Here's a party trick. Double a penny every day for thirty days and what do you have? Most people guess a few hundred dollars.
The real answer is over five million.
Nobody gets there on the first try, because our brains aren't built for this maths. We're wired for linear thinking. Glance at oncoming traffic and your body knows whether it’s safe to step off the kerb. That same brain, asked to picture something doubling every year, consistently underestimates how big the number gets by a factor of thousands.
For most of human history, this didn't matter much. Every tool we built, every institution we created, moved at a pace our linear intuitions could roughly track.
Now, for the first time, we have intelligence that doesn't think in a straight line. It compounds, feeds back on itself, accelerates. And it's arriving at the same moment as five or six other exponential curves, all hitting the steep part of the S at once.
I first wrote about this in April 2021, in a GMI feature I called The Exponential Age. It was the longest thing I'd ever put in the Monthly, and looking back, I didn't fully appreciate how much bigger the thing I'd spotted actually was...
What I Got Right, And What I Missed
Back in 2021 the observation was simple: fiat currency was being devalued faster than anyone was pricing in, and a handful of assets were the only things compounding fast enough to outrun it. Bitcoin and technology stocks, chief among them.
That part held up. What I underweighted was the scale of what was coming next.
Also, at the time I was looking at central bank balance sheets, mostly the Fed's. That was the right instinct, pointed at an incomplete target.
The real driver isn't any single central bank. It's Total Global Liquidity: every major central bank, every Treasury rolling debt, every commercial bank expanding credit to fund it, all moving together like a relay. When the Fed pauses, China or Europe picks up the baton.
But watch just one runner and you'll misread the whole race. Which is exactly why most people got 2017 wrong. The Fed was shrinking its balance sheet, and markets went vertical anyway, because China and Europe were doing the opposite. Total global liquidity rose. Nobody watching only the Fed saw it coming.
Global liquidity now expands at roughly 8% a year. Add ordinary inflation and your real hurdle rate, the return you need just to stand still, is closer to 11%.
The Part That's Actually New
The currency debasement story explains why your money buys less. It doesn't fully explain why everything feels like it's speeding up. Not just markets but the pace of change itself.
That's a separate force, layered on top, and it’s the reason the Exponential Age matters more now than it five years ago.
I named the curves back in 2021: artificial intelligence, robotics, solar and battery storage, biotechnology, blockchain networks. The list hasn't changed. What's changed is where they are on the curve.
In 2021 most of this was still theory. You could see it coming if you looked, but it hadn't arrived. Five years on, they've all accelerated at once, in the same handful of years, and started feeding each other.
That convergence changes everything.
Artificial Intelligence
Here's the thing about debasement that most people miss. The reason governments keep expanding debt isn't stubbornness or incompetence. It's demographics. Ageing populations, shrinking workforces, fewer people producing and more people drawing on the system. You can't grow your way out of that with human labour alone, so you borrow instead, and the balance sheet expands to compensate.
AI breaks that equation.
When an AI agent can do the work of a knowledge worker, and a humanoid robot can do the work of a manual labourer, you're no longer constrained by how many humans of working age exist. You've created synthetic demographics. The productivity curve that demographics was dragging down starts moving up again, and it does so without the debt expansion the system has needed for fifty years.
There's a deflationary force running alongside it too. When the marginal cost of intelligence drops toward zero, an enormous range of goods and services gets cheaper fast. That doesn't eliminate the debasement problem overnight, but it changes the maths. The 11% hurdle rate I described earlier starts to look different in a world where AI is compressing costs across the whole economy.
But it’s worth pausing on the pace of all this because it’s staggering. The length of autonomous tasks that AI can perform has been doubling roughly every seven months for the past six years. OpenAI's o3 already outperforms human PhDs in their own scientific domain. And development hasn't slowed.
Energy
There's a huge bottleneck sitting underneath all of this.
AI and robotics run on compute. Compute runs on electricity. And the scale of compute being built right now is so large that energy has become the binding constraint on the entire technological transition. Microsoft is looking at nuclear. Google is signing deals for geothermal. They aren't doing this for their carbon targets, they're doing it because they're running out of grid capacity to feed the machines.
China understood this first and moved hardest.
In 2024 alone China added more new solar capacity than the rest of the world combined.
Here's why that matters, and it comes down to a piece of economics most people have never heard of.
There's a thing called Wright's Law, first spotted watching aircraft factories back in 1936. It says that every time the total number of units ever produced doubles, the cost of making the next one falls by a consistent percentage. Workers get faster, defects fall, someone works out the panel runs just as well with less silver and thinner silicon…
Solar has followed Wright's Law more faithfully than almost any technology ever measured. Every doubling of global capacity knocks another 20-something percent off the cost. And this is where China closes the loop, because by manufacturing panels at absurd scale they drove cumulative global production through the roof, which shoved the entire world further down the curve, faster.
Solar is 90% cheaper than it was a decade ago, and it is nowhere near the bottom. That cost curve gives solar four properties nothing else in energy can touch.
It's cheap, it’s fast, it's decentralised, and it's scalable in a way fossil fuels fundamentally are not. Every other energy source runs into a wall somewhere in the supply chain. Solar's only limit is the size of the sky.
