Loading reports … Fetching the current index of long-form research.
Loading reports … Fetching the current index of long-form research.
Winning, Losing, and Competing in the 21st Century.
World growth stepped down in the 1970s from about 5 percent to about 3, and it has stayed near 3 percent since. That step is the master variable of twenty-first-century competition. When output grew fast, growth could fund every power's new commitments at once. A power gains position mainly by allocating more of its output to force, compute, and investment than its rivals do, and on force that held when the world grew at 5 percent as it does now. What the slower world changes is the price: each new commitment takes a larger share of each year's growth. The ledger reads each power by conversion, the share of the world's military spending, AI computing capacity, and investment it commands against its share of world output. On that ledger the United States leads, conditionally; China gains selectively; Europe underallocates; and Russia buys force by spending down its future.
Power keeps a balance sheet. Its assets are output, workforce, technology, energy access, alliances, and institutional depth; its liabilities are debt, adverse demographics, weak allocation, overextended commitments, and political erosion. The entry that has changed most is the growth line at the top, and its fall changes what the rest of the sheet is worth.
World real growth averaged 5.2 percent a year from 1961 to 1973. It stepped down through the 1970s, to 4.0 percent, and has averaged close to 3 percent in every decade since: 3.0 percent in the 1980s, 2.8 in the 1990s, 3.0 in the 2000s, and 3.2 in the 2010s. From 2020 to 2024 it averaged 2.6 percent, pulled down by the 2020 contraction; 2021 to 2024, which carries the 2021 rebound, averaged about 3.9 percent, and 2023 and 2024 grew 2.9 percent each. The step came in the 1970s, and for forty years the world economy has moved within half a point of 3 percent.1
Two objections deserve an answer. About half of the step down is slower population growth: per head, world output grew 3.1 percent a year from 1961 to 1973 and about 1.7 percent from 2020 to 2024, so living standards have slowed less than the headline. Power is paid for out of the total, though, because armies, data centers, and debt service are bought with aggregate output, so the total is the right line for this ledger. The second objection is sharper: relative power turns on the gaps between powers, and China rose fastest inside the 3 percent world. The gaps have narrowed too. China grew 8.5 points a year faster than the United States in the 2000s, 5.3 points faster in the 2010s, and 2.4 points faster from 2020 to 2024. China's output is 64 percent of America's at market exchange rates. At the gap of the 2000s it would have closed the rest of that distance in about five years; at the gap of 2020 to 2024 it needs about nineteen.2 Those years hold prices and exchange rates constant, so they measure what growth alone can do. At market rates the ratio has also moved with valuation: China's output was 62 percent of America's in 2015, 77 percent in 2021, and 64 percent in 2024, as its dollar value stalled and its GDP deflator fell in 2023 and 2024.3 Valuation has recently widened a gap that growth was closing slowly. Either way, a challenger that can no longer outgrow the leader quickly gains only what it takes from it.
Conversion is the share of a world total a power commands against its share of world output. This report measures it on three entries: military spending, AI supercomputer capacity, and inflows of foreign direct investment. A power converting at par holds the same share of the entry as of world output; above par, it holds a larger share of the entry than of world output. Two of the ratios are simpler than they look. The force ratio is a power's military burden divided by the world's: the United States spends 3.4 percent of GDP against the world's 2.4 percent, a ratio of 1.4. The capital ratio records where foreign investors place their money, which is a judgment on how well a system converts, and it misses capital raised at home.
A panel from 2015 to 2024 shows how those shares moved in the slower world. The largest change in any of the four powers' shares of world output was 2.1 points, the U.S. rise from 24.1 to 26.2 percent. On force, budgets moved the European Union's share of world military spending further, from about 10.5 to 13.3 percent, while its output share fell. Russia's share rose from about 3.9 to 5.5 percent on an output share that barely moved.4 Output shares move with growth, which no government sets; military shares move with budgets, which governments do set. The ratio has a limit. The Soviet Union spent a far larger share of its output on its military than its rivals for decades and lost the competition, so conversion counts only while liabilities grow more slowly than the assets beneath them.
