---
cursor:
  subagentId: "bc-de1f9f63-d955-5e03-8c05-5cb47ac08e3d"
---

# Facts for a GM-to-OpenAI concentration graphic

Checked 24 September 2026. Figures are nominal unless noted. Market values move daily.

## The claim the facts support

Participation changed its channel.

At the GM peak, a leading company put hundreds of thousands of people on one payroll, most of them hourly, and large-firm retirement was a pension the employer owed. Today the firms that hold the new market value employ far fewer people per dollar of value, and that value is equity. Equity sits with the top of the wealth distribution and, for OpenAI, mostly with a private shareholder list.

## The claim the facts do not support

- Big employers did not vanish. Walmart reports about 2.1 million employees. Amazon reported about 1.58 million at the end of 2025. Both are larger payrolls than GM ever had.
- OpenAI is not a bigger slice of current output than GM was. GM's 1978 sales were equal to about 2.7 percent of that year's GDP. OpenAI's own figure is $2 billion of revenue a month, about $24 billion a year, under 0.1 percent of current GDP.
- There are not fewer public companies than in the GM peak year. World Bank listed-company counts: about 2,401 in 1979, a peak of 8,090 in 1996, and 3,908 in 2025. The drop is from the 1990s, not from 1979.
- Index funds and 401(k) plans do hold Apple, Microsoft, and Nvidia. The holding is real. It is also very uneven. OpenAI is still private, so a normal S&P 500 fund does not own it.

## GM at the peak

| Fact | Number | Source |
| --- | --- | --- |
| US employment, 1979 | 618,365 | Company history as repeated by AP timelines (CBS, San Diego Union-Tribune, mlive). Largest private employer in the country that year. |
| Of which hourly | 511,000 | Transport Topics, citing the same peak. |
| Worldwide employment, 1979 | 853,000 | Same timelines. |
| 1978 sales | $63.221 billion | Fortune 500 list published in 1979. Washington Post, 19 April 1979: $63.2 billion, first on the list, ahead of Exxon. |
| 1978 profit | $3.508 billion | Same Fortune list. |
| Share of US cars | More than half | New York Times, 7 February 1978, on 1977 results. |
| US civilian employment, 1979 | 98.824 million | Economic Report of the President, 2006, table B-37. |
| GM's direct share of US employment | 0.63 percent, about 1 in 160 employed people | 618,365 / 98,824,000. |
| US GDP, 1978 | about $2.35 trillion | Average of the four 1978 quarterly GDP readings on FRED series GDP (BEA). |
| Sales relative to GDP | about 2.7 percent | $63.2 billion / $2.35 trillion. Sales are not value added. Purchased parts are inside the sales number, so this overstates GM's share of production. |
| Sales per worldwide employee | about $74,000 | $63.221 billion / 853,000. Years are adjacent (sales 1978, headcount peak 1979). |
| Same figure in 2025 dollars | about $370,000 | CPI-U annual average 65.2 in 1978 and about 322 in 2025 (Minneapolis Fed CPI table; ratio about 4.9). |
| Profit per worldwide employee | about $4,100 then, about $20,000 in 2025 dollars | $3.508 billion / 853,000, then the same CPI ratio. |
| GM employment, 31 Dec 2025 | 156,000 | Company filings, as compiled by Stock Analysis. |
| GM market value, 2026 | on the order of $75–80 billion | Public quotes in 2026 compilations. About $0.5 million of market value per employee. |

Around GM, the industry was also a mass employer. BLS: in 1979, 463,000 people worked in motor-vehicle assembly and 441,100 in parts. GM was the largest piece of that, not the whole of it.

A 2020 Center for Automotive Research study of a much smaller GM (about 84,000 US jobs) estimated 8.1 total US jobs per direct GM job, counting suppliers and the spending of those paychecks. That multiplier is for recent operations. It is not a 1979 census count.

