# The economics of AI and labor power

Research brief for the Exploit Conference keynote (Montreal, 28 September 2026).
Compiled 12 September 2026. All figures carry a source and a date. Check the market numbers (valuations, market caps, TAO price) on the day of the talk. They move.

Claim under test: **As AI replaces labor, humans lose their historic source of economic and political power (labor power). Value and capital concentrate in a small number of AI companies with very few employees.**

Key terms used below:
- **Labor share**: the slice of national income paid to workers as wages and benefits. The rest goes to capital (owners of machines, buildings, patents, shares).
- **Capital**: assets that produce income. Factories, data centers, chips, software, company shares.
- **TFP** (total factor productivity): output you get after you account for the labor and capital you used. A measure of "how smart" the economy is.
- **Run-rate revenue**: the latest month's revenue multiplied by 12.
- **Market cap** (market capitalization): share price multiplied by number of shares. What the market says a company is worth.
- **Capex** (capital expenditure): money spent to buy long-lived assets, such as data centers and chips.
- **VC** (venture capital): money invested in private start-ups.

---

## Executive summary (10 lines)

1. The US labor share hit **52.9% in Q2 2026**, the lowest since records began in 1947. It was 65.8% in 1947 and 62.8% in 2000. (BLS, 2026)
2. The global labor income share fell to **52.3%** by 2024, a shortfall of **$2.4 trillion a year** versus 2004. The ILO links the fall to automation. (ILO, Sep 2024)
3. Frontier theory agrees on direction: if AI can do all tasks, the labor share falls toward **zero** and wages **collapse** even while output grows more than ten-fold. (Trammell & Korinek 2023–26; Korinek & Suh 2024)
4. Acemoglu and Restrepo show automation already explains **50–70%** of US wage-structure change from 1980 to 2016, with only a **3.4%** cumulative TFP gain. Big inequality, small growth. (Econometrica, 2022)
5. Early evidence is arriving: employment of 22–25-year-olds in AI-exposed jobs is **19% below** its peers' path as of June 2026, up from 13% in 2025. (Brynjolfsson, Chandar & Chen, Stanford, Aug 2026)
6. Value per head is unprecedented: Anthropic is valued at **$965B** with roughly **2,300–5,200** staff; OpenAI at **$852B** with roughly **4,500–7,800**; NVIDIA earned **$215.9B** with **42,000** people. GM at its 1979 peak employed **853,000**. Walmart employs **2.1 million**.
7. Capital is concentrating: four hyperscalers guide **$720–745B** of 2026 capex. AI took **52.7%** of all global VC dollars in 2025. The Magnificent Seven hit a record **35%** of the S&P 500 in June 2026.
8. Who owns that capital: the top 1% of US households hold **49.9%** of corporate equities; the top 10% hold about **87%**. The bottom 50% hold **2.5%** of net worth. (Federal Reserve, Q2 2025)
9. The historical analogy is strong: during Engels' pause (1780–1840) British output per worker rose **46%** while real wages rose **12%** and the profit rate doubled. Wages caught up only after workers organized. (Allen 2009; Acemoglu & Johnson 2023, 2024)
10. Honest weak points: economy-wide job data show **no disruption yet** (Yale Budget Lab, Oct 2025; Humlum & Vestergaard 2025–26). Acemoglu himself forecasts only **0.53–0.66%** TFP gain from AI over ten years. The "AI takes 100% of capital" claim is a scenario, not a measurement.

---

## 1. Labor share, capital share, and "labor as leverage"

### The data

| Measure | Value | Date | Source |
|---|---|---|---|
| US nonfarm business labor share | 65.8% | Q1 1947 | BLS (2017 TED article) |
| US nonfarm business labor share | 62.8% | Q4 2000 | BLS (2017 TED article) |
| US nonfarm business labor share | 56.0% (then-record low) | Q4 2011 | BLS (2017 TED article) |
| US nonfarm business labor share | 54.4% | Q4 2025 | BLS Productivity and Costs, revised 24 Mar 2026 |
| US nonfarm business labor share | 53.7% (record low) | Q1 2026 | BLS Productivity and Costs, Q1 2026 |
| US nonfarm business labor share | **52.9% (record low)** | **Q2 2026** | BLS Productivity and Costs, preliminary, Q2 2026 |
| Global labor income share | 52.3% (down 1.6 pts from 2004) | 2022–2024 | ILO WESO Update, 4 Sep 2024 |
| Annual global labor income shortfall vs. 2004 share | $2.4 trillion (PPP) | 2024 | ILO WESO Update, 4 Sep 2024 |

The ILO also ran a 36-country study (2003–2019) and found that technology shocks raised productivity but cut the labor share. It says this pattern is "consistent with automation-based technological innovations." It warns generative AI could push the share down further if the pattern holds. (ILO, Sep 2024)

Note: BLS data for Q1 and Q2 2026 are preliminary and will be revised. The direction is not in doubt. The series has been falling for 25 years and is now at its lowest recorded point.

### Why the labor share is falling: the superstar-firm mechanism

Autor, Dorn, Katz, Patterson and Van Reenen (QJE, 2020) used US Economic Census micro-data from 1982 onward. Their finding: sales concentrate in a few "superstar firms" in each industry. These firms have high markups (price well above cost) and a low labor share of value added. When output shifts to them, the economy-wide labor share falls, even if the typical firm's labor share does not. They confirmed seven predictions of this model, in the US and abroad.

This matters for the talk. It shows the labor-share decline is not mainly about every firm paying workers less. It is about **reallocation toward firms that need very little labor**. AI labs are the extreme case of that firm type.

### The "labor as leverage" argument

Why did labor give ordinary people power? Three channels, each with a named source.

**1. Concentration of workers created political organization.** Acemoglu and Johnson (*Power and Progress*, 2023) quote the 1790s radical John Thelwall: "every large workshop and manufactory is a sort of political society, which no act of parliament can silence, and no magistrate disperse." Their thesis: shared prosperity "emerged because, and only when" workers organized and forced a different direction of technology and a different split of the gains. Electoral competition, trade unions, and worker-protection laws changed how wages were set in 19th-century Britain. Their line: "Today we need to do the same again."

**2. Labor is the state's main tax base.** Korinek and Juelfs (Brookings / NBER, 2022) note that governments rely mostly on taxing labor. If automation makes labor redundant, "governments will no longer be able to raise funds from taxes on labor." Korinek and Lockwood (NBER, 2026) extend this: transformative AI "may gradually erode the two main tax bases that underpin modern tax systems: labor income and human consumption." A state that no longer needs its citizens' labor or their taxes has less reason to answer to them.

**3. Low wages benefit capital owners, which shapes politics.** Korinek and Juelfs (2022) show a multiplier: when benefits are cut, people must work more, labor supply rises, wages fall further. They write that "greater supply of labor and lower wages benefit the owners of capital," which "may explain part of the political appeal to capitalists of cutting social benefits." The political fight over the labor share is not new; AI raises the stakes.

