What is this?

A long-range population scenario tool for 192 sovereign nations. It follows official UN World Population Prospects 2024 data through 2100, then uses our own stochastic extension model to explore what different long-run fertility assumptions would imply after the UN horizon.

Data Sources

• 2024-2100: UN World Population Prospects 2024 — crude birth rates, crude death rates, survival ratios from life tables, net migration for all 237 countries. Our implementation tracks the UN medium path within about 1.6% at tested reference years.
• Post-2100: Custom long-range scenario engine using an AR(1)-style process on crude birth and death rates, age-structure-sensitive mortality, migration decay, and stylized shocks. 100 TFR scenarios (0.1-10.0) were computed with 10M trajectories each on 8× A100 GPUs.
• Language data: CIA World Factbook, JovianHQ languages dataset
• Religion data: Pew Research Center Global Religious Composition

Methodology

The core model is a cohort-component projection in the same general family of methods used by the UN, but it is notthe UN's full internal projection system. Population is tracked in 21 five-year age groups by sex for each country. Each 5-year step:

1. Each age cohort survives and moves up one group (using UN life table survival ratios)
2. Births are computed from crude birth rates applied to total population
3. New births enter the 0-4 cohort with infant survival applied
4. Net migration is distributed across working-age groups

After 2100, the public TFR slider is converted into a simplified crude-birth-rate target rather than a full age-specific fertility projection. That makes the post-2100 results scenario illustrations, not an official extension of the UN medium variant.

Scenarios

We compute 100 Total Fertility Rate (TFR) scenarios from 0.1 to 10.0 in 0.1 increments, each with 10 million independent trajectories — 1 billion total across the full sweep. The interactive homepage focuses on the most interpretable public range and shows scenario bands (5th to 95th percentile) rather than a single forecast line.

Key presets:

• Continued Decline (TFR 0.7): Illustrative low-fertility scenario in which global fertility keeps falling past current trends.
• East Asia levels (TFR 1.0): Illustrative scenario anchored to the current neighborhood of South Korea (0.72), Japan (1.22), and China (1.01).
• European levels (TFR 1.4):Illustrative low-fertility scenario near today's European average.
• UN Medium Variant (TFR 1.84): UN WPP 2024 through 2100, then our custom long-range tail. In the published median path, population peaks at ~10.3B (2084), then declines to ~2.7B by 2499.
• Replacement (TFR 2.1): Illustrative scenario at exact replacement fertility. Population stabilizes long-term.
• Multi-Planetary (TFR 3.0): Purely illustrative high-fertility stress case rather than an evidence-based space forecast.

Compute

All projections computed on 8× NVIDIA A100-SXM4-80GB GPUs (640GB total VRAM) via the Microsoft for Startups GPU Cluster (Azure ML, Italy-North region). Global TFR sweep: 100 scenarios × 10M trajectories = 1 billion total (~4 hours). Per-country TFR sweep: 100 scenarios × 237 countries × 1M trajectories with adaptive per-country histogram resolution (~82 minutes). All code runs on PyTorch with CUDA.

Key Finding

Under the UN-medium-like assumption (TFR converging to about 1.84 after the UN horizon), the long-range extension points toward a much smaller world after the 21st-century peak. In that scenario family, crossing below 1 billion occurs well after the modeled 2500 horizon and should be read as a directional extrapolation, not a precise forecast date.

A large share of humanity already lives in below-replacement-fertility countries. That matters for long-run age structure and population momentum, but it does not determine a single future path on its own. Migration, mortality, education, policy, and regional divergence still matter.

Limitations

• Anything after 2100 is our custom scenario model, not an official UN projection.
• The post-2100 model uses a simplified CBR-based fertility heuristic (`CBR ≈ TFR × 5`), not age-specific fertility schedules or the UN's full bayesTFR framework.
• Migration is assumed to decay toward zero after 2100, which is a strong simplifying assumption for country-level paths.
• Uncertainty bands are outputs of this model's assumptions, not formal confidence intervals over all plausible futures.
• Dates shown after 2500 are extrapolated directional thresholds, not direct simulation output.
• Small countries and bloc aggregates have more approximate long-range values because of histogram resolution and aggregation shortcuts.
• The model includes only stylized shocks and omits many structural drivers, including education, policy responses, major wars, and transformative technologies.
• Upstream UN inputs also carry uncertainty, especially where recent census or register data are sparse. Only 114 of 237 UN countries/areas had post-2019 census data at the time of the WPP 2024 release.
• Post-2100 births are computed from total population × CBR, not from women aged 15-49 or age-specific fertility schedules. This loses the population-momentum effect that drives real demographic outcomes.
• Bloc-level percentiles (EU, BRICS, Africa, etc.) are summed from member-country percentiles, which is not statistically rigorous. The quantile of a sum is not the sum of quantiles.

Tech Stack

Next.js 16 + TypeScript + Tailwind CSS + SQLite (better-sqlite3) + Canvas 2D charts. GPU simulations in PyTorch (Python). Data pipeline: UN CSV → Python processing → SQLite → Next.js.