## How pay was shared then

BLS, in its ERISA-at-50 review: in 1979, 87 percent of full-time workers in medium and large private firms participated in a retirement plan. The typical plan was a defined-benefit pension. The employer owed a lifetime payment. The worker did not have to pick stocks.

BLS National Compensation Survey, data extracted 22 September 2026: 14 percent of private-industry workers had access to a defined-benefit plan in 2025. Access was 15 percent in 2024 and 2023.

An American Academy of Actuaries review of ERISA at 50: active participation in PBGC-insured single-employer defined-benefit plans fell from 29 percent of private wage-and-salary workers in 1980 to 7 percent in 2020. "Access" and "still accruing a benefit" are different measures. Both fell a long way.

GM's chairman paid himself under $1 million in 1977 ($975,000, New York Times, 15 April 1978) in a year the company earned $3.3 billion.

## OpenAI now

All of these are company statements from OpenAI's 31 March 2026 funding note, unless marked.

| Fact | Number |
| --- | --- |
| Post-money valuation | $852 billion |
| New capital committed | $122 billion |
| Revenue | $2 billion per month |
| Weekly ChatGPT users | more than 900 million |
| Paying subscribers | more than 50 million |
| Enterprise share of revenue | more than 40 percent |
| Profit | The company is not describing itself as profitable. The valuation is a price on future profit. A secondary account of the same round says heavy spending on chips, data centers, and hiring keeps it unprofitable. |

Headcount is not in that note. Reuters, 21 March 2026, citing the Financial Times: about 4,500 people, with a plan to reach about 8,000 by the end of 2026. Third-party scrapers publish higher counts. Those are not company figures.

Value per person at the reported 4,500: $852 billion / 4,500 = about $189 million. At the planned 8,000: about $107 million. This is a private valuation per employee, not pay, and not profit.

Revenue per person at 4,500 and a $24 billion annual run rate: about $5.3 million. That happens to match Nvidia's revenue per employee. The valuation does not.

### Who holds it

Official, from the 2025 recapitalization, still the last full split the company published: OpenAI Foundation 26 percent, Microsoft about 27 percent, and the other 47 percent with employees and investors. The 2026 round did not come with a new official table, so those shares have been diluted by some amount that is not public.

The March 2026 round was anchored by Amazon, Nvidia, and SoftBank, with Microsoft continuing. Other names in the company note: a16z, D. E. Shaw, MGX, TPG, T. Rowe Price, Altimeter, Appaloosa, ARK, BlackRock-affiliated funds, Blackstone, Coatue, Fidelity, Sequoia, Thrive, Temasek, UC Investments, and others.

New retail access is small next to the round: more than $3 billion from individuals through banks, out of $122 billion committed. ARK said some of its ETFs will hold OpenAI. A household S&P 500 fund still does not.

A reconstructed cap table circulated in April 2026 (Forbes, drawing on a table posted by Sheel Mohnot). It is not a filing. It estimates Microsoft at 26.8 percent (about $228 billion on $13 billion invested), the Foundation at 25.8 percent, SoftBank at 11.7 percent, Amazon at 4.7 percent, Nvidia at 3.5 percent, and about 165 current and former employees at 19.4 percent, about $165 billion. Use it only with that label.

Reach is broad. Ownership is a list of funds, three or four strategic companies, a foundation, and a few hundred employees with meaningful equity.