Acemoglu and Johnson (NBER, 2024, "Learning from Ricardo and Thompson") add the mechanism from E.P. Thompson: automation in early industrial Britain "forced workers into unhealthy factories with close surveillance and little autonomy." Their conclusion: "Wages are unlikely to rise when workers cannot push for their share of productivity growth." If AI is used for surveillance and worker control, it "will shift the balance of power between workers and managers."

---

## 2. Automation vs. augmentation: the economics

Two words to define. **Automation**: a machine does a task a person used to do. The person's labor is no longer needed for that task. **Augmentation**: a machine helps a person do a task better or faster, or creates new tasks for people. Augmentation can raise wages. Automation without new tasks tends to lower them.

### Acemoglu and Restrepo (task-based model)

- *Tasks, Automation, and the Rise in US Wage Inequality* (Econometrica, 2022): **50–70%** of changes in the US wage structure from 1980 to 2016 are explained by task displacement from automation. Automation cut real wages of male high-school dropouts by 8.8%. Automation contributed only a **3.4%** cumulative TFP gain over 36 years. Their summary: "major changes in wage inequality can go hand-in-hand with modest productivity gains."
- The key idea: automation has a **displacement effect** (bad for labor) and a **productivity effect** (good for everyone). Whether labor wins depends on whether new tasks are created for people. "So-so automation" (machines slightly better than people) displaces workers without much productivity gain.

### Acemoglu, *The Simple Macroeconomics of AI* (NBER, May 2024)

- Forecast: AI raises US TFP by **no more than 0.66%** over ten years. Adjusted for hard-to-learn tasks: **less than 0.53%**.
- But: "AI is predicted to widen the gap between capital and labor income." No evidence AI will reduce labor income inequality.
- Also warns some new AI tasks may have "negative social value (such as design of algorithms for online manipulation)."
- Use with care. This paper is the strongest **counter** to "AI will transform the economy." It is also strong **support** for "the gains will go to capital." Both parts are true in his model.

### Acemoglu and Johnson, *Power and Progress* (2023)

- Thesis: a thousand years of history show that technology raises living standards only when society forces a broad split of the gains. The Middle Ages gave cathedrals, not fed peasants. The first hundred years of industrialization gave stagnant wages.
- On AI: "The current path of AI is neither good for the economy nor for democracy, and these two problems, unfortunately, reinforce each other."
- Policy: build "countervailing powers," change the narrative from "machine intelligence" to "machine usefulness," and redirect technology to augment workers.
- Simon Johnson (MIT News, May 2023): "Many algorithms are being designed to try to replace humans as much as possible. We think that's entirely wrong."

### Autor, *Applying AI to Rebuild Middle Class Jobs* (NBER, Feb 2024)

- The optimist's case. Autor argues AI could extend expertise: let nurses do some of what doctors do, paralegals some of what lawyers do, junior coders some of what senior engineers do. This could rebuild the middle class that automation hollowed out.
- His own caveat: "My thesis is not a forecast but an argument about what is possible."
- Use this to show that the good path exists, but is a **choice**, not a default.

### Brynjolfsson, Chandar and Chen, *Canaries in the Coal Mine* (Stanford Digital Economy Lab)

The best early evidence, from ADP payroll data on millions of US workers.

- August 2025 version: workers aged 22–25 in the most AI-exposed jobs saw a **13%** relative employment decline since late 2022, controlling for firm-level shocks.
- November 2025 version: **16%**.
- August 2026 update (data through June 2026): **19%** below where it would be if it had kept pace with less-exposed peers. The gap "has widened steadily." It works through **reduced hiring**, not layoffs. It is concentrated where AI **substitutes** for tasks; where AI complements workers, employment is flat or rising.
- Their own caution: these are "early, descriptive indicators... rather than causal estimates." No economy-wide displacement yet.
- Occupations named: software developers, customer service representatives.

### Korinek and Suh, *Scenarios for the Transition to AGI* (NBER, March 2024)

- Model: work is a set of tasks of varying complexity. AI automates ever more complex tasks. Wages depend on a "race between automation and capital accumulation."
- Result: "if the complexity of tasks that humans can perform is bounded and full automation is reached, then wages collapse." This holds "no matter what savings behavior" the economy pursues.
- Baseline AGI scenario (full automation in 20 years): wages rise a little, then "collapse before full automation is reached." After the collapse, wages equal the return on capital, and the economy grows at **18% a year**.
- Aggressive AGI scenario (5 years): wage collapse "after about 3 years."
- Both AGI scenarios produce **more than ten-fold** output growth alongside collapsing wages. (Korinek, NBER Reporter, 2024)
- One hopeful scenario: a "bout of automation" followed by a long tail of hard tasks. Wages collapse, then recover around year 9 once capital catches up.

### Trammell and Korinek, *Economic Growth under Transformative AI* (NBER 2023, revised April 2026; Annual Review of Economics, June 2026)

- The **Kaldor facts**: for a century, rich economies have shown steady per-capita growth and a stable labor share. Their claim: transformative AI breaks both.
- "Fully automating production alone (so that machines can self-replicate) would dramatically raise the growth rate and lower the labor share."
- Direct quote: "the burst in output made feasible by a qualitative advance in automation will almost all accrue to capital, driving the labor share toward zero. It will probably be very inefficient (and so very difficult) to steer technology in a labor-augmenting direction permanently."
- Wages may rise or fall in absolute terms. The labor **share** falls either way.

### Aghion et al., *How Different Uses of AI Shape Labor Demand: Evidence from France* (AEA Papers & Proceedings, May 2025)

- French firm data 2017–2020. Firms that adopt AI grow employment and sales. The productivity effect beats displacement **within the firm**.
- But: "larger and more productive firms should be the great winners of the AI revolution." And: "the main risk for workers is likely to be displacement by workers at other firms using AI." They call for policy to prevent "increased market concentration and entrenched market power."
- Use this as an honest complication: firm-level studies often show AI adds jobs at adopting firms. The concentration risk is at the market level.

### Susskind, *A World Without Work* (2020) and *Growth: A Reckoning* (2024)

- Susskind argues "this time really is different" because AI reaches non-routine cognitive tasks.
- His answer is a "Big State" focused on distribution, not production, funded by taxing capital and "superstar firms that generate healthy profits while employing proportionally fewer workers." He proposes a **conditional basic income** tied to socially valuable non-market activity.
- Relevant as a foil: Susskind's answer runs through the nation-state. Bittensor's does not.

### IMF, *Gen-AI: Artificial Intelligence and the Future of Work* (Jan 2024)

- **40%** of global employment is exposed to AI. **60%** in advanced economies. About half of exposed jobs in advanced economies are at risk of lower demand; the other half may benefit.
- "Capital returns will increase wealth inequality."

### 2025–2026 work

- Brynjolfsson, Chandar & Chen, Aug 2026 update (above).
- Korinek & Lockwood, *Public Finance in the Age of AI: A Primer* (NBER 34873, 2026): if AI displaces labor, consumption taxes may become the main revenue tool; if AGI absorbs most resources, even that fails. They evaluate robot taxes, compute taxes, token taxes, sovereign wealth funds, windfall clauses, and "Universal Basic Capital."
- Humlum & Vestergaard, *Large Language Models, Small Labor Market Effects* (NBER 33777, revised March 2026): Danish data, precise null effects on earnings and hours (see section 6).
- Trammell & Korinek, Annual Review version (June 2026).

