A research project that will become a book
The Next Order is a research project on one question. It will become a small book for the informed reader, perhaps with a scientific paper beside it.
The question has two halves, and the project gives them equal weight. The first is the destination: which forms of political, economic and social organization are plausible successors to today's? The second is the mechanism: what drives a transition between orders, and by which pathways could today's world move?
What the project means by an order
An order is a durable configuration of twelve rows in three dimensions: political, economic and social.1 The project speaks of a new era when most rows have changed their state across its cases and the new configuration has held for two generations, about fifty years.
Five cases and the world
The project sees the world through five cases: the United States, the European Union, China, India, and the Global South as a cluster.2 In 2023 they held 91 percent of the world's people; their share of its output was 84 percent in 2022.3 From the cases the project derives a description of the world as a whole.
- People
- Output
- United States
- European Union
- China
- India
- Global South cluster
- Outside the five cases
The five cases in the world
Share of the world's people in 2023 and of the world's output in 2022
Data The United States holds 4.2 percent of the world's people and 15.0 percent of its output; the Global South cluster holds 46.2 and 27.1 percent.
Outside the five cases are 8.7 percent of the world's people and 16.0 percent of its output.
Sources: United Nations, World Population Prospects 2024 (CC BY 3.0 IGO), and Maddison Project Database 2023 (CC BY 4.0), both through Our World in Data (CC BY 4.0). The shares are the project's own sums.
The horizons
By 2040 signposts should be visible. 2060 is the transition window: by then the direction should be clear. 2100 is the far horizon: the shape of the next order, if there is one.
What the project does not assume
The project does not assume that a new era comes. Continuity is a full scenario, and so is collapse. Several orders side by side are a live possibility, not a failure of the question. Artificial intelligence is weighed as a possible actor in the transition, not only as a tool. The project argues for no preferred order.
How orders changed before
History enters as a source of base rates and mechanisms. The project uses eight earlier transitions as its reference class, from the first farms to the international order after 1945.4
- Agricultural (Neolithic)c. 10,000–
4,000 BCE millennia - Early states and empiresc. 3500 BCE–
1 CE centuries - Universal religions and classical empiresc. 800 BCE–
600 CE centuries - Antiquity to feudal Europec. 400–
900 CE 300–500 years - Feudal to early-modern commercial and sovereign-state societyc. 1350–
1650 about 300 years - Industrial-capitalistc. 1760–
1900 100–150 years - Nation-state with mass democracy and welfarec. 1848–
1950 about 150 years for the franchise; 1914–1950 for the welfare state - Post-1945 liberal international order (sub-era)1945–
1989 about 45 years
Eight reference transitions and an open future
Approximate periods in the core region of each transition; not drawn to a uniform scale
A reading of the literature This is how the project reads the literature, not a measurement: the dates are approximate, the durations run from first clear signs to consolidation, and the readings are disputed.
A small grid marks the rows of an order that the reading names as changed most prominently; a hollow cell is a disputed one. The last row is the life of an order, not a passage into one. Beyond today the line is open: no scenario is built yet.
What the data show so far
The data strand follows long-run series for each dimension and for the drivers of change. Some are shown here. They describe the past; one carries a projection of the United Nations. None is a forecast of the project.
- Closed autocracy
- Electoral autocracy
- Electoral democracy
- Liberal democracy
Who lives under which kind of regime
Share of the world's people by regime class
Data In 2025, 45.8 percent of the world's people lived in electoral autocracies and 7.5 percent in liberal democracies, by the classification of the V-Dem project.
In an electoral autocracy people vote in multi-party elections but lack some of the freedoms that make an election fair. Reclassifications of single countries stand behind most of the recent change: India is classed as an electoral autocracy since 2017, the United States as an electoral democracy, no longer a liberal one, since 2025. People in countries without regime data are not drawn.
Source: V-Dem, Regimes of the World, with population figures, through Our World in Data. Licence: CC BY 4.0 (Our World in Data); the V-Dem data CC BY-SA 4.0.