## The firms that now hold the market value

Employee counts are from Form 10-K. Market values are 2026 ranking figures (AlphaSense and similar) and move every day. Treat value-per-employee as the current order of magnitude.

| Company | People | Market value, 2026 rankings | Value per person |
| --- | --- | --- | --- |
| Nvidia | 42,000 (fiscal 2026, year ended 25 Jan 2026) | about $5.1 trillion | about $120 million |
| Alphabet | 190,820 (31 Dec 2025) | about $4.6 trillion | about $24 million |
| Apple | 166,000 (27 Sep 2025) | about $4.5 trillion | about $27 million |
| Microsoft | 223,000 (30 Jun 2026) | about $3.7 trillion | about $17 million |
| Amazon | about 1,576,000 (31 Dec 2025) | about $3.0 trillion | about $1.9 million |
| GM, 2025 | 156,000 | about $75–80 billion | about $0.5 million |
| GM, 1979 headcount with no comparable market-value pair in these sources | 853,000 worldwide | — | — |
| OpenAI | about 4,500 (Mar 2026 report) | $852 billion private valuation | about $189 million |

Nvidia fiscal 2026 revenue was $215.9 billion. Revenue per employee was $5.14 million, about 14 times GM's inflation-adjusted 1978 sales per worker.

Deutsche Bank strategist Jim Reid, as reported by Investopedia: when Nvidia was worth about $3.5 trillion with 36,000 people, that was above $90 million per employee, against roughly $18 million at Apple and $15 million at Microsoft at that time. His historical note matches the employment record used here. GM, the largest American company in the 1950s, employed about 600,000. Kodak later passed GM in market value with about one-sixth the workforce. GE in the 1970s employed about 400,000.

## Market value piled into a few names

- Magnificent Seven (Apple, Microsoft, Nvidia, Alphabet, Amazon, Meta, Tesla): 33.5 percent of S&P 500 market value on 10 September 2026 (History of Market live weights). Motley Fool: 33.9 percent in September 2026. MacroMicro: 34.03 percent in July 2026, combined value $22.75 trillion.
- That combined value is on the order of 70 percent of one year of US GDP. BEA annual GDP was $30.76 trillion in 2025. Q2 2026 GDP was a $32.5 trillion annual rate.
- RBC: the 10 largest S&P 500 companies were about 19 percent of the index at the end of 1990 and 40.7 percent at the end of 2025.
- Visual Capitalist, from a long index reconstruction: the top 10 were about 19 percent of S&P 500 value in 2010 and 38 percent in 2024. From 1880 to 2010 the typical reading was about a quarter.
- CRSP, 30 September 2025: top 10 names were 36.5 percent of the whole US market, versus 17.3 percent in January 1980. Top 30 names were 50.4 percent, versus 28.5 percent in 1980. The next 1,000 companies after the top 500 fell from 16.7 percent of market value to 7.7 percent.

## Who owns the equity

Federal Reserve Distributional Financial Accounts, Q1 2026. Category: corporate equities and mutual fund shares held by households. This category does not include stock inside 401(k) and similar defined-contribution accounts. Those are a separate line.

| Wealth group | Share of directly held equities and mutual funds |
| --- | --- |
| Top 0.1 percent | 24.2 percent |
| Top 1 percent | 50.2 percent |
| Next 9 percent (90th to 99th) | 37.2 percent |
| Top 10 percent combined | 87.4 percent |
| 50th to 90th percentile | 11.6 percent |
| Bottom 50 percent | 1.1 percent |

Dollar levels on the Fed's Q1 2026 chart for that same category: top 0.1 percent $13.3 trillion, next 0.9 percent $14.3 trillion, next 9 percent $20.5 trillion, middle 40 percent $6.4 trillion, bottom 50 percent $0.6 trillion.

Defined-contribution pensions are less concentrated, and the middle holds a real piece. Same chart, Q1 2026: middle 40 percent $6.3 trillion of a roughly $15.9 trillion total; bottom 50 percent $0.75 trillion; top 10 percent a bit over half.

## Labor's share of business output

BLS Productivity and Costs, Q2 2026: labor's share of nonfarm business output was 52.8 percent. BLS says that is the lowest reading since the series began in the first quarter of 1947.

The same release: unit profits of nonfinancial corporations rose at a 43 percent annual rate in the quarter, and 17.8 percent over the prior four quarters.