---

## 3. Superstar firms and winner-take-all: value per employee

### The AI companies

| Company | Valuation or market cap | Date | Employees | Date | Revenue | Date |
|---|---|---|---|---|---|---|
| **Anthropic** | $965B post-money (Series H, $65B raised) | 28 May 2026 | ~2,300 (end 2025, Fortune); ~3,830 (Mar 2026, Revelio); ~4,800–5,200 (Jun 2026, Crustdata/Tracxn) | 2025–2026 | Run-rate $47B (May 2026); reported >$65B (end July 2026); Q2 2026 revenue >$11.5B | 2026 |
| **OpenAI** | $852B post-money ($122B raised, led by SoftBank) | 31 Mar 2026; confirmed flat in $7B employee tender, Aug 2026 | ~4,500 core (CB Insights); 7,832 (Revelio, Mar 2026) | 2026 | Run-rate ~$40B (Aug 2026, Sacra); 2025 booked revenue $13.07B, 2025 operating loss $20.9B | 2026 |
| **NVIDIA** | ~$4.0T (non-affiliate float, 25 Jul 2025, 10-K); ~$5T+ during 2026 (press) | 2025–2026 | **42,000** (31,000 in R&D) | FY2026 10-K (year ended 25 Jan 2026) | **$215.9B** revenue; $130.4B operating income; 71.1% gross margin | FY2026 |

Anthropic has not published an official headcount. The range above comes from three workforce trackers plus Fortune. Say "a few thousand" in the talk. Do not say an exact number unless you cite one tracker.

### Per-employee arithmetic (compute on stage, cite the inputs)

- **NVIDIA**: $215.9B / 42,000 = **$5.1 million revenue per employee**. At a $5T market cap: **~$119 million of market value per employee**.
- **OpenAI**: $852B / 7,832 = **$109 million of valuation per employee** (or $189M per head at 4,500). $40B / 7,832 = **$5.1M revenue per employee**.
- **Anthropic**: $965B / 5,000 = **$193 million of valuation per employee** (or $420M per head at the 2,300 end-2025 figure). $65B run-rate / 5,000 = **$13M revenue per employee**.
- **Walmart**: $681.0B revenue / 2.1 million associates = **$324,000 revenue per employee** (FY2025 10-K). FY2026: $339,600 (Bullfincher, secondary).

So Anthropic produces roughly **40 times** Walmart's revenue per head, and NVIDIA about **16 times**. On valuation per head the gap is in the hundreds.

### The historical giants

| Company | Employees | Date | Source |
|---|---|---|---|
| General Motors | 624,000 worldwide (400,000+ hourly US workers) | 1955 | Progressive Policy Institute (2017), citing company history |
| General Motors | **853,000 worldwide; 618,365 in the US** (peak; largest US private employer) | 1979 | MLive GM history; PPI (2017) |
| General Electric | 215,000 | 1955 | PPI (2017) |
| IBM | 56,000 (up from 3,000 in 1919) | 1955 | PPI (2017) |
| DuPont | 87,000 | 1955 | PPI (2017) |
| Walmart | **2.1 million worldwide; 1.6 million US** | FY2024–FY2025 | Walmart 10-K (2025); corporate site |

The PPI paper's point: the five frontier firms of 1955 "more than quintupled their employment between 1919 and 1955." That growth "replace[d] low-wage jobs with jobs that offered middle-class incomes." That was how a frontier firm spread its gains: through **payroll**.

Comparison for the stage: **Anthropic at ~5,000 people is about 0.6% of GM's 1979 headcount, and about 0.25% of Walmart's, at a valuation near $1 trillion.** GM in 1979 was the largest private employer in America. The largest AI lab in 2026 would not make the top 1,000.

Ford: no verified headcount found in this pass. Ford's historic peak is often cited near 500,000 in the late 1970s. Confirm before use.

### Why economists expect this

Autor et al. (2020) already documented the pattern before generative AI: superstar firms have low labor shares and high markups. Arrieta-Ibarra, Goff, Jiménez-Hernández, Lanier and Weyl (AEA Papers & Proceedings, 2018) wrote that Facebook, Google and Microsoft had "market capitalization and value-added... similar to or greater than a firm like Walmart, yet they employ 1–2 orders of magnitude fewer workers" and that their labor shares are "a small fraction of the traditional average 60–70 percent." They add: "The 'future' such firms represent would validate Piketty's (2013) foreboding of high capital shares." AI labs push that pattern one more order of magnitude.

---

## 4. Capital concentration: capex, index gains, VC

### AI capex

- The four largest hyperscalers (Amazon, Alphabet, Microsoft, Meta) guide **$720–745 billion** of combined calendar-2026 capex: Amazon ~$220B, Alphabet $195–205B, Microsoft ~$175B, Meta $130–145B. (Company earnings calls, July 2026; compiled by TMT Finance and others.) Including Oracle's FY2027 guidance, about **$835B**.
- Caveat: these are company-wide capex numbers on three different accounting bases. Not all of it is AI. Most of the growth is.
- Cumulative estimate: roughly **$1.05 trillion** of big-four capex from January 2023 through June 2026. (MLQ reconstruction; secondary.)
- In Q2 2026 the four spent **$170.1B** on capex, equal to **99%** of their operating cash flow. (Axis Intelligence, secondary.) Almost every dollar they earn is going back into AI infrastructure.
- NVIDIA reports visibility into "more than $1 trillion in cumulative Blackwell and Rubin revenue from the start of 2025 through 2027." (NVIDIA Annual Review, 2026)

### Share of S&P 500 gains from AI names

- The Magnificent Seven (Alphabet, Amazon, Apple, Meta, Microsoft, Nvidia, Tesla) accounted for **62–63%** of S&P 500 returns in 2023, **53–55%** in 2024, and **46%** in 2025. (State Street, July 2026; FactSet / J.P. Morgan via Landmark Wealth, 2026)
- Their share of S&P 500 market cap peaked at about **35% in early June 2026**, described as "the highest concentration in index history," above the dot-com peak. (Institutional Investor / CME sponsored content, 2026; secondary.) It was **33.9%** ($23.8T) in September 2026. (Motley Fool, citing Stock Analysis.)
- Nvidia alone is **7.9%** of the S&P 500. It was under 1% of the Mag-7 total in 2015. (Motley Fool, Sep 2026)
- Honest note: in 2026 the group has **underperformed** the index. Leadership rotated to semiconductor suppliers (AMD, Micron, Broadcom). Concentration is rotating within the AI supply chain, not leaving it.

### VC concentration

- AI took **52.7%** of global VC deal value in 2025 and **31.4%** of deals. (PitchBook, Jan 2026)
- In the US, AI took **65.4%** of 2025 deal value and 39.4% of deals, both records. US AI investment grew from $73.0B in 2022 to **$222.1B** in 2025. (PitchBook-NVCA, 2025)
- In Q1 2025, **57.9%** of global VC dollars went to AI; **70.2%** in North America. OpenAI's single $40B round was more than half of all global AI funding that quarter. (PitchBook, April 2025)
- Fund concentration: the 10 largest VC funds captured about **43%** of all commitments in 2025. (PitchBook Q3 2025 Quantitative Perspectives)
- Anthropic's Series H alone was **$65B** (May 2026). OpenAI's March 2026 round was **$122B**. These two rounds exceed the entire annual VC market of most years.