World military spending reached a record $2.72 trillion in 2024. The burden, 2.4 percent of world GDP, is less than half the one the world carried in 1960. What has changed is the base it lands on: in a slower world a new defense line is paid out of output that already has an owner, and defense now competes with interest.
The United States spent $997 billion in 2024, 37 percent of the world total. SIPRI estimates China's spending at $314 billion, a figure above the official defense budget because it adds military items funded outside it. Russia raised spending 38 percent to an estimated $149 billion, 7.1 percent of its GDP and 19 percent of all government spending. Germany spent $88.5 billion and Poland $38.0 billion.5 In real terms the world total rose 9.4 percent, the steepest annual increase SIPRI has recorded since the end of the Cold War.
The record is in dollars. As a share of output, the world spent 6.0 percent of GDP on its militaries in 1960 and 3.8 percent in 1973, against 2.4 percent in 2024.6 The squeeze today comes from the direction and the base: the burden has risen from its 2018 low of 2.1 percent, and it lands on budgets that growth no longer relieves.
In June 2025 NATO allies agreed to raise core defense spending to 3.5 percent of GDP by 2035.7 For the European Union's members taken together, at 1.85 percent in 2024, that means adding about 1.65 points of GDP to defense over the eleven years to 2035, which holds the growth of everything else about 0.15 points a year below the growth of the whole economy. At the postwar rate of 5 percent that would have been a thirty-third of each year's growth; at the 3 percent the world has grown since the 1980s it is a twentieth. At the 1.2 percent the European Union averaged from 2020 to 2024 it is an eighth. The commitment is the same; the growth line decides what it costs.
The United States meets the same arithmetic through its debt. On the CBO's own accounts, net interest on the public debt passed Department of Defense military outlays in fiscal 2024, $949 billion against $826 billion, and reached $1,028 billion in fiscal 2025 against $868 billion, about 1.2 times. On OMB's outlay basis, which Depth, Speed, Duration uses, net interest was $880 billion against $874 billion for all national defense in fiscal 2024; the two bases differ in what they count, and both put interest ahead. The CBO projects a federal deficit of $1.9 trillion in fiscal 2026, 5.8 percent of GDP, with debt held by the public rising from 101 percent of GDP to 120 percent by 2036, and net interest rising from 3.3 percent of GDP in 2026 to 4.6 percent by 2036.8 Every added point of interest is a point of output that cannot go to force or compute. A state can defer the trade between security and solvency for a time, and in a slower world the deferral runs out sooner.
Most of the ledger is compressing toward tighter competition. Compute is the exception: in a slower world it is the one entry where a lead still compounds, and it is widening toward one leader. Capital, the entry that prices both upside and trust, sends a clearer signal about what investors are leaving than about where they are going.
American firms attracted $109.1 billion of private AI investment in 2024, 72 percent of the $150.8 billion invested worldwide and about twelve times China's $9.3 billion. The United Kingdom drew $4.5 billion. Since 2013 the United States has accumulated $470.9 billion of private AI investment, about four times China's $119.3 billion; the European Union has drawn roughly $50 billion over the same period. The United States also holds an estimated 74 percent of tracked AI supercomputer capacity, China 14 percent, and the European Union 4.8 percent, a five-to-one lead over China that is wider than the four-to-one lead in cumulative investment.9
Two cautions bound that reading. The capacity shares count AI supercomputers that can be tracked, a tenth to a fifth of the world's computing, and the source judges China's figures unreliable, so its share may be misstated. Even with both cautions, the gap is large. Compute is an input to productivity, defense and cyber operations, and research, which is why a lead in it can compound. The bottleneck is the full stack (chips, power, capital, and the institutions that mobilize them), and a follower catches up only by spending heavily for years on worse terms. The United States converts 26 percent of world output into about 74 percent of tracked capacity, a ratio near three; China converts 17 percent into 14 percent. On this entry the distance between first and second place is widening, the opposite of what the growth line is doing.