Autor, Dorn, Katz, Patterson, and Van Reenen, Quarterly Journal of Economics, 2020: the labor-share decline lines up with sales moving toward "superstar" firms. Those firms have high markups and a low labor share. The fall is mostly reallocation toward those firms, not a cut in the labor share at a typical firm.

## Global scale, and why "early" is the stark part

Checked 24 September 2026, from public research and company statements. Exposure is not the same thing as a firing.

### The planet versus a few payrolls

| Fact | Number | Source |
| --- | --- | --- |
| World output, 2025 | $118.2 trillion | IMF World Economic Outlook, April 2026, market exchange rates. |
| World output, 2026 projection | $126.3 trillion | Same. |
| Magnificent Seven market value, July 2026 | $22.75 trillion | MacroMicro. About one-fifth of one year of world output. |
| People employed worldwide, 2025 | about 3.7 billion | Secondary compilation of ILO and World Bank data (Statbase, May 2026). |
| OpenAI weekly users | more than 900 million | OpenAI, 31 March 2026. On the order of 1 in 9 people. |
| Users per OpenAI employee | about 200,000 | 900 million / about 4,500. |
| Employed people on earth per OpenAI employee | about 800,000 | 3.7 billion / 4,500. |

A model does not stop at a border. Once it can do a task, it can do that task in every country in the same week. A GM plant hired a city. The company that owns the model does not.

### The millions-of-people model is still here. It is the target.

| Payroll | People | What they sell |
| --- | --- | --- |
| Foxconn / Hon Hai | 610,000 in the 2025 sustainability report; the company says about 900,000 at peak manufacturing season | Electronics assembly. 2025 revenue about $260 billion. Net margin 2.3 percent. This is the GM shape, moved to Asia. |
| India technology industry | 5.8 million in FY2025 | Nasscom. Revenue about $283 billion, exports about $224 billion. Code, back office, engineering. |
| Philippines IT and business-process work | 1.9 million in 2025 | Industry reports via IBPAP. Customer service and back office. Target about 2.0 million in 2026. |

Those three payrolls are about 8 million people. OpenAI's reported headcount is about 4,500. The work those millions sell — code, customer support, documents, and, next, physical assembly — is the work agents and robots are being built to do.

### What is already visible

Stanford Digital Economy Lab and ADP, revised August 2026, data through June 2026 (Brynjolfsson, Chandar, Chen). No economy-wide wave of firings. Employment of workers ages 22 to 25 in the most AI-exposed occupations is about 19 percent below where it would be if it had kept pace with workers the same age in less-exposed occupations. In July 2025 that gap was 15 percent. The change comes from not hiring them, not from laying them off. It shows up where AI use replaces tasks, and it does not show up where AI use assists a person. The authors call these canaries, not a finished result.

ILO working paper, "Disruption without Dividend?": office jobs are a large share of the good formal jobs in poorer countries, and they are exposed. A study of a large online labor marketplace (Betai and Chen, 2025), cited there, finds generative AI lined up with a sharp drop in the volume and value of service work sent from rich countries to poorer ones, especially for skills that are easier to automate. A small group moved up to harder tasks. Most did not.

Anthropic Economic Index, March 2026, on its own usage: for about 49 percent of job types, at least a quarter of that job's tasks have already appeared as something Claude is doing. That is a map of tasks in their sample, not a count of workers removed. On the consumer chat product, use is still split: about 52 percent helping a person, 45 percent doing the task (November 2025). On the API, the way a company wires the model into a product, automated use has been about three-quarters. Coding is moving out of the chat window and into those automated pipelines. Anthropic says that move is the signal that the related jobs are closer to changing.

Usage is uneven. The top 20 countries account for about 48 percent of Claude use per person. The tool spreads first where the buyers are. The lost contract shows up where the work used to be sent.