### Who owns the capital

- Top 1% of US households: **49.9%** of corporate equities and mutual fund shares. Top 0.1%: **24.0%**. (Federal Reserve Distributional Financial Accounts, Q2 2025)
- Top 10%: about **87–88%** of corporate equity, up from 82% in 1989. (Fed DFA; Batty et al., 2020)
- Bottom 50% of households: **2.5%** of total net worth. (Fed DFA, Q2 2025)

So when "AI companies soak up the capital," the gains flow to the roughly 10% of households who own almost all the equity. This is the link between company concentration and household concentration.

---

## 5. Political economy: what happens to democratic leverage

### The theory

**Korinek and Juelfs, *Preparing for the (Non-Existent?) Future of Work* (Brookings / NBER, 2022):**
- "If there are significant wage declines, avoiding mass misery will require other ways of distributing income than labor markets, whether via sufficiently well-distributed capital ownership or via benefits."
- Work produces "positive externalities such as social connections or political stability." They cite Boix (2022), *AI and the economic and informational foundations of democracy*.
- Their policy: "a small UBI or distributed ownership program that automatically scales up as labor's share in the economy declines."

**Korinek, US Senate AI Insight Forum statement (1 Nov 2023):**
- Without institutional reform, AGI "would likely deliver mass impoverishment, with all the resulting implications for our political system."
- Proposes a "seed UBI" and "universal basic capital (UBC)," meaning broad ownership stakes in AI, "that allow all Americans to participate in the prosperity generated by advanced AI."

**Acemoglu and Johnson (2023, 2024):** power sets the direction of technology. Workers gained a voice when factories concentrated them. If AI removes the need for workers, it removes the base of that voice.

### The historical analogies

**Engels' pause** (named by Robert Allen, *Explorations in Economic History*, 2009):
- 1780–1840: British output per worker rose **46%**; real wages rose **12%**.
- Profit rate rose from about **10%** (late 1700s) to over **20%** (mid-1800s). Labor's share of national income fell.
- 1840–1900: output per worker up 90%, real wages up **123%**. Wages caught up.
- Allen's mechanism: technical change raised demand for capital faster than capital could be built. High profits financed the catch-up. Acemoglu and Johnson add the political mechanism: unions, the vote, and factory laws.
- Lesson for the talk: the first sixty years of the last great technology shift went almost entirely to capital. People got their share only after they organized. **The question is what organizing looks like when the machines no longer need the people.**

**Ricardo's change of mind** (Acemoglu & Johnson, NBER 2024): David Ricardo first believed machinery would help workers. He revised his view after watching cotton weavers' wages fail to rise for decades. Automation can raise wages "only when accompanied by new tasks that raise the marginal productivity of labor and/or when there is sufficient additional hiring in complementary sectors."

**Enclosure**: the 16th–19th century conversion of English common land into private holdings. It removed the peasant's independent means of subsistence and forced dependence on wage labor. Acemoglu and Johnson (2023) treat medieval and early-modern agriculture as their first case of gains captured by elites. The AI parallel: if the "commons" of human skill loses its market value, people lose their independent claim on output. This analogy is rhetorical, not a published economic model. Say so.

### The fiscal-state argument (strongest version of the political claim)

Korinek and Lockwood (NBER 34873, 2026): transformative AI erodes both labor-income and consumption tax bases. If a state cannot tax its citizens because they no longer earn, and AI systems produce most value, then the state's revenue depends on the owners of AI capital. That is a structural shift in who the state must listen to. This is the most rigorous published version of "humans become untethered from power."

### What the AI CEOs themselves say

- **Dario Amodei** (Anthropic CEO), Axios, 28 May 2025: AI could eliminate **half of all entry-level white-collar jobs** and push US unemployment to **10–20%** within one to five years. His scenario: "Cancer is cured, the economy grows at 10% a year, the budget is balanced — and 20% of people don't have jobs." He proposed a **3% "token tax"** on AI revenue for redistribution: "Obviously, that's not in my economic interest."
- **Sam Altman** (OpenAI CEO), *Moore's Law for Everything*, March 2021: proposed an "American Equity Fund" that taxes large companies **2.5% of market value each year, paid in shares**, and land 2.5% in cash, paying every adult about **$13,500 a year** within a decade.
- Use: the people building the technology are themselves proposing to share its capital. They propose to do it through the US Treasury. That is the fork in the road.

---

## 6. Counterarguments and honest weak points

The speaker should know these cold. A hostile questioner will raise them.

### 6.1 "There is no economy-wide job disruption yet."

- **Yale Budget Lab** (Oct 2025, with monthly CPS updates): "the broader labor market has not experienced a discernible disruption since ChatGPT's release 33 months ago." Occupational mix is shifting only about 1 point faster than during the internet era, and the shift began in 2021, before ChatGPT.
- **Humlum and Vestergaard** (NBER 33777, revised March 2026): Danish administrative data on 25,000 workers in 11 exposed occupations. "Precise null effects on earnings and recorded hours... ruling out effects larger than 2% two years after the launch of ChatGPT." Average time savings from chatbots: about **3%**.
- **Brynjolfsson et al. (Aug 2026)** themselves: "We find no evidence of widespread, economy-wide job displacement."
- Response: the claim is about the **direction and the destination**, not this quarter. The labor share is at a record low. The entry-level canaries are getting worse each year (13% → 16% → 19%). Engels' pause took sixty years to resolve.

### 6.2 "Acemoglu says AI's effect is small."

- His 2024 forecast: **0.53–0.66% TFP** over ten years. He calls the big-transformation claims exaggerated.
- Response: the same paper says AI will "widen the gap between capital and labor income." The thesis does not need fast growth. It needs the gains to go to capital. Also, his figure is a forecast from 2023-era task-exposure data. Anthropic's revenue went from ~$9B (end 2025) to a $65B run-rate (July 2026), a pace no 2024 model assumed.

### 6.3 "Firms that adopt AI hire more."

- Aghion et al. (2025, France) and many firm-level studies find AI-adopting firms grow employment.
- Response: Aghion et al. say the risk is **between firms**, not within them, and that "larger and more productive firms should be the great winners." That is the concentration claim, stated by the optimists.

### 6.4 "Augmentation will win, like Autor says."

- Autor's 2024 paper is a serious case that AI could rebuild the middle class.
- Response: Autor calls it "not a forecast but an argument about what is possible." Brynjolfsson's data show the automation applications are the ones cutting jobs. Trammell and Korinek say steering technology toward augmentation permanently is "very inefficient (and so very difficult)."

### 6.5 "AI companies are not that small. They have thousands of staff and are hiring fast."

- True. OpenAI's tracked headcount roughly tripled from 2023 to 2026. Anthropic's grew about tenfold.
- Response: the point is the **ratio**. Even at 8,000 people, OpenAI is 1% of GM's 1979 workforce at a valuation many times GM's. And the AI labs are hiring people to build the thing that reduces hiring elsewhere.

