multi-payer system generates enormous administrative overhead. Estimates of total waste in U.S. healthcare — from pricing failures, administrative complexity, overtreatment, and fraud — range from $600 billion to $1.9 trillion per year according to research compiled in the American Journal of Public Health.

Regulatory Capture via Lobbying: Healthcare sector lobbying expenditures grew by more than 70% from 2000 to 2020. In 2024, pharmaceutical and health products companies alone spent $294 million on lobbying. This spending is an investment: it protects pricing power, blocks regulatory reform, and maintains the complexity that sustains the intermediary layers.

Exhibit C: The CEO-to-Worker Extraction Engine

In 1965, the average CEO of a top-350 U.S. firm earned 21 times what their typical worker made. By 2000, that ratio hit 376:1. In 2024, it’s 281:1 across the S&P 500 — and among the 100 largest low-wage employers, IPS found the average ratio is 632:1.

The mechanism behind this divergence is not mysterious. From 1978 to 2024, CEO compensation grew by more than 1,000%. Typical worker pay grew by 26%. Economywide productivity over the same period grew 74.8%. Workers produced dramatically more per hour — but almost none of the gains went to wages.

Where did the gains go? Substantially to stock buybacks, which function as an upward wealth pump:

• From 2019–2024, the 100 largest low-wage employers spent $644 billion on stock buybacks — repurchasing their own shares to inflate stock prices and, consequently, the stock-based compensation of their executives.
• A majority (56 of 100) of these firms spent more on buybacks than on long-term capital improvements.
• Lowe’s alone spent $46.6 billion on buybacks in six years — enough for an annual $28,456 bonus for each of its 273,000 employees. Instead, median worker pay declined.
• U.S. corporations are on track to exceed $1 trillion in annual buybacks.

Stock buybacks were considered a form of market manipulation and were effectively illegal before 1982, when the SEC adopted Rule 10b-18 creating a safe harbor for repurchases. Since then, they’ve become the primary mechanism by which corporate profits are converted into executive wealth.

At least 32 U.S. billionaires owe their wealth to these low-wage companies, with a combined net worth of $827 billion. This is wealth that was created by workers, captured by the firm, and redirected upward through stock-based compensation.

Exhibit D: The Data Harvest — Social Media, AI, and the Newest Toll Bridge

The previous three exhibits describe toll bridges that have been accumulating for decades. This one is different. It was built in under twenty years, and it may already be the most efficient extraction machine in human history.

You Are the Product

In 2024, Meta Platforms — the company that owns Facebook, Instagram, WhatsApp, and Messenger — generated $164.5 billion in revenue. Nearly all of it came from advertising. The company reported 3.35 billion daily active users across its platforms. The math is straightforward: Meta’s global average revenue per person (ARPP) reached $49.63 in 2024. In the U.S. and Canada, that figure was roughly $227 per user per year.

That’s the price tag on your attention, your social graph, your location data, your purchase intent, your political leanings, and every scroll, click, and hover you’ve made since you signed up for a free account.

The word “free” deserves scrutiny here. Meta’s services are free in the way a livestock pen is free to the cattle. You aren’t the customer — you’re the raw material. Shoshana Zuboff, the Harvard scholar who coined the term “surveillance capitalism,” describes it precisely: companies like Google and Meta collect human experience as “free raw material for hidden commercial practices of extraction, prediction, and sales.”

The scale of this extraction is astonishing. U.S. digital advertising spending alone exceeded $101 billion in 2024. The global data brokerage industry — the firms that buy, aggregate, and resell personal data — has grown to an estimated $315 billion. None of the humans whose data fuels this industry receive meaningful compensation.

The Efficiency of Extraction

What makes Big Tech’s extraction model unique is its extraordinary efficiency — the ratio of revenue to employees. Traditional rent-seeking industries like finance and healthcare at least employ large workforces to extract their tolls. Tech platforms extract comparable or larger sums with a fraction of the labor:

• Nvidia generated $3.6 million in revenue per employee in 2024, and $2 million in profit per employee — more than double its nearest tech competitor.
• Apple produced $2.38 million per employee.
• Meta generated roughly $2.3 million per employee.
• Alphabet (Google) hit $1.91 million per employee.

Compare that to the U.S. median income of around $60,000. Nvidia generates as much profit per worker in one day as a typical American earns in a year. This isn’t a commentary on the talent of Nvidia’s engineers — it’s a measure of how much leverage data-driven and platform-based business models exert over the broader economy.

AI: The Second Harvest

And now, the extraction model is being extended. The explosion of generative AI has created a new class of extraction: the mass scraping of human-created content — text, images, video, code, music — to train AI models that generate enormous revenue without compensating the creators.

The numbers are staggering and recent:

• OpenAI’s annualized revenue surged from $200 million in early 2023 to $13 billion by late 2025.
• Anthropic grew from $87 million in early 2024 to $7 billion by late 2025.
• xAI reached $500 million by mid-2025.

This revenue was built on training data that was overwhelmingly obtained for free. Common Crawl, the primary training dataset for most large language models, described itself as contributing “82% of raw tokens used to train GPT-3.” That data — billions of web pages, books, articles, forum posts, and creative works — was scraped without compensation to the people who wrote it.

The emerging market for AI training data is only beginning to put a price on what was taken. When HarperCollins struck a deal with Microsoft in late 2024 to license nonfiction titles for AI training, the price was $5,000 per title for three years. That’s a data point that reveals the magnitude of the unpaid extraction: if a single book is worth $5,000 to train on, and GPT-4 was trained on the equivalent of millions of books, articles, and creative works, the total uncompensated value runs into the billions.