### What the forward estimates actually say

These are not forecasts of a single unemployment rate. They answer different questions.

| Who | What they measured | The number |
| --- | --- | --- |
| ILO, 2025 global index | Share of workers in a job with some generative-AI exposure | 1 in 4 worldwide. 3.3 percent in the highest band. 34 percent of jobs in high-income countries, 11 percent in low-income countries. Women are more exposed than men. The ILO's own reading: most jobs will change, because a job is a bundle of tasks and some tasks stay human. |
| Goldman Sachs | Jobs exposed, and US hours | About 300 million jobs worldwide exposed. In the US, tasks equal to about 25 percent of work hours. Their 2026 note says the labor story of the year is AI, and that the first pressure is on people in their 20s and 30s entering knowledge and content work. |
| McKinsey, 2025, "Agents, robots, and us" | Technical potential of tools that already exist, US only | Agents could cover tasks that take 44 percent of US work hours. Robots, 13 percent. Together, 57 percent. Their midpoint for what is actually adopted by 2030 is about 27 percent of hours. They say this is not a prediction of job losses. |
| World Economic Forum, employer survey, 2025 | How employers think tasks will be split by 2030 | Now: 47 percent of tasks mainly a human, 22 percent mainly a machine, 30 percent both. By 2030 they expect those three shares to be close to even. Most of the drop in human-only work is automation. |

### What the people building it say

The scare lines and the walk-backs are both public. A graphic that uses only the scare lines is picking a side the speakers themselves have blurred.

- Dario Amodei, Anthropic, 2025: AI could wipe out half of entry-level white-collar jobs within one to five years, and unemployment could go to 10–20 percent. In 2026 he said he is "still the same order of concern." He also offered the other story: automate 90 percent of a job and the remaining 10 percent expands until people are doing ten times as much. Both sentences are his.
- Mustafa Suleyman, Microsoft AI: most professional tasks done sitting at a computer — accounting, law, marketing, project management — automated within 12 to 18 months. Human-level performance on most professional tasks.
- Jim Farley, Ford: AI cuts the number of white-collar jobs in the US in half.
- Sam Altman, earlier: AI will probably replace most of the jobs people do today, and some categories will be "totally gone." May 2026, at a Commonwealth Bank of Australia event: he was "delighted to be wrong." He does not expect the jobs apocalypse some companies in the field talk about. Entry-level white-collar cuts have been smaller than he expected. Fortune noted the softer language from both Altman and Amodei as their companies moved toward public markets.
- Jensen Huang, Nvidia, has called the AI-layoffs story nonsense. Yann LeCun has said Amodei does not know the labor record of past technologies.
- Daron Acemoglu, MIT: a careful task model puts the extra productivity over ten years at roughly half a percent to about 1 percent of output. He does expect the gap between capital income and labor income to widen.

Challenger, Gray & Christmas: through April 2026, employers cited AI for on the order of 50,000 announced job cuts. That is real and small next to a workforce of hundreds of millions. The large number in the data is the missing hire, not the announced layoff.

### The sentence that is fair

The firing of everyone has not happened. The mechanism that makes a person feel replaced has started, and it is global. A young worker in an exposed job is not hired. A service contract that used to go to Manila or Bengaluru is not renewed. The model that did the task is owned by a company with a few thousand people, and the same model works everywhere. The tools in use now are the early ones. ChatGPT is under four years old. Studies of "exposure" were mostly written about a chatbot. The product now shipping is an agent that finishes the task.