### 6.6 "Concentration is already unwinding. The Mag 7 lagged the index in 2026."

- True for the seven names. Leadership rotated to semiconductor suppliers.
- Response: the rotation stayed inside the AI supply chain. NVIDIA is still 7.9% of the S&P 500. Two private companies just raised $187B in two rounds.

### 6.7 "The AI companies lose money. This is a bubble."

- OpenAI's 2025 operating loss was **$20.9B** on $13.07B of revenue. Hyperscaler capex equals 99% of their operating cash flow.
- Response: this is a real risk to the **market-cap** numbers, not to the **labor-share** argument. If the bubble pops, capital destruction hits equity owners; the automation that already happened does not un-happen. But the speaker should not lean on valuations as if they were earnings.

### 6.8 "Humans have always found new work."

- The strongest historical prior. Every previous automation wave created new tasks.
- Response: Korinek and Suh's model shows this holds **only if** the tail of hard-to-automate tasks is thick enough. AGI by definition thins that tail to zero. Susskind: "this time really is different" because the machines now reach cognitive, non-routine tasks. This is a bet on the technology, and the speaker should say it is a bet.

### 6.9 Weak points in the "100% of capital" framing

- "AI companies soak up almost 100% of the valuable capital in the world" is not a measured fact. Measured facts: 52.7% of VC, ~35% of the S&P 500 (Mag 7), $720–745B of capex from four firms. Large, not total. Say "the majority of new capital formation" or "a record share," not "100%."
- Trammell and Korinek's "labor share toward zero" is a model result under full automation, not a data point.

### 6.10 Weak points in the Bittensor link (see section 7)

- Bittensor's total market cap is on the order of **$2 billion** (September 2026; verify). NVIDIA's is about $5 trillion. The scale gap is roughly 2,500 to 1.
- Bittensor pays for **contribution** (compute, models, data, validation). It is not a dividend to people who cannot contribute. It does not, by itself, solve displacement.
- TAO ownership distribution is not documented in this brief. If it is concentrated, the "open ownership" claim weakens. Get the data.
- The chain today runs on a sudo key controlled by Jacob (project context). A critic will say a permissionless network with a single upgrade key is not yet the alternative it claims to be. The talk should own this.

---

## 7. Where a decentralized, open-ownership network fits

### The economists' answer is "broad ownership of AI capital." They disagree on the mechanism.

| Thinker | Proposal | Mechanism | Runs through |
|---|---|---|---|
| **Anton Korinek** (Senate statement 2023; Korinek & Juelfs 2022; Korinek & Lockwood 2026) | "Universal Basic Capital": broadly distributed equity in AI companies. "Predistributing AI gains through ownership." Plus a "seed UBI" that scales with the labor-share decline. | Government grants or mandates | The state |
| **Thomas Piketty** (*Capital in the Twenty-First Century*, 2013/2014) | When the return on capital exceeds growth (r > g), wealth concentrates. Remedy: progressive global wealth tax; later, broad capital endowments. | Taxation | The state |
| **Eric Posner & Glen Weyl** (*Radical Markets*, 2018) and **Arrieta-Ibarra, Goff, Jiménez-Hernández, Lanier & Weyl** (AEA P&P, 2018) | "Data as labor": pay people for the data that trains AI. Create the missing market. Data unions for bargaining power. | New markets, collective bargaining | Firms and unions |
| **Jaron Lanier** (*Who Owns the Future?*, 2013) | Micropayments to people whose data creates value. "Data dividends." | Platform-level payments | Platforms |
| **Sam Altman** (*Moore's Law for Everything*, 2021) | American Equity Fund: 2.5% of large companies' market value per year, paid in shares to every adult. | Tax paid in equity | The state |
| **Dario Amodei** (Axios, May 2025) | 3% token tax on AI revenue, redistributed. | Tax | The state |
| **O'Keefe, Cihon, Garfinkel, Flynn, Leung & Dafoe** (GovAI, *The Windfall Clause*, 2020) | AI firms pre-commit to donate a large share of any "historically unprecedented" profits. | Voluntary contract | Firms |
| **Daniel Susskind** (2020) | "Big State" for distribution; conditional basic income. | Tax and transfer | The state |
| **Korinek & Juelfs** (2022) | Income "via sufficiently well-distributed capital ownership or via benefits." | Either | Either |

Pattern: nearly every serious proposal to share AI capital requires **a national government to tax and redistribute**, or **a private AI firm to volunteer**. Korinek and Lockwood (2026) list "Universal Basic Capital (ensure broad ownership of AI companies)" and "windfall clauses (voluntary commitments)" as the private mechanisms. Both depend on the goodwill or the jurisdiction of the very entities that are concentrating the capital.

### Where Bittensor fits: the third mechanism

Bittensor is a different mechanism from all of the above. It is neither tax-and-transfer nor corporate charity. It is **ownership through open contribution**.

What the network does (from the project context and public documentation):
- Anyone can mine, validate, or build a subnet. No permission needed.
- New TAO is issued to those who provide AI work the network values. Fixed supply of **21 million**, same schedule as Bitcoin. First halving **December 2025**, cutting issuance from ~7,200 to ~3,600 TAO per day. (Grayscale Research, 2025)
- **No pre-mine.** No insider allocation. (Project context; Grayscale.)
- About **129 active subnets** as of late 2025, up from ~32 in early 2025. (Grayscale Research, 2025)

How that maps onto the economists' framework:

1. **It answers Korinek's "Universal Basic Capital" without a state.** UBC asks: how do ordinary people come to own a piece of AI capital? Korinek's answer is a government program. Bittensor's answer is: the capital is issued directly to whoever contributes, under rules nobody can change unilaterally (once the sudo key is removed). Frame it as "permissionless basic capital." Be clear this is your framing, not Korinek's.

2. **It is the "data as labor" idea made general.** Weyl and Lanier want people paid for the data that trains AI. Bittensor pays for compute, models, data, and validation. It is a market for AI inputs where the payment is ownership in the network itself, not a wage. This is the closest published economic lineage. Weyl and Posner did not write about Bittensor. Do not imply they did.

3. **It attacks the superstar-firm mechanism directly.** Autor et al. (2020) show the labor share falls because value pools in a few low-labor-share firms. A network where the "firm" has no shareholders, and value accrues to a distributed set of contributors and token holders, is structurally the opposite of a superstar firm. Whether it wins is an open question. Its structure is the right shape.

4. **It restores a version of Thelwall's "political society."** Thelwall said the factory gave workers power because it gathered them. Bittensor gathers contributors on a network with shared economic interest and a shared stake in the rules. It is a place to organize when the factory no longer exists. This is the strongest rhetorical bridge to Acemoglu and Johnson's "countervailing power." It is a metaphor, not a measurement.

5. **It is jurisdiction-free.** Every state-run proposal (Korinek, Piketty, Altman, Amodei, Susskind) works only inside one country's borders, and only if that country's government chooses to act. Korinek and Lockwood note tax bases erode as AI advances. Bittensor is the only mechanism in the table that does not depend on a state's willingness or capacity to tax. This is the "third path" from the project context, stated in economic terms.

