The Concentration of Wealth

Executive Summary

Concentration Without Precedent

Wealth concentrations are as old as recorded history. This one is structurally different. Three things make today's moment unique: the concentration is driven by an efficiency breakthrough in frontier technology so large it has no historical precedent; revenue is rotating out of every other industry and into the firms that control that breakthrough; and the margins those firms generate are being redeployed at unprecedented scale to cement that dominance further.

The result is that a small number of firms now hold the infrastructure of intelligence itself — and the cash flows to keep concentrating it. Nearly half the growth in most American retirement accounts traces back to a single category: consumer technology, monetized largely through advertising and AI. The value creation is extraordinary and worth celebrating. The question this article presses is what we owe one another when structural power of this scale gets concentrated in so few hands.

This report covers economic concentration only. For the religious and moral profile of the people running these firms, see the companion study The Religious Profile of Technology.

Section 1

History of Concentrations

Wealth concentration has a long history. Understanding today's moment requires understanding the four prior eras that shape how economists, regulators, and the public read concentration when they see it. Each prior era was bounded by something — a state grant, a physical limit, an expectation without earnings. Today's is bounded by none of them.

The First Monopolist

The Dutch East India Company (VOC), founded 1602, was the world's first publicly traded company — and the first to achieve monopoly-scale concentration. At its peak in the 1630s–1640s, the VOC's inflation-adjusted market capitalization was roughly $8 trillion in today's dollars, larger than any company in history. It controlled trade routes, military force, currency issuance, and treaty-making. The French East India Company and the British South Sea Company (which collapsed in the 1720 bubble) followed the same template: state-granted monopoly, single-sector dominance, extreme capital concentration.

The Railroad Era

By the late 19th century, railroads made up the dominant share of U.S. equity markets — roughly 60–70% of all publicly traded market value at peak. Standard Oil, Carnegie Steel, and J.P. Morgan's banking empire formed a second layer of concentration in energy and finance. The era produced the first major U.S. antitrust response: the Sherman Act of 1890 and the subsequent breakup of Standard Oil in 1911. Concentration was real, but bounded by the physical limits of the underlying industries — you could only lay so much track, refine so much oil.

The Dot-Com Era

Between 1995 and 2000, broad tech's share of U.S. equity more than doubled — from ~14% to ~37%. The sector P/E reached 65×. This was concentration driven by expectation rather than earnings: a bet on what the internet would become, not what it had yet produced. The correction was swift. By 2002, tech share had fallen to 19% and the P/E compressed to 32×.

The 2000 crash is usually remembered as a warning about overvaluation. The more important lesson is that the underlying thesis was correct — just early. The companies that survived (Amazon, eBay, Google) and the ones that came next (Facebook, Apple's iPhone arc, Netflix) built exactly the internet-scale businesses the bubble had bet on. The market was right about the direction, wrong about the timing, and structurally underestimating how concentrated the winners would be.

Today's Technology

Today's tech share of ~54% of U.S. public equity exceeds the dot-com peak and exceeds the railroad era's peak share of its own market. Unlike the VOC (state monopoly), railroads (physical limits), or dot-com (earnings-free), today's concentration is supported by genuine earnings at a scale no prior era produced. The P/E is elevated (~40×) but not detached from reality. That is what makes the current moment structurally different — and more durable.

Section 2

The Fundamentals of Today's Concentration

What a Kilowatt Now Does

In 1850, a steam engine converting 1 kWh of energy could produce approximately 0.3 hours of equivalent human physical labor. In 2025, an NVIDIA H100 GPU converting the same 1 kWh produces approximately 8,600,000,000,000 hours of equivalent human computational work — a roughly 28-trillion-fold improvement in output per unit of energy input.

No prior technological transition produced anything close to this ratio. The internal combustion engine improved on steam by roughly 3–5× in useful work per unit of energy. The transistor and integrated circuit improved on IC-engine-era electromechanical computing by roughly 10,000×. The jump from 1970s transistor computing to 2025 AI accelerators is roughly 1,000,000×. Each transition produced a wave of market concentration; each wave was larger than the last because the underlying transformation function kept compounding.

The implication is direct. When 1 kWh of energy can generate 8.6 trillion hours of computational work, the firm that controls the infrastructure to do that conversion — the GPU clusters, the data centers, the training pipelines — captures economic value at a rate no prior form of capital can match. A factory requires land, raw materials, labor, logistics. A GPU cluster requires electricity and software. The marginal cost of the next unit of output approaches zero in a way that physical capital never did.

