The Math Nobody Wants to Do

I've been building a projection model for the last two months.

Not from vibes. Not from Twitter threads. Not from "AI will change everything" platitudes that mean nothing.

From demographic data. Fiscal math. Historical transitions. AI cost curves. And the developing world angle that almost every Western analyst completely ignores.

I'm going to walk you through what I found. Some of it surprised me. Some of it scared me. And one part of it made me genuinely angry.

Here's what the next 20 years actually look like.

This Has Happened Before

Before I show you the projection, I need to show you the pattern. Because this has happened before.

Not with AI. But the structural dynamics — a new technology displacing the primary source of human economic value — have played out multiple times.

And every time, it played out the same way.

Fifty Years of Getting Poorer

During the first Industrial Revolution, productivity surged.

Wages didn't.

For fifty years.

Between 1780 and 1840, output per worker rose 46%. Real wages rose 12%.

The wealth went to the people who owned the machines. Workers got poorer relative to what the economy was actually producing.

Let that sit for a second. Fifty years. Two entire generations of people watched the economy grow while their lives didn't improve.

That's not a recession, not a rough patch. That's a structural betrayal.

Economists call this the Engels' Pause. I call it the closest thing we have to a preview of what's about to happen.

Eventually — around 1840 — it resolved.

Capital accumulation caught up with the technology. Wages exploded. Between 1840 and 1900, workers captured more than their share of the productivity gains.

But it took half a century.

If you're an optimist, modern institutions compress that timeline. Safety nets. Democratic feedback loops. Faster information. Maybe 15-20 years instead of 50.

If you're a realist, you look at how governments handled 2008, COVID, and climate — and you wonder if 15-20 years is wishful thinking.

I'm using 15-25 years as my working range. I hope I'm wrong on the high end.

The Part That Should Scare You

Between 2001 and 2018, the US lost about 353,000 manufacturing jobs per year to Chinese competition. The "China Shock."

Here's the thing. The macro economy healed. Economists said it worked. Non-manufacturing sectors absorbed the workers by 2011. GDP kept growing. The charts looked fine.

The people didn't heal.

The affected communities never recovered. Not in 5 years. Not in 10. Not in 17.

And the political consequences — populism, Trump, Brexit — didn't emerge until 15 years after the initial displacement.

Fifteen years.

This is the part that should scare you. Not the displacement itself. The delay. The fact that the statistics will say "everything's fine" for a decade while millions of people live through something the numbers don't capture.

And then the politics explodes.

Remember this when you see headlines in 2030 saying "AI transition proceeding smoothly."

Everything Is Faster Now

One more piece of context.

Electricity took 93 years to go from invention to 50% adoption.

Cars took 40.

Broadband took 9.

Smartphones took 5.

AI is on the smartphone curve.

Enterprise adoption hit 78-88% within three years of ChatGPT.

So take every historical timeline I just described — the 50-year Engels' Pause, the 17-year China Shock — and compress it.

Same dynamics. Faster.

That either means the resolution comes faster too. Or the damage accumulates faster than anyone can respond.

I think it's the second one.

I'd love to be wrong.

The Collision

OK. Now the projection.

Here's where the 2030s get truly weird. Two forces are about to collide that have literally never collided before in human history.

Force 1: AI displaces working-age cognitive labor. Fewer workers needed for the same output.

Force 2: The population ages. Fewer workers available to support more retirees.

Read that again.

The economy simultaneously needs fewer workers AND has fewer workers.

On paper, that sounds like it solves itself. AI fills the gap. Nobody gets hurt. Clean transition.

In practice, it doesn't work like that.

Because the workers AI replaces aren't the workers who are retiring.

AI replaces 35-year-old analysts, writers, developers, marketers. Retirement removes 65-year-old workers across all sectors. The mismatch is total. Displacement still happens even as the overall labor pool shrinks.

So what do developed countries actually face in the 2030s?

More retirees drawing benefits. Fewer knowledge workers paying taxes. Social Security's trust fund depletes in 2033.

That's not my projection — that's the Social Security Administration's own math. Federal debt crosses 118% of GDP by 2035.

And these numbers don't even account for AI shrinking the tax base.

Labor's share of GDP is already at 53.8%. Historic low. Heading toward 48-50%.

If that happens, income tax revenue projections are too optimistic. The fiscal math breaks from both directions at once.

More people needing money from the government. Less money going to the government.

At the same time.

Nobody in Washington is seriously planning for this.

A Billion People With No Escalator

This is the part that makes me angry.

Because almost nobody in Silicon Valley or Washington or Brussels is talking about it. And it's the most consequential dimension of the entire AI transition.