Storage was always the counter-argument, and storage is closing in fast. Tesla's Megapack business has been growing 50-70% a year with new factories coming online to keep up. Grid-scale batteries have got cheap enough, quickly enough, that most people simply haven't caught up to what it means yet.
Then the feedback loop kicks in, and this is the part that should make you sit up. AI optimises the energy grid, which makes power cheaper. Cheaper power makes compute cheaper. Cheaper compute makes AI better. Which then optimises the grid again. These curves aren't running side by side. They're multiplying each other.
Crypto
Bitcoin's relationship to global liquidity is well documented at this point. Roughly 90% of its price movement since 2012 traces back to the liquidity cycle. That part of the thesis is intact and, if anything, tighter than I originally had it.
But there's a second argument for crypto that barely existed in 2021 and is now becoming unavoidable.
AI agents need to transact. Millions of them, eventually billions, buying services, allocating resources, settling between themselves at machine speed. Human financial infrastructure, with its clearing houses and correspondent banks and three-day settlement windows, simply cannot support that. You cannot run an agent economy on the existing rails.
Crypto can. Programmable, trustless, instant settlement, no counterparty required. Blockchain is the only financial infrastructure that actually scales with a hyperintelligent economy. The adoption argument for crypto was always compelling. The inevitability argument is something different again.
The Convergence
Here's where it gets interesting...
Every previous technology wave arrived alone and took decades to diffuse. The internet was one curve. Mobile was one curve. Each transformed the economy, but sequentially, with enough breathing room that institutions could adapt.
This is the first time multiple exponentials are hitting the steep part of the S at the same moment, and crucially, feeding each other. AI designs better chips. Better chips train better AI. Cheaper energy powers more compute. More compute optimises the energy grid. Crypto settles the transactions between autonomous agents that neither humans nor banks ever see.
None of these needs the others to keep climbing. Together, they climb faster than any one of them would alone.
Hyperscaler capex alone is running north of $600 billion a year, up 36% on the prior year, and that figure doesn't touch Tesla, xAI, the frontier labs, or the sovereign buildouts coming out of the Gulf. As a share of GDP, corporate balance sheets are now outspending what governments committed to build the atomic bomb.
The Double Exponential
That compounding has a name, and it's the real reason the brain can't keep up.
A single exponential already beats us. That's the penny. But when the curves feed each other the way I've just described, you don't get a steeper single exponential. You get a double one, growth on top of accelerating growth, and there's a specific mechanism driving it...
Think of it as a ladder. Sarnoff's Law said a broadcast network's value grows in a straight line with its number of viewers. Metcalfe's Law said a network where any two people can connect is worth far more, growing with the square of the users, n². Reed's Law goes further again. In a network where participants can form groups, value scales with two to the power of n, because the number of possible coalitions explodes far faster than the number of pairs.
For most of history Reed's Law was a curiosity, because the nodes in any network were people. And people are slow, scarce, and can only be in so many groups at once.
Now the nodes are intelligent, tireless, and can spin up a million copies of themselves. AI agents form coalitions, dissolve them and reform them at machine speed, in numbers no human network could reach. This is the first time the nodes in the network are themselves intelligent, and the first real expression of Reed's Law at the scale of an entire economy. Two to the power of n isn't a steeper straight line. It's a curve that keeps bending even after you've flattened everything onto a log scale.
That's what this chart shows.
So to come back to the penny one more time. A single exponential is already more than the mind can hold, which is why nobody guesses five million. A double exponential is a different order of problem. No intuition survives it, no mental picture, no evolved instinct. You can't draw this curve in your head, and neither can I.
That's what's actually changed since 2021. Not the technologies. I had those on the page. What I underestimated was that they'd stop behaving like separate curves and fuse into a single one that bends off the top of the chart. This is fucking wild, and we're still on the flat part of it.
Where This Leaves You
So what do you actually do with all this?
Follow the thread to its end.
If you accept the premise that synthetic labour replaces human labour, that AI agents and robots become the productive units the economy runs on, then you have to accept where the returns go. They flow to whoever owns the machines and the rails they run on.
The question stops being "how do I keep my job safe from the machine" and becomes "how do I own a piece of the machine." The same logic that says AI takes the work also tells you exactly where the value pools, and it pools somewhere you can actually buy into.
There's a name for this idea once you scale it up to a whole society: universal basic equity. People holding a direct stake in the machines, so the productivity gain flows back to them as owners rather than as workers drawing a wage. It's one of the more serious answers to what happens when wages stop working, and I write about it properly as part of the Economic Singularity, the 2030 to 2032 window where all of this converges into a genuine phase change and the old models stop describing reality. Whether that transition is smooth or violent depends on choices being made right now.
I'm not telling you what happens next. I'm showing you what's already happening: liquidity expanding at a rate you can measure, adoption curves you can plot, a double exponential bending off the top of the chart, and a small set of assets sitting directly on top of all of it. Call parts of it a bubble if it makes you feel better...
The maths says otherwise.
This is the Exponential Age. Where it leads next is a separate framework of mine, Economic Singularity, and that's where I get into what to actually do about all of it.
More at raoulpal.com