World foreign direct investment rose 14 percent in 2025 to an estimated $1.6 trillion. Flows to developed economies climbed 43 percent to $728 billion and flows to developing economies fell 2 percent to $877 billion. The headline overstates the vote. More than $140 billion of the rise passed through a few financial centers, and net of those swings UNCTAD puts the increase at about 5 percent. Beneath the conduits the direction is clear. Inflows to China fell for a third consecutive year, by 8 percent to $107.5 billion, 6.7 percent of the world total against 17 percent of world output. Inflows to the European Union rose 56 percent to $239 billion, much of it through the same conduits. North America drew $347 billion, up 2 percent: 21.6 percent of world inflows, less than the U.S. share of world output on its own.10
The fall in China's inflows has several causes the data cannot separate: U.S. and allied export controls and investment screening, slower growth, a narrower interest-rate differential, and policy risk. Legibility may be one of them: research using night-time lights finds that autocracies systematically overstate GDP growth, and a decline in inflows is consistent with investors pricing that doubt, though nothing here isolates it.11 Whatever the mix, three years of decline in the world's second-largest economy put China furthest below par on this entry. The United States is below par on it too, which is the entry's main limit: much of America's compute buildout is financed at home and never appears in these flows, so the capital ratio understates U.S. conversion and weighs least in the verdicts.
Political stability is the least visible asset on the balance sheet and the one the others depend on. In a slower world, where each year adds less new capital to compete for, it decides whether output, compute, and capital can compound, and it erodes in financing costs and forgone investment long before it shows up in growth.
Stable states earn a premium that headline GDP does not show. Switzerland ranked first on the IMD's 2025 World Competitiveness Ranking, ahead of Singapore, Hong Kong, and Denmark; the United States ranked 13th and China 16th.12 The asset is easy to spend without anyone noticing, because its erosion is gradual and its price is paid later.
The World Bank's political-stability score for the United States, on a scale of 0 to 100, stood at 77 in 2015 and fell to 64 by 2024, in two steps: 2016 to 2017 and 2019 to 2020. It has stayed below its 2013 to 2015 range in every year since. On the underlying estimate, scaled so that the average country scores zero, the United States fell from 0.66 in 2015 to below zero in every year from 2020, at −0.11 in 2024. The indicator aggregates survey and expert judgments of the likelihood of political instability and politically motivated violence. That is one input to the trust capital prices, and only one, and none of the decline reverses the U.S. lead on output, compute, or capital depth. It does narrow the distance between the United States and the next-best places to put money.13
The score has not recovered since 2020, and nothing in the capital account yet separates its effect from rates, tariffs, and the dollar. The account does record that in a year when data centers drew more than $270 billion of announced cross-border investment, North American inflows rose 2 percent. A premium that thins without a measured price is the kind a power spends before it notices.
In a slower world each power’s gains are paid out of a thinner margin of growth, so the ledger gives four verdicts and no single scoreboard. Each is read against the same criteria: conversion on force, compute, and capital, and whether liabilities are growing faster than conversion.
Four tests, applied in order, sort the powers, and each power takes the first verdict it meets. A power is winning if it converts above par on at least two of the three entries; the condition attached names the liability that could reverse it. It is spending its position down if it converts above par on force alone, at a military burden more than twice the world's. It is gaining selectively if it converts below par but still outgrows the leader and carries a military burden below the world's, so it has both growth and room to raise conversion where it chooses. It is underallocated if it converts below par and does not outgrow the leader. The tests were written with these four powers in view, and each lands in a different category, so the sort classifies four cases; it is not an independent prediction, and a wider panel of powers is the test it has not yet faced.
The United States produces 26 percent of world output and commands 37 percent of military spending, 72 percent of private AI investment, and about 74 percent of tracked AI supercomputer capacity. It converts above par on force and compute and below par on inbound capital, which is why it leads, conditionally. The condition is on the liability side: debt held by the public at 101 percent of GDP and rising, net interest that already exceeds military outlays and is projected to keep rising, and a stability score that fell about 13.5 points in nine years. The base case is continued leadership while those liabilities grow more slowly than the asset base.