While the developed world ages, Africa's working-age population explodes.

By 2040, Africa will have 1.8 billion working-age people. That's more than India and China combined.

After 2040, sub-Saharan Africa is the only region on earth where the labor force is still growing.

Now. Normally, a massive young population is a gift. It's exactly what powered East Asia's economic miracle from 1960 to 2000. South Korea. Taiwan. China. Young workers. Rising productivity. Export-driven growth. Economists call it the demographic dividend.

Africa should be next.

Except every escalator that previous countries rode to the middle class has been removed.

The manufacturing escalator is gone.

China already dominates. Automation is making it even harder for new entrants. The window that lifted East Asia shut behind them.

The services escalator is breaking.

India's IT sector — $200 billion in exports, 5.4 million workers — was the service-economy version of a factory floor.

The Philippines' BPO industry employs 1.82 million people. Eight percent of the entire country's GDP.

These were the pathways. From poverty to middle class. For millions of real people.

When an AI coding agent costs less than an Indian developer, that pathway narrows.

When an AI handles customer service calls better than a Manila call center, that pathway disappears.

So by the 2030s you have this: 1.8 billion people. Majority young. In a world where neither manufacturing nor services outsourcing offers a viable way up. No fiscal cushion. Limited infrastructure. Growing population.

This is not about whether American software engineers find new jobs. That's a rich-world problem with rich-world solutions.

This is about whether a billion young Africans have any economic pathway at all.

I work in cross-border founder mobility. I sit across the table from people navigating these systems every day.

I've watched the doors narrow in real time. I've seen what happens when a country's development model breaks and nobody has a replacement.

It's not theoretical to me.

The migration pressure this creates is enormous. The UN projects peak international migration in 2040-2045. Most of it flowing from young, economically constrained regions toward aging, labor-short economies.

Whether that's managed as an opportunity or a crisis depends entirely on politics.

And if the China Shock is any guide — where political consequences arrived 15 years late and hit like a freight train — politics won't handle it well.

The Cost Curves Don't Lie

Let's talk about what AI can actually do by 2035. Because this is the most predictable part of the entire projection.

Inference costs dropped 280x between November 2022 and October 2024.

I'll say that again. 280x. In two years.

GPT-level performance went from $20 per million tokens to $0.40. That's 50x cheaper in three years.

By 2030, inference is commodity-priced. The marginal cost of AI thinking converges with the cost of electricity.

By 2035, running an AI agent that does the cognitive work of a $100K-a-year knowledge worker probably costs less than $500 a month. Maybe less than $100.

At that point, the economic argument for hiring humans to process information is just... gone. Not weakened. Not pressured. Gone.

The same way the argument for hiring humans to do arithmetic disappeared when calculators showed up.

And on capability — by 2030, AI handles 70-80% of current knowledge work. Not perfectly. Not creatively. But well enough that most businesses don't need a human for it.

By 2035, that's 85-90%.

The remaining 10-15% is genuinely creative, deeply relational, or requires physical presence.

That's it. That's what's left.

Then Comes the Physical World

The cognitive disruption comes first. The physical disruption follows with a 5-10 year lag.

Tesla's Optimus. Figure AI partnering with BMW and Hyundai. Humanoid robots at scale by 2028-2032.

Why this matters: the gig economy and in-person service work — which is where the first wave of displaced knowledge workers will land — starts facing its own automation by the mid-2030s.

The "I'll just do something with my hands" fallback has an expiration date.

Not immediately. But it's coming. And it means the physical world isn't a permanent refuge from cognitive displacement. It's a temporary one.

Two Futures

By the mid-2030s, the early disruption has settled into a new baseline.

GDP is growing. Probably faster than before. AI productivity gains are real.

But the growth goes to capital and compute. Not labor. I think of this as Ghost GDP — output that shows up in the statistics but doesn't circulate through households.

Former $150K knowledge workers are earning $60-80K in hybrid roles. Not unemployed. Underemployed. The median household feels poorer even as the economy technically expands.

This is where the road forks.

Future A: We Figure It Out

40% probability. I want to be wrong about those odds.

Policy response is late but adequate. AI-funded transfers. Compute taxes. Immigration reform. New institutions.

New job categories have matured — AI operations, human verification, experience design, care work. Employment stabilizes, even if average income is lower in real terms than 2025.

The twist: real purchasing power might actually improve. Because AI has cratered the cost of healthcare, education, legal services, food. You earn less but everything costs less too.

Migration pathways between young, labor-surplus countries and aging, labor-short ones have been formalized. Messy. Political. But functional.