The values as a table
| Year | Closed autocracy | Electoral autocracy | Electoral democracy | Liberal democracy |
|---|---|---|---|---|
| 1900 | 77.1 | 13.3 | 2.5 | 0.8 |
| 1910 | 76.3 | 14.3 | 2.5 | 1.0 |
| 1920 | 47.9 | 32.8 | 13.2 | 3.5 |
| 1930 | 76.8 | 3.6 | 13.0 | 4.2 |
| 1940 | 81.4 | 5.0 | 6.6 | 3.1 |
| 1950 | 71.2 | 11.0 | 9.5 | 8.0 |
| 1960 | 48.5 | 16.4 | 24.1 | 10.7 |
| 1970 | 54.4 | 10.9 | 17.8 | 16.9 |
| 1980 | 45.3 | 21.0 | 17.7 | 15.9 |
| 1990 | 38.1 | 19.9 | 26.7 | 15.3 |
| 2000 | 29.4 | 17.6 | 35.6 | 17.4 |
| 2010 | 24.2 | 25.7 | 33.4 | 16.7 |
| 2020 | 23.4 | 44.5 | 19.1 | 13.0 |
| 2025 | 27.6 | 45.8 | 19.0 | 7.5 |
- the case named above the panel
- the other cases
United States2022: 58,488
European Union2022: 38,051
China2022: 19,238
India2022: 7,766
Global South cluster2022: 9,954
Output per head
International dollars at 2011 prices, logarithmic scale
Data By these estimates output per head in the United States was 3.0 times China's in 1820, 19 times in 1950 and 3.0 times in 2022.
The values of the European Union and of the Global South cluster are means of their members, weighted by population; the cluster has estimates from 1950. For the early decades many years have no estimate; a dot is the estimate of a single year.
Source: Maddison Project Database 2023 (Bolt and van Zanden), through Our World in Data. Licence: CC BY 4.0.
The values as a table
| Year | United States | European Union | China | India | Global South cluster |
|---|---|---|---|---|---|
| 1820 | 2,674 | 1,896 | 882 | – | – |
| 1850 | 3,632 | 2,225 | 858 | 947 | – |
| 1870 | 4,803 | 2,532 | 945 | 850 | – |
| 1900 | 8,038 | 3,758 | 972 | 955 | – |
| 1913 | 10,108 | 4,517 | 985 | 1,073 | – |
| 1950 | 15,240 | 5,750 | 799 | 987 | 1,962 |
| 1980 | 29,611 | 17,950 | 1,930 | 1,495 | 4,433 |
| 2000 | 45,886 | 27,031 | 4,730 | 2,753 | 5,721 |
| 2022 | 58,488 | 38,051 | 19,238 | 7,766 | 9,954 |
- the case named above the panel
- the other cases
- projection of the United Nations, medium variant
United States2023: 1.62
European Union2023: 1.41
China2023: 1.00
India2023: 1.98
Global South cluster2023: 2.97
Births per woman
Total fertility rate
Data In 2023 the fertility rate was 1.0 in China, 1.4 in the European Union, 1.6 in the United States, 2.0 in India and 3.0 in the Global South cluster; in 1950 it had been 5.8, 2.7, 3.1, 5.7 and 6.2.
The dashed lines are the medium projection of the United Nations to 2100, not a forecast of the project. The values of the European Union and of the Global South cluster are means of their members, weighted by population.
Source: United Nations, World Population Prospects 2024, through Our World in Data. Licence: CC BY 3.0 IGO (United Nations); CC BY 4.0 (Our World in Data).
The values as a table
| Year | United States | European Union | China | India | Global South cluster |
|---|---|---|---|---|---|
| 1950 | 3.09 | 2.68 | 5.81 | 5.73 | 6.18 |
| 1975 | 1.81 | 2.08 | 3.57 | 5.20 | 5.75 |
| 2000 | 2.03 | 1.44 | 1.63 | 3.35 | 3.83 |
| 2023 | 1.62 | 1.41 | 1.00 | 1.98 | 2.97 |
| 2050* | 1.64 | 1.50 | 1.18 | 1.76 | 2.33 |
| 2075* | 1.64 | 1.55 | 1.28 | 1.71 | 2.06 |
| 2100* | 1.65 | 1.57 | 1.34 | 1.69 | 1.90 |
- notable models
- largest so far
Computing power used to train AI models
Floating-point operations in training, logarithmic scale
Data The largest training run on record used 1.45 × 1018 floating-point operations in 2007 and 9.4 × 1026 in 2026: a doubling every 7.9 months on average between these records.