### Honest limits of the link

- **Scale.** ~$2B network vs. ~$5T NVIDIA and ~$1.8T combined for OpenAI and Anthropic. Bittensor is a proof of mechanism, not a counterweight, yet.
- **It rewards contribution, not need.** A displaced 55-year-old paralegal cannot mine a subnet. Bittensor addresses "who owns AI capital," not "how does a person with no marketable contribution eat." Korinek's UBI half of the answer still needs a mechanism. Say so.
- **Ownership concentration inside Bittensor is not documented here.** Fair launch does not guarantee dispersed holdings today. Publish the distribution or expect the question.
- **The sudo key.** A single upgrade key is the opposite of the "rules nobody can change" claim. The project context says this is a known gap with a path to proof-of-stake or proof-of-work. The talk should state the current state and the path, in that order.
- **Token value is speculative.** "Ownership of AI capital" only means something if the token captures real demand for AI work, not just emissions. A skeptic will cite the gap between subnet activity and paying customers.
- **No mainstream economist has endorsed a token network as the UBC mechanism.** The published literature stops at "broad capital ownership." The step from there to "permissionless network" is Bittensor's argument to make. Make it as an argument, with the caveats above, and it will hold. Make it as a settled fact and it will not.

### Suggested framing for the talk

"The economists agree on the diagnosis: AI shifts income from labor to capital, and capital is owned by very few. They agree on the cure: broad ownership of AI capital. Every one of their proposals runs through a government or through the goodwill of the companies doing the concentrating. Bittensor is the one design where ownership is issued directly to anyone who contributes, under open rules, in no country. It is small. It is not finished. It is the only one of its kind."

---

## Quotable facts

Each line: fact, source, date.

- US labor share was 52.9% in Q2 2026, the lowest since the series began in 1947. — BLS Productivity and Costs, preliminary, 2026.
- US labor share was 65.8% in Q1 1947 and 62.8% in Q4 2000. — BLS, The Economics Daily, 2017.
- Global labor income share fell 1.6 points from 2004 to 52.3% in 2024; workers received $2.4 trillion less than under a stable share. — ILO WESO Update, 4 Sep 2024.
- Automation explains 50–70% of US wage-structure change 1980–2016, with only 3.4% cumulative TFP gain. — Acemoglu & Restrepo, Econometrica, 2022.
- AI will raise US TFP by at most 0.66% over ten years, and "is predicted to widen the gap between capital and labor income." — Acemoglu, NBER 32487, May 2024.
- Under full automation, "the burst in output... will almost all accrue to capital, driving the labor share toward zero." — Trammell & Korinek, NBER 31815, rev. April 2026.
- In the aggressive AGI scenario, wages collapse after about 3 years while output grows more than tenfold. — Korinek & Suh, NBER 32255, March 2024.
- Employment of 22–25-year-olds in AI-exposed jobs is 19% below its peers' path as of June 2026, up from 13% in August 2025. — Brynjolfsson, Chandar & Chen, Stanford, Aug 2026.
- 40% of global jobs and 60% of advanced-economy jobs are exposed to AI. — IMF Staff Discussion Note, 14 Jan 2024.
- Anthropic: $965B valuation, $65B raised, run-rate revenue $47B. — Anthropic press release, 28 May 2026.
- Anthropic run-rate revenue reportedly passed $65B by end of July 2026, from ~$9B at end of 2025. — GraniteShares / AlphaSense summaries of company statements, 2026 (secondary; verify against S-1 when public).
- Anthropic had 2,300 employees at end of 2025 versus NVIDIA's 42,000. — Fortune, 18 Jun 2026.
- OpenAI: $852B valuation, $122B round, 31 Mar 2026; $7B employee tender at the same price, Aug 2026. — Sacra; ValueAdd VC, 2026.
- OpenAI 2025 operating loss $20.9B on $13.07B revenue. — ValueAdd VC, 2026 (secondary).
- NVIDIA FY2026: $215.9B revenue, $130.4B operating income, 42,000 employees. — NVIDIA 10-K and Q4 FY2026 press release, 25 Feb 2026.
- NVIDIA: "visibility into more than $1 trillion in cumulative Blackwell and Rubin revenue" 2025–2027. — NVIDIA Annual Review, 2026.
- GM employed 853,000 worldwide at its 1979 peak, 618,365 in the US. — MLive GM history; PPI, 2017.
- Walmart employs 2.1 million people; revenue $681.0B; about $324,000 per employee. — Walmart 10-K, FY2025.
- Facebook, Google and Microsoft matched Walmart's value with "1–2 orders of magnitude fewer workers." — Arrieta-Ibarra, Goff, Jiménez-Hernández, Lanier & Weyl, AEA P&P, 2018.
- Four hyperscalers guide $720–745B of 2026 capex. — Company earnings calls compiled by TMT Finance, July 2026.
- Big-four Q2 2026 capex of $170.1B equaled 99% of their operating cash flow. — Axis Intelligence, 2026 (secondary).
- AI took 52.7% of global VC deal value in 2025 and 65.4% of US VC deal value. — PitchBook / PitchBook-NVCA, Jan 2026.
- OpenAI's $40B round was over half of all global AI VC in Q1 2025. — PitchBook, 17 Apr 2025.
- Magnificent Seven drove 62% of S&P 500 returns in 2023, 53% in 2024, 46% in 2025. — State Street, 20 Jul 2026; J.P. Morgan / FactSet data.
- Magnificent Seven peaked at ~35% of S&P 500 market cap in early June 2026, a record. — Institutional Investor, 2026 (sponsored; secondary).
- Nvidia alone is 7.9% of the S&P 500. — Motley Fool citing Stock Analysis, Sep 2026.
- Top 1% of US households own 49.9% of corporate equities; top 0.1% own 24.0%; bottom 50% own 2.5% of net worth. — Federal Reserve DFA, Q2 2025.
- Engels' pause: 1780–1840 British output per worker +46%, real wages +12%; profit rate doubled. — Allen, Explorations in Economic History, 2009.
- "Every large workshop and manufactory is a sort of political society, which no act of parliament can silence." — John Thelwall (1790s), quoted in Acemoglu & Johnson, *Power and Progress*, 2023.
- "Wages are unlikely to rise when workers cannot push for their share of productivity growth." — Acemoglu & Johnson, NBER 32416, May 2024.
- Avoiding mass misery "will require other ways of distributing income than labor markets, whether via sufficiently well-distributed capital ownership or via benefits." — Korinek & Juelfs, Brookings/NBER, 2022.
- Without reform, AGI "would likely deliver mass impoverishment." Proposes "universal basic capital." — Korinek, US Senate AI Insight Forum, 1 Nov 2023.
- Transformative AI "may gradually erode the two main tax bases... labor income and human consumption." — Korinek & Lockwood, NBER 34873, 2026.
- AI could cut half of entry-level white-collar jobs and push unemployment to 10–20% in 1–5 years. Proposes a 3% token tax. — Dario Amodei, Axios, 28 May 2025.
- American Equity Fund: 2.5% of large-company market value per year, paid in shares, to every adult. — Sam Altman, *Moore's Law for Everything*, Mar 2021.
- "The broader labor market has not experienced a discernible disruption since ChatGPT's release 33 months ago." — Yale Budget Lab, Oct 2025.
- AI chatbots: "precise null effects on earnings and recorded hours... ruling out effects larger than 2%." — Humlum & Vestergaard, NBER 33777, rev. Mar 2026.
- Bittensor: 21 million TAO cap, no pre-mine, first halving December 2025, ~129 subnets. — Grayscale Research, 2025; project context.