AI-driven scientific breakthroughs start landing. Drug discovery timelines go from 10-15 years to 3-5. New materials. Better batteries. Real new economic activity.

This is the version where the Engels' Pause lasts 15 years instead of 50.

Future B: We Don't

60% probability.

Structural unemployment at 7-10%.

But underemployment is the bigger story — 20-30% of workers in roles significantly below their skill level. The downshift never reverses for the first generation.

No real transfer system. Piecemeal programs. Extended unemployment benefits that don't address a structural shift.

The top 5% captures an absurd share of output. The bottom 60% gets by on AI-driven cost reductions.

Debt-to-GDP past 130% and climbing. No political consensus on anything.

Then robotics hits the service sector around 2035-2039. The in-person jobs that absorbed displaced knowledge workers start getting automated. Second wave of displacement. This time hitting people who thought they were safe.

Populism doesn't peak. It hardens.

Why is this more likely? Because the Engels' Pause lasted 50 years. Because the China Shock's affected communities never recovered.

Because political response to the Industrial Revolution took 60-80 years. Chartist movement didn't emerge until 80 years after mechanization started.

Modern institutions are faster. But they're not 10x faster.

The honest expectation is a messy, prolonged transition. Not a clean landing.

What Survives

Regardless of which future plays out, by the 2040s the dust settles on a clear picture. Five kinds of work remain.

Creators.

People who use AI to build things that couldn't exist before. Small number. High income. Like artists who mastered the printing press in the 1500s.

Taste and judgment.

Humans who decide what matters. Editors, curators, strategists. AI generates the options. Humans choose.

Relationship and trust.

Doctors, advisors, negotiators, leaders. Where the human connection IS the product. AI assists. But you can't automate the handshake.

Physical presence.

Care workers, trades, hospitality, emergency services. Partially automated. Still significantly human through 2045.

Maintenance.

Keeping AI systems and robots running. Less like an old IT job. More like being a power plant operator.

Everything else — information processing, analysis, writing, coding, customer service, routine legal, routine medical, routine financial — is AI. Not partially. Fully.

The Real Challenge Isn't What You Think

Here's what keeps me up at night. And it's not poverty.

By the 2040s, the cost of living has genuinely dropped.

AI-driven healthcare is cheaper. Education is basically free. Legal services, financial advice, food, energy — all cheaper.

Someone earning $40K in 2045 might live as well as someone earning $80K today.

Material comfort isn't the problem.

Purpose is.

Work gives people identity. Structure. Social connection. Meaning. A reason to get up. When traditional employment shrinks, the demand for meaning doesn't shrink with it. It intensifies.

The economy of 2045 doesn't have a scarcity problem. It has a purpose problem.

And no policy paper, no economic model, no Silicon Valley pitch deck has a solution for that.

What Doesn't Change

Even in the most disrupted version of this future, some things hold.

People still consume.

Machines don't eat. Don't travel. Don't raise kids. Don't need vacations. The consumer economy restructures but doesn't vanish.

People still need meaning.

If anything, the demand grows. Community, creativity, caregiving, connection — these become economic forces, even if they don't look like "jobs."

Geography still matters.

Where you are determines whose laws apply, what currency you use, what healthcare you get, what community surrounds you.

As countries diverge on AI policy, jurisdiction becomes more important. Not less.

Relationships compound.

AI can't replace the trust between two people. The handshake. The shared risk. The dinner where you decide to build something together. Still human. Still scarce. Still the most valuable thing in any economy.

Scarcity shifts. It doesn't disappear.

When intelligence is abundant, what becomes scarce? Attention. Trust. Taste. Judgment. Physical presence. Emotional connection.

The economy of 2045 prices these things the way the economy of 2025 prices information processing.

So Here We Are

I'm writing this in February 2026.

The earthquake hasn't hit yet. But the tremors are everywhere if you look.

Fiverr's active buyers down 21% since 2022. Writing demand down 33%. Translation down 19%. Software job postings down 21%.

Forty-two percent of recent graduates underemployed. Planned hiring collapsed 34% year over year.

This is before the real wave.

The displacement is already running 5-10x higher than official statistics capture. Freelancers. Contractors. International knowledge workers. New graduates. Being displaced right now. With zero visibility in the numbers that policymakers watch.

I don't have a neat solution. Nobody does. Anyone who tells you they do is selling something.

But I know what the structural math says.

I know what the historical pattern looks like.

I know that the last time this happened, it lasted fifty years — because nobody understood what was happening while it was happening.

This time we can see it.

The question is whether anyone does anything about it.