## A clean set of sentences for the graphic

1. In 1979, GM alone employed 618,365 people in the United States, about 1 in 160 employed Americans, and 511,000 of them were hourly.
2. In 1978 that company sold $63 billion, equal to about 2.7 percent of US GDP, and earned $3.5 billion. Sales per worker were about $74,000, roughly $370,000 in 2025 prices.
3. In 1979, 87 percent of full-time workers at medium and large private firms were in a retirement plan, usually a pension the company owed.
4. In 2025, 14 percent of private-industry workers had access to that kind of plan.
5. Nvidia, worth about $5 trillion, employs 42,000 people. That is about $120 million of market value per person, and $5.1 million of revenue per person.
6. OpenAI, valued at $852 billion in March 2026, had on the order of 4,500 people. That is about $190 million of private value per person. It reports $2 billion of revenue a month and more than 900 million weekly users.
7. Seven companies are about one-third of the S&P 500. Ten companies are about 41 percent. In 1990 the top ten were about 19 percent.
8. The top 10 percent of households hold about 87 percent of directly owned stocks and mutual funds. The bottom half hold about 1 percent of that category.
9. Labor's share of nonfarm business output is 52.8 percent, the lowest since 1947.
10. The largest payrolls today are bigger than GM's, not smaller. Walmart's is about 2.1 million people. The new fact is how little of the new value those payrolls, or any broad payroll, receive.
11. About 3.7 billion people are employed on earth. OpenAI's reported staff is about 4,500, and more than 900 million people use ChatGPT in a week. That is about 200,000 users per employee.
12. Seven companies are worth about $23 trillion, roughly one-fifth of one year of world output.
13. Foxconn still employs on the order of 600,000 to 900,000 people to assemble electronics. India tech employs 5.8 million. The Philippines' outsourced-service industry employs 1.9 million. That is the global GM. It is also the work agents are being pointed at.
14. The early evidence is a closed door, not a mass layoff. US workers ages 22 to 25 in the most exposed jobs are about 19 percent below their less-exposed peers. Service work sent to poorer countries has already dropped where the task is easy to automate.
15. Tools that exist now could, in theory, cover 57 percent of US work hours. The sober midpoint for 2030 is about 27 percent of hours actually automated. One in four workers worldwide is in a job with some exposure. The highest-risk band is 3.3 percent. The people building the models have said both "half of entry-level white-collar jobs" and "I was wrong." The graphic should show the door closing while the tools are still early, and it should not print a single unemployment forecast as fact.

## If the ladder stays up: control versus cash

This is a scenario, not a forecast. The assumption is the one in the question: a few firms keep the compute and the frontier models, users cannot switch to a cheap substitute, and the pay for automated tasks is not handed back to workers or competed away into lower prices.

Two different questions get two different sizes.

Control means how much activity has to pass through them. Cash means how many dollars they keep. Card networks already show the split. A huge share of payments passes through a few firms. The fee they keep is small. The ladder-up case is the same gateway with a fee the size of a wage.

### Where the money is now

Latest reported full years. Fiscal years do not line up. Amazon’s figure is the whole company, stores included, not the cloud alone.

| Firm | Revenue | Operating profit |
| --- | --- | --- |
| Microsoft, year to June 2025 | $282 billion | $129 billion |
| Alphabet, calendar 2025 | $403 billion | $129 billion |
| Meta, calendar 2025 | $201 billion | $83 billion |
| Nvidia, year to 25 January 2026 | $216 billion | $130 billion |
| Amazon, calendar 2025 | $717 billion | $80 billion |

Microsoft, Alphabet, Meta, and Nvidia together earn about $470 billion of operating profit. Adding Amazon brings the five to about $550 billion. World output in 2025 was $118 trillion (IMF, market exchange rates). So the profit is about 0.4 to 0.5 percent of one year of world output.

Their stock prices are a different object. The seven largest tech firms were worth about $23 trillion in mid-2026, about a fifth of one year of world output. That price is a claim on future profit. It is not profit they have collected.

The gateway is already narrow:

- Synergy Research, second quarter 2026: Amazon, Microsoft, and Google took 28, 20, and 15 percent of cloud infrastructure spending. Together, 63 percent of a $143 billion quarter.
- Nvidia still has on the order of 70 to 80 percent of AI-accelerator revenue. Trade estimates, not an official census. The custom chips taking the rest are mostly Google, Amazon, Microsoft, and Meta. Leaving Nvidia means staying inside the same club.
- A Goldman asset-management note in 2026 said about 90 percent of today’s AI profit pool sits in chips, memory, and manufacturing, and that AI-related capital spending in 2026 was expected to pass $750 billion. OpenAI’s own revenue is about $2 billion a month, and the company is not describing itself as profitable. The cash is at the shovel layer.