Structural Revenue Support in Addition to Multiple

Today's concentration differs from the dot-com era because it is grounded in genuine earnings, not expectation. In 2000, tech's 37% market share was built on a P/E of 65× — extraordinary multiples applied to minimal earnings. Today's 54% share sits on a P/E of ~40× applied to roughly $4.5 trillion in annual revenue. The valuation is elevated; the earnings foundation is real and growing.

Revenue rotating into the tech sector from other industries matters as much as the efficiency itself. Advertising budgets that once funded print, broadcast, and radio now flow to Google and Meta. Retail margins that once went to physical stores now flow to Amazon. Enterprise software spending that once supported dozens of niche vendors now consolidates around a handful of platforms. The concentration of equity market share is downstream of an underlying shift in where the economy's revenue base is going — and that shift is structural, not cyclical.

Section 3

The Buyback Pivot

The story of tech concentration is not only about market cap. It is also about what the firms at the top do with the cash that concentration produces. Around 2013, the largest software-tech firms began shifting structurally — away from reinvesting into R&D, toward returning capital to shareholders via buybacks. By 2022, the top seven were returning 94% of net income this way, up from 4% a decade earlier.

The mechanism

Tech firms in their growth phase reinvest cash into R&D — funding the next generation of products that drive future growth. The 2000–2012 period in the chart above shows this: buybacks negligible or negative, R&D growing in both absolute and proportional terms.

Around 2013, the largest firms entered a different phase. Apple's $50B buyback announcement in March 2013 inflected the curve. After the 2017 Tax Cuts and Jobs Act, the acceleration became structural. By 2022, the seven firms in the sample were collectively returning roughly $200B/year via buybacks — 94% of net income.

Now Reinvesting into Aggressive CapEx to Cement Dominance

The same firms that redirected cash from R&D to buybacks have now pivoted again — this time toward AI infrastructure at a scale that dwarfs both. The attention economy — Google search ads, Facebook/Instagram feed ads, YouTube pre-rolls — generated the cash surpluses now being redeployed into compute. The firms that monetized human attention are the same firms now buying the infrastructure to automate human cognition.

Combined CapEx by the four largest hyperscalers grew from ~$93 billion in 2020 to ~$384 billion in 2025, with guidance for 2026 projecting ~$700 billion. This is a structural redeployment of attention-economy cash flows into the compute layer of the intelligence economy.

Section 4

The Eras of Market Growth

If you plot inflation-adjusted U.S. total market value from 1900 to today and slice that growth by the dominant technology era driving it, five distinct periods emerge. Each era was powered by a different transformation of energy into economic output. Each produced a different concentration peak. The current era — intelligence — is the steepest.

The pattern is consistent across eras, but the trajectory is not linear. Each new wave concentrated more sharply than the last. The pre-industrial era concentrated in railroads, steel, and oil — bounded by track laid and ore extracted. The industrial era spread across electrification, autos, aviation, and finance, broadening the market. The information era re-concentrated around PC and internet platforms. The attention era narrowed further, to mobile and ad-funded social media. The intelligence era is narrowest of all — a handful of hyperscalers controlling the GPU clusters that everything else now runs through.

The same platforms generating these returns are also the primary shapers of how children understand the world, what they desire, and who they become. For a deeper look at those dynamics, see our companion study on What Shapes Children.

Section 5

Conclusion and What We Can Do

The concentration documented in this article is structural, not accidental. The Dutch East India Company had a state monopoly. The railroads had physical limits. The dot-com era had expectation without earnings. Today's concentration is built on a real efficiency breakthrough — and is being captured by a small number of firms at a scale and speed that compounds each year. The transformation function keeps improving. The revenue base is real. The capital is being redeployed into infrastructure that will further widen the gap.

This study does not argue that technology's wealth creation is bad. It is extraordinary, and it has materially improved life for hundreds of millions of people. It does argue that we should look at the full picture: who captures the gains, what happens to the industries and workers displaced by the rotation, and what becomes of a society whose foundational infrastructure — compute, attention, search, retail — is controlled by a handful of firms.

Abundance and concentration both keep rising. The question is no longer whether technology produces wealth at unprecedented scale — that question is settled. The question is what we ask of the people and institutions that now hold structural control over so much of common life.