---

## Sources

### Official statistics and filings
- BLS, Productivity and Costs, Second Quarter 2026, Preliminary: https://www.bls.gov/news.release/PDF/prod2.PDF
- BLS, Productivity and Costs, First Quarter 2026: https://www.bls.gov/news.release/prod2.htm
- BLS, Productivity and Costs, Q4 and Annual 2025, Revised (24 Mar 2026): https://www.dol.gov/newsroom/economicdata/prod2_03242026.pdf
- BLS, "Labor share of output has declined since 1947," The Economics Daily (2017): https://www.bls.gov/opub/ted/2017/labor-share-of-output-has-declined-since-1947.htm
- ILO, World Employment and Social Outlook: September 2024 Update: https://www.ilo.org/sites/default/files/2024-10/WESO%20September%202024%20Update%20-%20Final.pdf
- ILO WESO Sep 2024 Technical Annex: https://www.ilo.org/sites/default/files/2024-08/WESO%20September%202024%20-%20Technical%20Annex.pdf
- IMF, Gen-AI: Artificial Intelligence and the Future of Work (SDN/2024/001, 14 Jan 2024): https://www.imf.org/-/media/files/publications/sdn/2024/english/sdnea2024001.pdf
- Federal Reserve, Distributional Financial Accounts: https://www.federalreserve.gov/releases/z1/dataviz/dfa/index.html
- Fed DFA series via ALFRED (release 19 Sep 2025): https://alfred.stlouisfed.org/release?rid=453&rd=2025-09-19
- Batty et al., "Introducing the Distributional Financial Accounts" (2020): http://www.kamilasommer.net/DFA_NBER_June2020.pdf
- NVIDIA Form 10-K, fiscal year ended 25 Jan 2026: https://www.sec.gov/Archives/edgar/data/1045810/000104581026000021/nvda-20260125.htm
- NVIDIA Q4 and FY2026 results press release (25 Feb 2026): https://investor.nvidia.com/news/press-release-details/2026/NVIDIA-Announces-Financial-Results-for-Fourth-Quarter-and-Fiscal-2026/default.aspx
- NVIDIA 2026 Annual Review: https://s201.q4cdn.com/141608511/files/doc_financials/2026/ar/2026-Annual-Report-Web.pdf
- Walmart Form 10-K, FY2025: https://stock.walmart.com/sec-filings/all-sec-filings/content/0000104169-25-000059/0000104169-25-000059.pdf
- Walmart, "How many people work at Walmart?": https://corporate.walmart.com/askwalmart/how-many-people-work-at-walmart
- Anthropic, "Anthropic raises $65B in Series H funding at $965B post-money valuation" (28 May 2026): https://www.anthropic.com/news/series-h

### Academic papers and books
- Acemoglu, D. (2024). The Simple Macroeconomics of AI. NBER WP 32487: https://www.nber.org/papers/w32487
- Acemoglu, D. & Restrepo, P. (2022). Tasks, Automation, and the Rise in US Wage Inequality. Econometrica 90(5): https://onlinelibrary.wiley.com/doi/full/10.3982/ECTA19815
- Acemoglu, D. & Johnson, S. (2023). Power and Progress: Our Thousand-Year Struggle Over Technology and Prosperity. PublicAffairs. Excerpt: https://shapingwork.mit.edu/power-and-progress/
- Acemoglu, D. & Johnson, S. (2024). Learning from Ricardo and Thompson. NBER WP 32416 / Annual Review of Economics 16: https://www.nber.org/papers/w32416
- Aghion, P., Bunel, S., Jaravel, X., Mikaelsen, T., Roulet, A. & Søgaard, J. (2025). How Different Uses of AI Shape Labor Demand: Evidence from France. AEA Papers & Proceedings 115: https://www.aeaweb.org/articles?id=10.1257/pandp.20251047
- Allen, R. C. (2009). Engels' Pause. Explorations in Economic History 46(4): https://www.nuff.ox.ac.uk/Users/Allen/engelspause.pdf
- Arrieta-Ibarra, I., Goff, L., Jiménez-Hernández, D., Lanier, J. & Weyl, E. G. (2018). Should We Treat Data as Labor? AEA Papers & Proceedings 108: https://www.diegojimenezh.com/assets/pdf/published/ArrietaIbarra_et_al_DataAsLabor.pdf
- Autor, D. (2024). Applying AI to Rebuild Middle Class Jobs. NBER WP 32140: https://www.nber.org/papers/w32140
- Autor, D., Dorn, D., Katz, L., Patterson, C. & Van Reenen, J. (2020). The Fall of the Labor Share and the Rise of Superstar Firms. QJE 135(2): https://doi.org/10.1093/qje/qjaa004
- Brynjolfsson, E., Chandar, B. & Chen, R. (2025–26). Canaries in the Coal Mine? Stanford Digital Economy Lab (Aug 2026 update page): https://digitaleconomy.stanford.edu/publication/canaries-in-the-coal-mine-six-facts-about-the-recent-employment-effects-of-artificial-intelligence/ ; Nov 2025 PDF: https://digitaleconomy.stanford.edu/app/uploads/2025/12/CanariesintheCoalMine_Nov25.pdf ; Aug 2025 SIEPR page: https://siepr.stanford.edu/publications/working-paper/canaries-coal-mine-six-facts-about-recent-employment-effects-artificial
- Humlum, A. & Vestergaard, E. (2025, rev. Mar 2026). Large Language Models, Small Labor Market Effects. NBER WP 33777: https://www.nber.org/system/files/working_papers/w33777/w33777.pdf
- Korinek, A. & Juelfs, M. (2022). Preparing for the (Non-Existent?) Future of Work. Brookings / NBER WP 30172: https://www.brookings.edu/articles/preparing-for-the-non-existent-future-of-work/ ; Milken Review summary: https://www.milkenreview.org/articles/the-non-existent-future-of-work
- Korinek, A. (2023). Written statement, US Senate AI Insight Forum on Workforce (1 Nov 2023): https://www.brookings.edu/wp-content/uploads/2023/12/Korinek_Senate_Statement_11.01.2023.pdf
- Korinek, A. (2024). The Economics of Transformative AI. NBER Reporter 2024:4: https://www.nber.org/reporter/2024number4/economics-transformative-ai
- Korinek, A. & Lockwood, L. (2026). Public Finance in the Age of AI: A Primer. NBER WP 34873: https://www.nber.org/papers/w34873
- Korinek, A. & Stiglitz, J. (2017/2019). Artificial Intelligence and Its Implications for Income Distribution and Unemployment. NBER WP 24174: https://www.nber.org/papers/w24174
- Korinek, A. & Suh, D. (2024). Scenarios for the Transition to AGI. NBER WP 32255: https://www.nber.org/papers/w32255 ; arXiv: https://arxiv.org/html/2403.12107
- O'Keefe, C., Cihon, P., Garfinkel, B., Flynn, C., Leung, J. & Dafoe, A. (2020). The Windfall Clause. GovAI: https://cdn.governance.ai/Windfall-Clause-Report.pdf
- Piketty, T. (2014). Capital in the Twenty-First Century. Harvard University Press.
- Posner, E. & Weyl, E. G. (2018). Radical Markets. Princeton University Press. Brookings summary of "data as labor": https://www.brookings.edu/articles/should-we-treat-data-as-labor-lets-open-up-the-discussion/
- Susskind, D. (2020). A World Without Work. Penguin: https://www.penguin.co.uk/books/306864/a-world-without-work-by-susskind-daniel/9780141986807 ; LARB interview on the "Big State" and conditional basic income: https://lareviewofbooks.org/blog/interviews/13652/
- Susskind, D. (2024). Growth: A Reckoning. Penguin: https://www.penguin.co.uk/books/446381/growth-by-susskind-daniel/9780141998725
- Trammell, P. & Korinek, A. (2023, rev. Apr 2026). Economic Growth under Transformative AI. NBER WP 31815: https://www.nber.org/system/files/working_papers/w31815/w31815.pdf ; Annual Review of Economics (June 2026): https://www.annualreviews.org/content/journals/10.1146/annurev-economics-051624-061002
- Yale Budget Lab (Oct 2025). Evaluating the Impact of AI on the Labor Market: Current State of Affairs: https://budgetlab.yale.edu/research/evaluating-impact-ai-labor-market-current-state-affairs