The figures are estimates of the source. A dot is one notable model. The line joins the models that used more computing power in training than any model before them.
Source: Epoch AI, Notable AI Models. Licence: CC BY.
The values as a table
| Day | Model | Operations |
|---|---|---|
| 1950-07-02 | Theseus | 40 |
| 1957-01-01 | Perceptron Mark I | 6.9 × 105 |
| 1959-02-01 | Pandemonium (morse) | 6.0 × 108 |
| 1960-03-30 | Perceptron (1960) | 7.2 × 108 |
| 1987-06-06 | NetTalk (transcription) | 2.8 × 1010 |
| 1989-11-27 | Handwritten digit recognition network | 1.8 × 1011 |
| 1989-12-01 | Zip CNN | 1.5 × 1012 |
| 1992-05-01 | TD-Gammon | 1.8 × 1013 |
| 1994-12-02 | Predictive Coding NN | 1.9 × 1013 |
| 1997-11-15 | LSTM | 3.2 × 1013 |
| 2000-11-28 | Neural LM | 6.3 × 1015 |
| 2007-06-22 | SB-LM | 1.4 × 1018 |
| 2013-01-16 | DistBelief NNLM | 2.6 × 1018 |
| 2014-06-18 | SPPNet | 3.4 × 1018 |
| 2014-09-04 | VGG16 | 1.2 × 1019 |
| 2014-09-10 | Seq2Seq LSTM | 5.6 × 1019 |
| 2014-12-03 | SNM-skip | 3.0 × 1020 |
| 2015-10-01 | AlphaGo Fan | 3.8 × 1020 |
| 2016-01-27 | AlphaGo Lee | 1.9 × 1021 |
| 2016-09-26 | GNMT | 6.6 × 1021 |
| 2018-05-02 | ResNeXt-101 32x48d | 8.7 × 1021 |
| 2019-09-17 | Megatron-BERT | 2.2 × 1022 |
| 2019-10-23 | T5-11B | 3.3 × 1022 |
| 2019-10-30 | AlphaStar | 1.1 × 1023 |
| 2020-01-28 | Meena | 1.1 × 1023 |
| 2020-05-28 | GPT-3 175B (davinci) | 3.1 × 1023 |
| 2021-08-11 | Jurassic-1-Jumbo | 3.7 × 1023 |
| 2021-09-03 | FLAN 137B | 2.0 × 1024 |
| 2022-03-15 | GPT-3.5 (davinci-002) | 2.6 × 1024 |
| 2022-06-29 | Minerva (540B) | 2.7 × 1024 |
| 2023-03-15 | GPT-4 (Mar 2023) | 2.1 × 1025 |
| 2023-12-06 | Gemini 1.0 Ultra | 5.0 × 1025 |
| 2025-02-17 | Grok 3 | 3.5 × 1026 |
| 2025-02-27 | GPT-4.5 | 3.8 × 1026 |
| 2025-07-09 | Grok 4 | 5.0 × 1026 |
| 2026-09-03 | GPT-6 Astra | 9.4 × 1026 |
How the project works
The project follows rules that a reader can hold it to.
- Every citation is verified before it is used.
- Every number traces to a dataset or a model.
- Every probability is logged with a written rationale and a criterion by which it can later be scored.
- What is unknown is said.
Five strands of work feed the book: theory, data, models, scenarios and probabilities. The book joins them.
Where it stands
On 16 November 2026 the project held notes on 197 works. From them it has compared 31 theories of how orders change, catalogued 39 mechanisms of transition and profiled 36 proposed or feared successor orders. It held 96 data series. A first cycle of framing, reading, data and models is done. The phase under way builds scenarios and attaches probabilities to them.
No scenario is built and no probability is stated yet, so this page states no finding about the future. The project has no deadline and no publication date.