§Real math · Global wealth

How Much Money Is In The World?

The chart every Bitcoiner shares says one thousand trillion dollars. I checked every line of it against the original sources. One line is out by $157 trillion, and it is not the line you would guess.

There is a chart that goes round Bitcoin Twitter every few months. It shows every asset class on earth as a coloured rectangle, and a tiny orange square in the corner labelled Bitcoin.

The version doing the rounds says one thousand trillion dollars.12 I wanted to know if that was real.

So I rebuilt it. Every line, from the original report rather than from the chart.

The short answer: about $910 trillion

That is the total value of everything people hold as a store of value. Property, bonds, shares, cash, gold, collectibles and Bitcoin.

It is close to the number in the chart. But it gets there differently, and one of the differences matters a great deal.

The three numbers you need

$910 trillion of gross stores of value. $570 trillion of net household wealth after debts.10 $144 trillion of actual money.4

All three are correct. They answer different questions, which is why people argue about this without ever disagreeing.

Every store of value on earth, drawn to scale · $910 trillion Real estate $393T 43.2% Bonds $161T 17.7% Equities $152T 16.7% Broad money $144T 15.8% Gold $31T Art & co… $27T Bitcoin, 0.18% Rebuilt from the original reports. Bitcoin is the orange sliver in the corner.
Our own version of the chart, rebuilt from the original reports. Every block is drawn to its true area, so Bitcoin really is that small.

Rebuilding the number, line by line

Here is every asset class with its actual source. The middle column is the chart everyone shares, the right column is what the original reports say.

Asset classThe chartRebuilt from sourceVerdict
Real estate
Residential, commercial and farmland
$370T $393.3T1 Sound
Bonds
Every tradeable debt security
$318T $160.7T2 Out by $157T
Equities
Every listed company on earth
$135T $151.9T3 Sound
Broad money
169 countries, 99% of world GDP
$129T $144.0T4 Sound
Gold
220,000 tonnes above ground
$22T $31.1T56 Stale by 41%
Art, cars, collectibles
The softest number here
$27T $27.0T22 Unsourced
Bitcoin
20.09M coins
$2T $1.6T25 Live
Total $1,003T $909.6T

The bond line is not bonds. It is all the debt in the world

This is the error, and it is an easy one to make. The world's actual bond market is $160.7 trillion.2

The $318 trillion in the chart is close to the total global debt figure, which was $348 trillion at the end of 2025.7 That number includes bank loans and mortgages, not just bonds.

Which creates a problem. The mortgages inside it are claims on the houses already counted in the real estate line.

That single line is overstated by $157 trillion. The chart's total only comes out $93 trillion too high, because its other lines are understated and partly cancel the error.

Why that matters to you

Picture a £300,000 house with a £200,000 mortgage. The chart counts the house at £300,000 and the loan at £200,000, so £500,000 of the world's wealth.

There is still only one house. The mortgage just decides who owns how much of it.

Gold is the other one worth knowing about. The chart priced it when gold was near $3,300, and it is $4,391 today.6

So the correct figure is $31 trillion, not $22 trillion. Since gold is the yardstick everyone measures Bitcoin against, that is a 41% understatement of the thing being measured against.

The bond line is not bonds Bonds, as charted what the popular chart says $318T Bond securities, actual SIFMA, every tradeable debt security $160.7T Total global debt IIF. Includes bank loans and mortgages $348T The chart used a total-debt number in a bond row. The mortgages inside it are claims on houses already counted elsewhere.
The one line that does not survive checking. A total-debt figure was used in a bond row.

There are three honest answers, and they differ by three times

This is the part nobody explains. Two sensible people can disagree about this by a factor of three without either of them being wrong.

It depends entirely on which pile you point at.

Which numberFigureWhat it actually measures
Net household wealth $570T10 What families own after every debt is subtracted. UBS puts personal wealth above $500T on its own method.11
Gross stores of value $910T Every pool a saver could move money out of. The right one for asking what Bitcoin could capture.
The full balance sheet $1,800T10 Every asset plus every mirrored financial claim. Too big, because it counts the same value repeatedly.

Net worth is the right measure of how rich the world is. Gross stores of value is the right measure of how much money is looking for a home.

The money grows 7% a year. The economy grows 3.

This is the most important number on the page, and it is measured rather than forecast. Global broad money went from $26 trillion in 2000 to $144 trillion in December 2025.4

That is 7.09% a year, across 169 countries covering 99% of world output. Two other sources land in the same place.

What is growingRateOver
Global broad money7.09%2000 to 2025, 169 countries4
US M2, median6.60%Since 196016
The world economy4.73%2005 to 2025, nominal17

Look at the last row. The money grows faster than the things money buys, every single year.