### Company data, market data, and press (secondary; verify before quoting on stage)
- Sacra, OpenAI profile (valuation, revenue, IPO filing): https://sacra.com/c/openai/
- ValueAdd VC, "OpenAI Hits $40B ARR" (Aug 2026): https://valueaddvc.com/blog/openai-revenue-2026-20b-arr-4b-month-path-to-profitability
- Revelio Labs, OpenAI headcount: https://www.reveliolabs.com/companies/openai-foundation/employees
- Revelio Labs, Anthropic headcount: https://www.reveliolabs.com/companies/anthropic-pbc/employees
- Tracxn, Anthropic profile: https://tracxn.com/d/companies/anthropic/__SzoxXDMin-NK5tKB7ks8yHr6S9Mz68pjVCzFEcGFZ08
- Crustdata, Anthropic headcount: https://profiles.crustdata.com/company/anthropic
- Fortune, "Dario Amodei has only 1 direct report" (18 Jun 2026; 2,300 employees at end 2025): https://fortune.com/2026/06/18/anthropic-ceo-dario-amodei-one-direct-report-unconventional-management-structure-start-up-success/
- TechCrunch, "Anthropic raises $65 billion" (28 May 2026): https://techcrunch.com/2026/05/28/anthropic-raises-65-billion-nears-1t-valuation-ahead-of-ipo/
- GraniteShares, Anthropic IPO explainer (run-rate table): https://graniteshares.com/research/anthropic-ipo-2026-explained-from-965-billion-to-a-possible-2-trillion-listing/
- AlphaSense, Anthropic IPO preview: https://www.alpha-sense.com/resources/research-articles/anthropic-ipo-preview/
- Axios, "Sleepwalking into a white-collar bloodbath" (Amodei interview, 28 May 2025): https://www.axios.com/2025/05/28/ai-jobs-white-collar-unemployment-anthropic
- Sam Altman, Moore's Law for Everything (Mar 2021): https://moores.samaltman.com/
- CNBC on Altman's proposal (17 Mar 2021): https://www.cnbc.com/2021/03/17/openais-altman-ai-will-make-wealth-to-pay-all-adults-13500-a-year.html
- Progressive Policy Institute, "An Analysis of Job and Wage Growth in the Tech/Telecom Sector" (2017; GM, GE, IBM, DuPont historical headcounts): https://www.progressivepolicy.org/wp-content/uploads/2017/09/PPI_TechTelecomJobs_V4.pdf
- MLive, "A brief history of General Motors Corp." (1979 peak employment): https://www.mlive.com/business/2008/09/a_brief_history_of_general_mot.html
- Bullfincher, Walmart revenue per employee: https://bullfincher.io/companies/walmart/revenue-per-employee
- TMT Finance, "2026 hyperscaler capex tops US$700bn": https://www.tmtfinance.com/intel/2026-hyperscaler-capex-tops-us700bn-analysis
- MLQ, "Big Tech's 2026 Capex Range Reaches $720 Billion to $745 Billion": https://mlq.ai/news/big-techs-2026-capex-range-reaches-720-billion-to-745-billion/
- Axis Intelligence, AI Capex Tracker 2026: https://axis-intelligence.com/ai-capex-tracker/
- State Street, "Magnificent 7 no longer moving as one trade" (20 Jul 2026): https://www.ssga.com/us/en/individual/insights/mind-on-the-market-20-july-2026
- Landmark Wealth Management, Mag 7 return-share data (FactSet / J.P. Morgan): https://landmarkwealthmgmt.com/articles/magnificent-7-performance-in-the-sp-500-how-mag-7-stocks-continue-to-shape-market-returns-and-earnings-growth/
- Institutional Investor / CME, "Is the S&P 500's Concentrated Rally Starting to Diversify?": https://www.institutionalinvestor.com/article/sponsored-content/sp-500s-concentrated-rally-starting-diversify
- Motley Fool, "The Magnificent Seven's Market Cap vs. the S&P 500" (Sep 2026): https://www.fool.com/research/magnificent-seven-sp-500/
- Reuters, "Magnificent 7 results set to test broadening US stock market" (29 Jul 2026): https://www.reuters.com/business/magnificent-7-results-set-test-broadening-us-stock-market-2026-07-29/
- PitchBook, "AI eats up 58% of global venture dollars" (17 Apr 2025): https://pitchbook.com/news/articles/ai-startups-57-9-percent-global-venture-dollars-fear-of-missing-out-drives-up-dealmaking-q1-2025
- PitchBook, Q3 2025 Quantitative Perspectives: https://pitchbook.com/news/reports/q3-2025-quantitative-perspectives-a-fork-in-the-road
- PitchBook, Q4 2025 AI VC Trends: https://pitchbook.com/news/reports/q4-2025-ai-vc-trends
- GamesBeat summary of PitchBook 2025 full-year VC data: https://gamesbeat.com/global-vc-deals-grew-in-2025-thanks-to-ai-investments-pitchbook-first-look/
- Grayscale Research, "Bittensor on the Eve of the First Halving" (2025): https://research.grayscale.com/reports/bittensor-on-the-eve-of-the-first-halving-research
- MIT News, "An AI challenge only humans can solve" (Power and Progress, 17 May 2023): https://news.mit.edu/2023/power-and-progress-book-ai-inequality-0517