Last year China had 7.92 million births, the fewest since 1949. In the same year it installed 295,000 industrial robots, more than the rest of the world combined, and for the first time in history overtook Japan as the largest robot maker in the world. These two numbers are more connected than they look.
Because demography doesn't have to be only about how many people there are. It can also be about the ratio of the productive to the unproductive.
Baseline
The best mainstream baseline is the UN's World Population Prospects 2024. The world has about 8.2 billion people today. The UN expects a peak of around 10.3 billion in the mid-2080s, then a slow decline to about 10.2 billion in 2100.
More interesting than the median is the spread. The UN puts 2100 at about 10.2 billion. The Lancet (IHME) expects more like 8.8 billion. The difference between the two scenarios is roughly all of today's China. That's the difference between two completely different worlds, and both scenarios are made by professionals. So where does the spread come from?
The key variable is fertility
Almost all of the spread comes from how fertility develops. Globally it has fallen dramatically in a few decades, and today it's about 2.25 children per woman. In its medium variant the UN assumes that some of the countries with extremely low fertility will at least partly bounce off the bottom over time. The last few years show, though, that the ultra-low regime is sticky, and the inertia is tricky.
And pronatalism? Historically it moves tenths, not whole children. Europe has talked about low birth rates for decades and nothing happens in the data: a fertility rate of 1.34 in 2024 and 3.55 million births, the fewest since 2001. Singapore runs the most aggressive pronatalist policy in the world, and last year its fertility fell to a new low of 0.87. Nobody has shown yet how to get from 1.3 to 2.1. The numbers say we don't know how.
And meanwhile the window is closing. Once the cohort of women of reproductive age shrinks, pronatalism loses any leverage at all. Even if it started working in a few years, in many countries it's already too late to rebuild a large population.
Three pivots
- Africa. Key from a global point of view, because that's where there's the most room for decline. Sub-Saharan Africa's fertility is still high, around 4.3, but the trend is down. If that decline speeds up, with urbanisation, girls' education, healthcare and contraception, the global curve falls significantly lower. This is where it's decided whether we get a world of 8.8 or 10.2 billion.
- India. Fertility has been falling since the 1960s and according to UNFPA is now at 1.9, below replacement. The population is about 1.46 billion, peaking at around 1.7 billion in roughly 40 years, then declining.
- China. The population fell for the fourth year in a row, last year by 3.39 million to 1.405 billion. Births are at their lowest since 1949. China's biggest problem is the total cost of a child: housing, education, work pressure, uncertainty.
And now the other side
An economy's output is roughly the number of people working times their productivity. When the workforce falls by some percent a year and productivity grows by another, the net effect is the difference.
We have one 25-year natural experiment: Japan. Its labour force peaked in 1995 at 87 million and by 2024 had fallen 16% to 74 million. The ratio of old people to working people more than doubled, from 21 to 49%, and it's heading for 74% by 2060. Brutal starting conditions.
And what happened? At the level of a single worker, Japan held up better than expected. Capital per worker grew, there are more machines per head, and growth in productivity per hour worked stayed roughly at Europe's level. That's the optimistic mechanism in practice: automation partly covers the missing hands.
But the same Japan is also a warning. In dollars, GDP per head fell, and Japan dropped below the G7 average for the first time since the 1980s. And hourly productivity growth didn't speed up, it slowed, from about 1.2% to about 0.3%. So Japan half confirms the thesis and half doesn't.
Where it hurts most
Not in aggregate GDP. That's manageable. The problem is elsewhere.
- Fiscal maths. A dependency ratio of 49%, heading for 74%, means fewer and fewer people paying for more and more pensioners and patients. That's a ticking bomb, not “a bit less gross product”.
- Who owns the robots. When capital replaces labour, the benefit flows to the owners of capital, not to payslips. But pensions and healthcare have historically been paid for from taxes on labour. A smaller, ageing population plus productivity captured in capital, in robots, means the erosion of what the whole thing is financed from.
On top of that, I wouldn't play down social cohesion during the transition, or the geopolitics of young against ageing regions.
Why do I still think that this time it will show up in the data, and positively? The capital intensity of automation is falling, robotics and running AI models are getting cheaper by orders of magnitude, and cheaper energy from new sources will add something too.
Conclusion
Back to the beginning. Demography is neither good nor bad. It's a vector.
People love black-and-white narratives because they share well. History is merciless to them, though. Look at the printing press, or at manufacture. When the printing press came, scribes rightly feared for their craft, but within 50 years of Gutenberg more than 9 million books were produced in Europe. A luxury for a chosen few became the medium that literacy and science stand on today.
In the same way, weavers smashed mechanical looms because the machines took their work, and their fear was justified: the number of hand weavers in Britain fell from about 250,000 to a few thousand. But the work didn't disappear, it moved. First into mechanised factories, where the machines themselves went from 2,400 in 1813 to 250,000 by 1850, partly because cheap cloth pushed demand for clothing up. Then into industries the revolution had only just created: railways, engineering, steel, building cities. Jobs that hadn't existed a generation earlier.
Welcome to the new world.