China produces 17 percent of world output and converts below par on all three entries: 12 percent of military spending on SIPRI's estimate, 14 percent of tracked capacity, and 7 percent of inflows. Market exchange rates understate what its yuan budgets buy at home, so its force ratio of 0.7 is a floor. It still grows twice as fast as the United States, at 4.9 percent against 2.5 from 2020 to 2024, and it spends 1.7 percent of GDP on its military, so it has room to raise its conversion where it chooses. That is a selective gain. Its liabilities are a population that has fallen every year since 2022, a five-to-one gap in tracked compute, and a third year of falling inflows.14
The European Union produces 17.5 percent of world output and converts below par on every entry: about 13 percent of military spending, 4.8 percent of tracked capacity, and 15 percent of inflows, a share inflated by conduit flows. It grew 1.2 percent a year from 2020 to 2024, below the U.S. rate, so it does not outgrow the leader. Its institutional depth, human capital, and single market are real assets, and it underallocates them to defense and compute. Russia produces 2 percent of world output and commands 5.5 percent of military spending, the highest force ratio on the sheet, at a burden of 6.9 percent of GDP on the World Bank series against the world's 2.4. It pays for that ratio with 7.1 percent of GDP on SIPRI's measure and 19 percent of all government spending, buying near-term force by spending down future capacity.
The four verdicts are the thesis at work, and the cost of a commitment shows why. The 0.15 points of GDP a year the European Union's members taken together need to reach NATO's core target (note 7) would have taken a thirty-third of each year's growth at the postwar rate; at the world's 3 percent it takes a twentieth, and at the Union's own 1.2 percent an eighth. The 2.8 points of world military spending the European Union added from 2015 to 2024, and the 1.6 points Russia added (note 4), were bought out of that thinner margin, so every rival that answers them pays from the same small increase.
Exhibit 5. The United States converts above par on force and compute; Russia converts on force alone
Share of world total · ratio to output share in brackets
| Output | Military | Compute | FDI inflows | Verdict | |
|---|---|---|---|---|---|
| United States | 26.2% | 36.7% (1.4×) | 74% (2.8×) | Under 21.6% (<0.83×) | Winning, on condition of debt and stability |
| China | 16.8% | 11.6% (0.7×) | 14% (0.8×) | 6.7% (0.4×) | Gaining selectively: growth and headroom |
| European Union | 17.5% | 13.3% (0.8×) | 4.8% (0.3×) | 14.9% (0.9×), conduit-inflated | Underallocated |
| Russia | 2.0% | 5.5% (2.8×) | n.a. | n.a. | Spending its position down |
Source: World Bank, WDI (output, 2024); SIPRI, April 2025 (military, 2024; EU a Power Curve estimate); Haag, FEDS Notes, October 2025 (tracked AI supercomputer capacity, as relayed from TBI, Omdia, and Epoch AI); UNCTAD, GITM No. 50 (FDI inflows, 2025; U.S. bounded by North America). n.a.: no comparable figure.
The best opposing reading uses the same evidence. On purchasing-power weights the world grows faster than 3 percent, China's military buys far more than its dollar total suggests, the burden is less than half its 1960 level, and the U.S. lead rests on one boom in one sector, so the slack this report calls gone may be merely mismeasured. The ledger concedes the second point, which is why China's force ratio is read as a floor, and the fourth, which is why the break conditions watch compute first. It answers the other two: budgets, chips, and capital cross borders at market prices, and a burden rising from a low base onto higher debt is a squeeze whatever its level.
A sharper objection is that conversion is only a definition: the force ratio is a relative burden, and nothing here shows that a higher ratio produces more power. The ledger accepts the first half and does not claim causation. It claims that, on force, position moves through allocation more than through growth, and a panel of 63 countries tests that against a faster decade. Changes in a country's military burden relative to the world's account for 85 percent of the variance in its shift in share of world military spending from 1963 to 1973, when world growth averaged 5.3 percent, and for 86 percent from 2014 to 2024, when it averaged 2.8 percent; in both decades allocation moved the share further than growth did for 50 of the 63 countries.15 On force, then, allocation was already decisive in the fast world. What the slower world adds is the price, since each allocation now takes a larger share of each year's growth. Whether a high ratio buys lasting power depends on the liabilities beneath it, which is why Russia's ratio reads as a cost and the Soviet case as a warning. In a 3 percent world each power faces one test, whether it can take share faster than its liabilities grow. Only the United States passes it today, and it passes on condition.
