How much does people still offline cost?
people still offline (% of people) · % of people · measured 2005–2025 · fitted live from 21 observations by the Many Minded cost-curve engine.
The last measured value is 26.4% (2025, ITU ICT Indicators (via World Bank / OWID)). Over the fitted window the series falls 7.9% a year (80% interval −9.0% to −6.8%) — a halving every 8.4 years (80%: 7.4–9.9). Carried forward on that fit, 2030: 17.5% (80%: 15.8%–19.4%).
the curve
what this series measures
share of the WORLD population NOT using the internet (100 − ITU individuals-using-the-internet, world aggregate) — not coverage, not subscriptions.
the gate — what crossing this threshold changes
Central fit: ~2045. 80% window: 2043–2050 · n=11. That is 2.4 halvings away at today's value.
what the fit says
| quantity | value | how it is computed |
|---|---|---|
| fitted rate | −7.9%/yr (80%: −9.0% to −6.8%) | Farmer–Lafond drift over the trailing window · n=11 points, 10 years |
| halving time | 8.4 years (80%: 7.4–9.9) | implied by the fitted drift |
| curve check | accelerating (−3% → −8%/yr) | first half of the record versus the second, judged against the direction that helps this metric (falling) |
| regime break | ~2017 (−4% → −9%/yr, F=206) | Chow-style best single breakpoint, kept only above a conservative sup-F threshold |
Not shown for this curve, because the machinery returns nothing: Wright's law (no cumulative-deployment series for this technology).
the projection, with its interval
Log cost as a random walk with drift: the forecast variance grows with horizon (τ + τ²/m), which is why these bands widen instead of staying parallel. The middle column is the least useful number on this page; the interval is the claim.
| year | central fit | 80% interval | |
|---|---|---|---|
| 2026 | 24.3% | 23.4% to 25.3% | near horizon |
| 2027 | 22.4% | 21.1% to 23.7% | near horizon |
| 2028 | 20.6% | 19.1% to 22.2% | near horizon |
| 2029 | 19.0% | 17.4% to 20.8% | the band is already wide here |
| 2030 | 17.5% | 15.8% to 19.4% | the band is already wide here |
| 2031 | 16.1% | 14.3% to 18.1% | the band is already wide here |
| 2032 | 14.8% | 13.0% to 16.9% | far horizon — read the interval, not the middle |
| 2033 | 13.7% | 11.8% to 15.8% | far horizon — read the interval, not the middle |
Projected to 2033, the horizon the whole engine uses.
questions this page answers
How much does people still offline cost?
26.4% as of 2025, the latest measured value in the series (ITU ICT Indicators (via World Bank / OWID)). The fitted trend has it falling 7.9% a year, with an 80% interval of −9.0% to −6.8%.
How fast is people still offline falling?
−7.9% a year over the fitted window, an 80% interval of −9.0% to −6.8% — a halving every 8.4 years (80%: 7.4 to 9.9 years). Fitted from 11 observations spanning 10 years.
What will people still offline be in 2030?
The central fit says 17.5%, inside an 80% interval of 15.8% to 19.4%. The interval is the forecast; the middle number is only its midpoint. Bands widen with horizon because shocks accumulate — a constant-width band would be overconfident.
When will people still offline reach ≤ 5% of people offline?
The central fit says ~2045, with an 80% window of 2043–2050 from 11 observations. Crossing it is what the engine calls "universal internet use (ITU’s ≥95% bar)", firing: free information and free intelligence in every hand. Treat the window, not the year, as the claim.
Is the decline in people still offline accelerating or slowing?
Splitting the record in half, the fitted rate went from −3% to −8% a year — accelerating. A Chow-style test finds a regime break around 2017 (−4% → −9% a year, F=206). Regime breaks, not window choice, are what dominate this method's errors — which is why every projection here carries an interval.
Where does this data come from?