Appendix A

Methodology

The concentration time series uses a single broad tech definition across the full 1900–2025 period, assembled from four overlapping data sources of decreasing precision as you go further back in time.

Definitions

• Tech share %: Share of total U.S. public-equity market value held by companies in the GICS Information Technology sector plus Amazon (Consumer Discretionary), Alphabet and Meta (Communication Services), and Tesla (Consumer Discretionary) — companies classified elsewhere in GICS but functionally tech.
• Est. Tech P/E: Sector-weighted estimate of the price-to-earnings ratio for the broad tech grouping. Directional indicator of valuation regime, not a precise multiple.
• Est. Tech Revenue: Aggregate annual revenue of in-scope companies, both nominal USD and inflation-adjusted to 2025 USD using BLS CPI.

Period-by-period data quality

Companies included in the historical reconstruction

The pre-1999 reconstruction back-traces today's tech sector to its predecessor companies:

• Computing/hardware: IBM (CTR pre-1924, IBM 1924+), Texas Instruments (1939+), Hewlett-Packard (1939+), DEC (1957–1998), Intel (1968+), Microsoft (1975+), Apple (1976+), Dell (1984+).
• Telecom/comms equipment: Western Electric (Bell), Motorola (1928+), Cisco (1984+).
• Internet/software era: Oracle (1977+), Adobe (1982+), Sun (1982–2010), AOL, Amazon (1994+), Google/Alphabet (1998+), Salesforce (1999+), Facebook/Meta (2004+).

Source data

File: Tech_Concentration_Timeseries.csv / .xlsx. Five columns: Year, IT Sector %, Broad Tech %, Est. Tech P/E, Est. Tech Revenue (USD billions), Est. Tech Revenue (2025 USD billions).

• 1999–2025: S&P GICS sector data; S&P Dow Jones Indices.
• 1957–1999: Reconstructed from S&P 500 individual constituent weights for companies that would today be tech-classified. Cross-checked against Ibbotson SBBI sector returns and NYU Stern historical sector series.
• 1925–1957: Reconstructed from Dow Jones Industrial Average composition and Cowles Commission historical data; rough estimates.
• Pre-1925: Zero by construction.
• Revenue figures: Aggregated 10-K filings for in-scope companies (post-1934, when SEC filings began). Pre-1934 revenue is approximate from contemporary financial press.
• 2025-dollar adjustment: BLS Consumer Price Index, all-items, 1913–present.

Appendix B

Limitations and Open Questions

1. Pre-1957 data is reconstructive. The modern S&P 500 did not exist before 1957, and GICS sector classification did not exist before 1999. Estimates for the 1925–1957 period are derived from Dow Jones composition and contemporary press; estimates pre-1925 are zero by construction. Treat with ±3-percentage-point error bars.
2. "Tech" is a moving definitional target. The IT sector excludes Amazon (Consumer Discretionary), Alphabet and Meta (Communication Services), Tesla (Consumer Discretionary). The Broad Tech figure includes them. Either definition is defensible; both are reported.
3. The P/E figures are sector-weighted estimates and should be read as directional indicators of valuation regime, not precise multiples.
4. Revenue figures for pre-1980 are approximate. Many of the in-scope companies did not exist in modern form; predecessor entities (RCA, CTR/IBM, Bell Labs) had revenue structures that don't map cleanly to modern segments.
5. Data is static, not live. The series is committed as anchored interpolation rather than pulled from a live data provider (e.g., FMP, Bloomberg, FactSet). The chart is reproducible offline and does not depend on any API. Direct access to historical S&P GICS weights would tighten the 1999–2025 portion of the series, but the headline figures (~54% in 2025, dot-com peak comparison) are robust to the level of precision available here.

Appendix C

References

• S&P Dow Jones Indices. S&P 500 Sector Weights and Constituents, 1999–present.
• MSCI / S&P. Global Industry Classification Standard (GICS), 1999–present.
• Damodaran, A. (NYU Stern). Historical sector data archive.
• Ibbotson, R. G., & Sinquefield, R. A. Stocks, Bonds, Bills, and Inflation (SBBI) Yearbook, annual series.
• Shiller, R. J. Historical U.S. stock market data, 1871–present.
• Bureau of Labor Statistics. Consumer Price Index, 1913–present.

Appendix D

Project Files

Companion project: TechReligiousProfile/ — religious composition of tech CEO leadership and tech workforce vs. other sectors and U.S. population.