Multiple lawsuits are now underway — from publishers, musicians, visual artists, and news organizations — alleging that AI companies built multi-billion-dollar businesses on content they never paid for. As one UK publishing rights organization testified: on average, publishers are targeted by 15 different AI bots extracting data for model training.

The Attention Tax

There’s one more dimension of tech extraction that doesn’t show up in revenue figures: the attention cost.

Americans spend an average of over 2 hours per day on social media. That’s roughly 730 hours per year — equivalent to about 18 full work weeks. This time is the raw material that platforms convert into advertising revenue. If you valued that time at even minimum wage ($7.25/hour), the aggregate value of American attention captured by social media platforms exceeds $1.2 trillion per year.

The platforms return a fraction of this captured value to content creators. YouTube shares approximately 55% of ad revenue with creators. Most other platforms share far less, or nothing. TikTok’s creator fund was widely criticized for paying fractions of a cent per view. Instagram and Facebook pay creators only through select programs. The overwhelming majority of value generated by user-created content flows to the platform, not the creator.

This is the defining feature of the data extraction toll bridge: unlike Wall Street or the healthcare system, the raw material — your data, your attention, your content — is obtained for free or at near-zero marginal cost. The users do the work of creating the product (content), attracting the audience (engagement), and generating the signal (behavioral data). The platform collects the toll.

The Aggregate Toll

Let’s do rough arithmetic on the total extraction:

These categories overlap — some healthcare excess flows through finance, some CEO extraction is embedded in healthcare prices, and digital advertising is funded from the budgets of companies across all sectors. But even with conservative estimates and deduplication, the picture is clear: somewhere between $2 and $3 trillion per year is consumed by intermediation and extraction beyond what a reasonably efficient economy would require.

That’s $6,000 to $9,000 per American, per year, in economic resources that are consumed by toll collection rather than value creation.

What This Series Will Cover

In subsequent articles, we’ll map each extraction layer in detail:

• Part 2: The Finance Plumbing — A detailed flow diagram of how money moves through Wall Street, who takes what at each step, and the specific mechanisms of extraction (HFT, PE fees, credit card interchange, payment processing).
• Part 3: The Healthcare Maze — Tracing a single dollar of healthcare spending from employer to patient, through every intermediary layer, with data on margins at each step.
• Part 4: The Data Harvest — How social media, AI companies, and data brokers turned human experience into the world’s most profitable raw material — and what the content creators, users, and artists get in return.
• Part 5: The Real Estate Rent Machine — How housing became a financial asset rather than shelter, and the agents who profit from that transformation.
• Part 6: The Regulatory Capture Loop — Mapping the revolving door between industry and regulators, and estimating the economic cost of regulatory capture.
• Part 7: Who Builds the Bridges? — Possible reforms, from Philippon’s fintech thesis to single-payer healthcare to data dividends and buyback regulation, and their estimated economic impact.

A Note on Framing

This analysis is descriptive, not prescriptive. It uses the word “extraction” not as a moral judgment but as an economic measurement: the difference between what an industry charges and what a competitive, efficient version of that industry would charge for the same service.

Many of the people who work in finance, healthcare administration, and corporate management are doing their jobs well within the rules of the system. The problem isn’t individual behavior — it’s system design. The toll bridges exist because they were built — by deregulation, by regulatory capture, by the gradual accumulation of market power.

The question isn’t whether these tolls exist. The data makes that clear. The question is what to do about them.

Brian James Curry is an Enterprise AI Strategist, Data and Marketing Scientist, and Applied Researcher. He is the founder of Vector1 Research, an independent research lab focused on marketing economics, causal systems, cognitive AI architecture, and the economic implications of intelligent systems. He also founded KC AI Lab, a data science community and incubator for applied machine learning education.

Sources & Data

• Philippon, T. (2015). “Has the US Finance Industry Become Less Efficient?” American Economic Review, 105(4), 1408–1438.
• Greenwood, R. & Scharfstein, D. (2013). “The Growth of Finance.” Journal of Economic Perspectives, 27(2), 3–28.
• CMS National Health Expenditure Data (2024). Centers for Medicare & Medicaid Services.
• Peterson-KFF Health System Tracker (2025). Health spending comparisons.
• IPS Executive Excess Report (2025). Institute for Policy Studies.
• EPI CEO Pay Series (2024). Economic Policy Institute.
• OpenSecrets Federal Lobbying Data (2024).
• FRED Economic Data, Bureau of Economic Analysis.
• Shrank, W.H., Rogstad, T.L., Parekh, N. (2019). “Waste in the US Health Care System.” JAMA, 322(15), 1501–1509.
• Zuboff, S. (2019). The Age of Surveillance Capitalism. PublicAffairs.
• Meta Platforms 10-K Annual Report (2024). SEC Filing.
• Visual Capitalist / Multiples (2025). Revenue per employee analysis.
• Epoch.ai (2025). AI company revenue estimates.
• Harvard Kennedy School (Fall 2025). AI and the Surveillance Economy.
• ManageEngine Insights (2026). “Ads in LLMs and the Expansion of the Surveillance Economy.”
• Grand View Research (2025). U.S. Digital Advertising Market.
• TechPolicy Press (2025). “How the Emerging Market for AI Training Data is Eroding Big Tech’s Fair Use Defense.”

Next in the series: Part 2 — The Finance Plumbing