ITU ICT Indicators (via World Bank / OWID). 21 observations spanning 2005–2025. Curated benchmark history, extended by a weekly authoritative fetch and by news figures fact-checked against their source before they may touch a fit. Both the observation ledger and the fitting code are public, and the engine publishes its own calibration score and its misses.
the raw numbers
Every observation behind the fit, unrounded by us and unsmoothed. This table is here on purpose: graphs make people underestimate exponential change, and the raw series beside the curve is the one correction shown to work.
| year | % of people | change |
|---|---|---|
| 2005 | 84.4% | — |
| 2006 | 82.8% | −1.9% |
| 2007 | 79.8% | −3.6% |
| 2008 | 77.2% | −3.3% |
| 2009 | 74.7% | −3.2% |
| 2010 | 71.6% | −4.1% |
| 2011 | 69.1% | −3.5% |
| 2012 | 66.7% | −3.5% |
| 2013 | 64.7% | −3.0% |
| 2014 | 62.6% | −3.2% |
| 2015 | 60.1% | −4.0% |
| 2016 | 56.4% | −6.2% |
| 2017 | 53.7% | −4.8% |
| 2018 | 50.6% | −5.8% |
| 2019 | 46.1% | −8.9% |
| 2020 | 39.9% | −13.4% |
| 2021 | 36.2% | −9.3% |
| 2022 | 33.0% | −8.8% |
| 2023 | 30.8% | −6.7% |
| 2024 | 28.8% | −6.5% |
| 2025 | 26.4% | −8.3% |
where this comes from
- Source: ITU ICT Indicators (via World Bank / OWID) — refetched weekly by the engine's benchmark cron.
- Basis: measured. Curated benchmark history, refreshed by the weekly authoritative fetch, extended by figures our daily editor extracts from the news — each of which is fact-checked against its source article before it may touch a fit.
- Observations: n=21, spanning 2005–2025. Ledger last updated 2026-09-07.
- Fit: the Farmer–Lafond stochastic model — log cost as a random walk with drift over a 10-year trailing window, Student-t tails, and (for series under 15 points) a volatility floor at their drift-volatility prior. The code is readable source; the observation ledger is public JSON.
- Reuse: this compiled series and the numbers on this page are published under CC BY 4.0 — take them, cite the page. The underlying source keeps its own terms (Our World in Data publishes under CC BY 4.0).
- Accuracy: the engine grades its own intervals in public — every feasible hindcast re-fitted on data as it stood, scored for whether the actual landed inside the 80% band. The calibration score is on the engine page, misses included.
the rest of the graph
Other CONNECTION curves:
Recent briefs from CONNECTION, the daily record of what moved:
- Ukraine tests its own small comms satellite after Starlink scare 2026-10-29Himera and GomSpace are testing Obria Nano in orbit, a first step toward a Ukrainian-controlled communications constellation.
- EU backs a Greenland subsea cable to route around US and Red Sea chokepoints 2026-10-29The Pan-Arctic Cable would be the first direct Europe-Greenland link, with possible later extensions to Canada and Asia. Only the first leg is in development.
- IEEE Smart Cities Program Expands Technical Collaboration for Urban Services 2026-10-18City governments increasingly use data-driven approaches to improve traffic and waste management through IEEE's technical framework.
- Data Center Permitting Test Case Targets Internet Affordability Barriers 2026-10-18Permitting delays for large data centers could be a key test case for reducing internet access costs
- Iran targets Starlink as conflict-era internet curbs emerge 2026-10-18Iran’s move to restrict Starlink access during active conflict signals growing friction in global connectivity
- Mobile Money in Africa: Access Gained, Investment Limited 2026-10-18Mobile money systems in Africa improve financial access but don't enable significant capital building for local populations.
Every tracked curve: utility solar, installed · battery pack price · battery cell price · solar module price · disk storage · human genome sequencing · transistors per chip · fastest supercomputer · DRAM memory · solid-state storage · onshore wind electricity · solar electricity (LCOE) · space launch, to LEO · wheat yield · offshore wind electricity · AI task horizon (METR) · frontier AI training compute · modular housing, turnkey · frontier-grade intelligence · AI video, finished minute · humanoid robot, capable unit · cheapest humanoid, list price · farm labor share · healthcare spend per person
← the whole engine: build clock, cascade, calibration · the archive · how we know