| Variable | What a move would mean | Reading | Falsifying trigger | Horizon |
|---|---|---|---|---|
| World real growth | Faster growth restores slack | 2.9% in 2023 and 2024 | Above 3.5% for 3 years | WB GEP, Jan 2027 |
| U.S. net interest | The U.S. condition starts to bind | 1.2× DoD military, FY2025 | Above 4% of GDP | CBO, early 2027 |
| U.S. share of AI compute | The one widening entry closes | About 74% | Below 60% | 2028 |
| U.S. stability score | The U.S. premium recovers | 64 of 100, 2024 | Above 70 | WGI, Sep 2027 |
Every figure is as published by March 31, 2026, the data cutoff, and each headline figure was reopened at its primary source, with two exceptions: North America's $347 billion of 2025 inflows is carried as UNCTAD's monitor reports it, and the compute-capacity shares are carried as the Haag FEDS Note relays them from the Tony Blair Institute, Omdia, and Epoch AI. World Bank series were read in their September, October, and November 2026 vintages; values first published after the cutoff, such as the 2025 governance score, are not used, and UNCTAD's mid-2026 revision of 2025 flows is set aside on the same rule. The two treatments differ for a reason: the World Bank vintages revise years already in the panel, while UNCTAD's revision replaces the projected 2025 values the ledger scores, so taking it would mix two cutoffs inside one exhibit. Growth rates are real, at market exchange rates, and averaged over periods from annual data; market rates are the right weights for this ledger because budgets, chips, and capital cross borders at market prices. Conversion ratios divide a power's share of a world total by its share of world output at market rates in 2024: military spending is for 2024 (SIPRI, current dollars; the EU share is the World Bank burden times EU GDP, since SIPRI publishes no EU aggregate), compute is tracked AI supercomputer capacity through July 2025, an estimated 10 to 20 percent of the world's aggregate, and FDI inflows are UNCTAD's 2025 projections from three quarters, with the U.S. share bounded by the North American figure. The 2015 comparison in Part I uses the World Bank burden and output series for both years. Interest and defense are compared on the CBO's own outlay categories for the same fiscal years. The ratios describe where each power stands and how far a decade of budgets moved it; they do not establish that a higher ratio causes a gain in power. The 2025 FDI split and announced greenfield values are given in Part III, note 10, and Tables A4 and A5.
| Dataset | Source | Frequency | Coverage | Retrieved | Used in |
|---|---|---|---|---|---|
| World real GDP growth | World Bank, WDI | Annual | 1961 to 2024 | Oct 2026 vintage | Exhibit 1, Table A1 |
| Growth by power | World Bank, WDI | Annual | 2000 to 2024 | Oct 2026 vintage | Part I, Table A1 |
| Military expenditure | SIPRI Military Expenditure Database | Annual | 2024 | Apr 2025 | Part II, Exhibit 5, Table A2 |
| Military burden | World Bank, WDI (SIPRI) | Annual | 1960 to 2024 | Oct 2026 vintage | Exhibit 2, Parts I and V |
| Private AI investment | Stanford HAI, AI Index Report 2025 | Annual | 2013 to 2024 | Apr 2025 | Exhibit 3, Table A3 |
| AI supercomputer capacity | Haag, FEDS Notes | One-off | Through Jul 2025 | Oct 2025 | Parts III and V, Exhibit 5 |
| FDI and greenfield value | UNCTAD, GITM No. 50 | Annual | 2024 to 2025 | Jan 2026 | Exhibit 5, Tables A4 and A5 |
| Political stability score | World Bank, WGI | Annual | 2013 to 2024 | Sep 2026 vintage | Exhibit 4, Table A6 |
| U.S. interest and defense outlays | CBO, Monthly Budget Review | Annual | FY2023 to FY2025 | Nov 2025 | Parts II and V |
| U.S. deficit, debt, interest | CBO, Budget and Economic Outlook | One-off | 2026 to 2036 | Feb 2026 | Parts II and V |
| Output by power | World Bank, WDI | Annual | 2015 to 2024 | Oct 2026 vintage | Part I, Exhibit 5 |
World growth by year, 2020 to 2024: −2.9, 6.5, 3.4, 2.9, and 2.9 percent. World output per head grew 3.1 percent a year in 1961 to 1973 and 1.7 percent in 2020 to 2024. The European Union averaged 1.2 percent in 2020 to 2024.