Adding these profits does not double-count. Each firm’s operating profit is what is left after it pays its suppliers. Adding their sales would double-count, because the cloud firms buy the chips.

### The pool that can be moved

Global labor income is 52.4 percent of world output (ILO, 2024). Applied to the IMF’s $118 trillion, that is on the order of $60 trillion a year paid for work. That product is an order of magnitude, not an ILO dollar release.

Not all of it can move. Farms, most care, most construction, and most in-person service stay put. The ILO’s own exposure index puts some generative-AI exposure on 1 in 4 workers worldwide, and on 11 percent of jobs in low-income countries. The dollar wages sit in the rich countries, and those jobs are more exposed (34 percent of high-income employment).

Published sizes for the movable piece:

- McKinsey, 2025: tools that exist could in theory cover 57 percent of US work hours (agents 44, robots 13). Their midpoint for hours actually automated by 2030 is about 27 percent. They value that midpoint at about $2.9 trillion a year in the United States, and they say it is not a job-loss forecast.
- Goldman Sachs, 2023: widespread adoption could raise yearly world output by about 7 percent, almost $7 trillion in the prices of that note. On the IMF’s 2026 world output of $126 trillion, 7 percent is about $9 trillion a year. That figure is new production. It is not, by itself, a profit for the model owners. Goldman later said the split among chip firms, model firms, the companies that use the tools, workers, and customers is not known.
- Acemoglu’s task model is the small case: extra output over ten years of roughly half a percent to about 1 percent. On today’s world economy that is under $2 trillion a year, and he still expects the gap between capital income and labor income to widen.

### Ladder-up ranges

If the club keeps about half the value of the work that moves, and sends the rest to the firms that deploy the tools or to customers:

- Tied to McKinsey’s US midpoint, plus a partial follow-on in other rich countries: on the order of $4 to $7 trillion of work value a year worldwide, and about $2 to $3.5 trillion a year kept by the club.
- If the club keeps most of that work value, because users cannot leave and workers are not paid for the task: about $4 to $7 trillion a year.
- If Goldman’s extra output is also captured as profit rather than passed on: the club’s claim can sit in the same $4 to $9 trillion band. These ranges overlap. They are not things to add together.

The IMF’s 2031 world output, at market exchange rates, is $158 trillion. Against that:

- $2 to $3.5 trillion of profit is about 1.5 to 2 percent of world output.
- $4 to $7 trillion is about 3 to 4.5 percent.
- A ceiling, if something close to the technical maximum of rich-country computer work is both automated and captured, is on the order of $10 trillion a year, about 6 percent of that future world output. That ceiling needs three things at once: the tools get used on most of the hours they could do, no cheap substitute appears, and new jobs do not absorb the pay.

For scale: the club’s operating profit now is about $0.5 trillion. The middle of the ladder-up range is several times that. It is still a few percent of the economy, not most of it.

### What “under their control” can mean without owning the GDP

Most computer-based work in rich countries, and the export office work in India and the Philippines, would have to run on a few models and a few clouds. That can be a large share of the world’s dollar wages and a much smaller share of the world’s workers. It does not put a farm, a hospital ward, or a mine inside the club.

The cash number stays far below the control number unless the fee is a wage. If models become cheap and easy to swap, the gateway can stay in a few hands and the fee can fall toward the cost of electricity and chips. That is the outcome that shrinks the projection. The custom-chip programs at Google, Amazon, Microsoft, and Meta do not, by themselves, shrink it. They move the gateway from Nvidia into the same small club.