| Period | World | China | United States |
|---|---|---|---|
| 1961 to 1973 | 5.2 | ||
| 1970s | 4.0 | ||
| 1980s | 3.0 | ||
| 1990s | 2.8 | ||
| 2000s | 3.0 | 10.4 | 1.9 |
| 2010s | 3.2 | 7.7 | 2.4 |
| 2020 to 2024 | 2.6 | 4.9 | 2.5 |
Source: World Bank, World Development Indicators, series NY.GDP.MKTP.KD.ZG and NY.GDP.PCAP.KD.ZG, October 2026 vintage; Power Curve analysis. Period averages of annual rates. Behind Exhibit 1.
The United States accounted for 37 percent of the world total. Russia's figure is at market exchange rates, a 38 percent real rise on 2023, 7.1 percent of GDP, and 19 percent of government spending.
| Country | Spending |
|---|---|
| United States | 997 |
| China, SIPRI estimate | 314 |
| Russia, SIPRI estimate | 149 |
| Germany | 88.5 |
| Poland | 38.0 |
| World | 2,718 |
Source: SIPRI Military Expenditure Database and press release, April 2025, current dollars. Part II and Exhibit 5.
The U.S. share of world private AI investment is 72.4 percent, and the U.S. to China ratio is 11.7 times. From 2013 to 2024 the United States accumulated $470.9 billion and China $119.3 billion; the European Union drew roughly $50 billion.
| Economy | Investment |
|---|---|
| United States | 109.1 |
| China | 9.3 |
| United Kingdom | 4.5 |
| World | 150.8 |
Source: Stanford HAI, AI Index Report 2025, chapter 4; Haag, FEDS Notes, October 2025, for the EU figure. Behind Exhibit 3.
More than $140 billion of the world increase passed through financial centers; net of those flows UNCTAD puts the increase at about 5 percent. The monitor does not list the United States separately.
| Destination | $ billion | Share of world | Change |
|---|---|---|---|
| Developed economies | 728 | 45.3% | +43% |
| Developing economies | 877 | 54.6% | −2% |
| North America | 347 | 21.6% | +2% |
| European Union | 239 | 14.9% | +56% |
| China | 107.5 | 6.7% | −8% |
| World | 1,606 | 100% | +14% |
Source: UNCTAD, Global Investment Trends Monitor No. 50, January 2026, annex tables 1 and 2. 2025 is projected from three quarters. Part III and Exhibit 5.
Data centers were more than one-fifth of all announced greenfield value; the leading hosts were France, the United States, and the Republic of Korea.
| Sector | Value | Change |
|---|---|---|
| Data centers | 270+ | |
| Semiconductors | 166 | +35% |
| Renewable energy | 197 | −28% |
| Extractive industries | 33 | −22% |
Source: UNCTAD, Global Investment Trends Monitor No. 50, January 2026. 2025 is projected from three quarters. Part III.
| Year | Score | Year | Score | Year | Score |
|---|---|---|---|---|---|
| 2013 | 75.6 | 2017 | 67.9 | 2021 | 63.6 |
| 2014 | 77.2 | 2018 | 70.9 | 2022 | 62.9 |
| 2015 | 77.1 | 2019 | 66.8 | 2023 | 63.1 |
| 2016 | 71.5 | 2020 | 63.4 | 2024 | 63.6 |
Source: World Bank, Worldwide Governance Indicators, "Political Stability and Absence of Violence/Terrorism," governance score, series GOV_WGI_PV.SC, September 2026 vintage. Behind Exhibit 4.
Hart, Brendan. 2026. “The Arithmetic of Power.” The Power Curve, October. https://brendanhart.com/reports/arithmetic-of-power.