That gap does not vanish. It lands in prices, which is where your house price came from and why your savings account has never once beaten it.

It takes about three pounds of new debt to produce one pound of new economy. Measured, 2005 to 2025

Here is that claim checked. Global debt rose by $208 trillion between 2005 and 2025,7 and global output rose by $71 trillion.17

That is $2.92 of new credit for every $1 of new output. Jeff Booth quotes roughly four to one from his own period, and measured over the last twenty years it is closer to three to one.15

A system that needs three pounds of credit to make one pound of economy cannot repay that debt out of the economy. It gets inflated away instead, which is the 7% you just looked at.

The money supply against the economy it is supposed to measure $40T $79T $119T $158T 2000 2005 2010 2015 2020 2025 Economy, $118T Money, $144T Money compounds at 7.09% a year. The economy it buys grows slower. The gap lands in prices.
Twenty-five years of both. The widening gap between the two lines is where your house price came from.

Why $10 million a coin is not the big number you think

Here is the mistake almost everyone makes, on both sides of the argument. They argue about a price and never convert it back into a share.

Start with the 7% above. What happens if Bitcoin does nothing at all except keep pace with the money supply for forty years?

The answer is $1.2 million a coin

At 7% a year, $81,092 becomes $1.21 million by 2066. And Bitcoin's share of the world's savings is exactly what it is today, 0.18%.

It won nothing. It just was not diluted.

So a million-dollar Bitcoin is not a bull case. It is the null result.

Which means most of that number is not Bitcoin going up. It is the pound going down.

What each headline price is really asking for gold today, 3.4% $1 million less than Bitcoin holds today 0.15% $2 million +1.3 points over the money supply 0.31% $5 million still under one percent 0.77% $10 million under half of gold 1.5% $20 million just under gold today 3.1% $45 million Saylor’s 7% share 6.9% Strip out forty years of debasement and the scary numbers become ordinary shares.
The same six prices people argue about, converted into what they actually claim about the world.

Run the scary numbers backwards

Take the headline prices people throw around. Strip out forty years of debasement, and express what is left as a share of the world's stores of value.

Headline price 2066
inflation left in
Same price in today’s moneyNeeds headline/yrOver money supplyShare of the world
$1 million$0.22M6.5%−0.5 points0.15%, less than today
$2 million$0.44M8.3%+1.3 points0.31%
$5 million$1.09M10.9%+3.9 points0.77%
$10 million$2.18M12.8%+5.8 points1.5%, under half of gold
$20 million$4.36M14.8%+7.8 points3.1%, just under gold today
$45 million$9.80M17.1%+10.1 points6.9%, which is Saylor's 7% share. He said $13M by 2045; the $45M is that share carried to 206613
The trap in this table, named directly

$10 million here is not the same as the $9.9 million you will see quoted for Saylor's base case. One is a headline price with forty years of inflation in it. The other is in today's money.

Saylor's 7% is $9.9M in today's money, which is $45M as a headline price. A $10M headline price is only $2.18M in today's money, so it buys 1.5% of the world, not 7%.

They look like the same number and they are four and a half times apart. Forty years of 3.88% debasement multiplies every headline price by 4.59. Nothing on this page is a headline price unless the column says so.

Ten million dollars a coin is Bitcoin becoming half of gold. Twenty million is Bitcoin becoming gold.24

Not gold plus the bond market. Just gold, the thing that already exists and that nobody finds shocking.

Both halves of this are true

The same logic cuts the other way, and it should. Going from $81,000 to $10 million is a 123-fold rise in the number, but only a 27-fold rise in what it buys you.

The other 4.6 times is the currency shrinking underneath it. A price target is always a claim about two things at once, and it never says which one is doing the work.

Before the model: what will things actually cost?

Every price on this page is in today's money. A £4 coffee means "costs what £4 buys you now", not "the ticket says £4".

That matters because the ticket price in 2066 will be far higher. Forty years of debasement multiplies every price tag by 4.59 times.

So why does the coffee still say £4?

It does not. The 2066 ticket price of a £4 coffee is about £22. What has not changed is how much of your earnings it takes.

Quoting everything in today's money is the only way to compare 2066 with a price you actually know. Both numbers are shown in the model below.

The Chart of the Century, turned into rates

Mark Perry's chart tracks 14 categories of US prices from 2000 to 2024 against overall inflation.29 It is the best empirical answer to "what gets cheaper".

I converted every category into a real rate per year, which is what you need to project anything forward.

It is not machines against people. It is competition against protection. Every category, converted to a real rate per year, against inflation AND against wages 0% = tracked inflation exactly wages, +0.75%/yr real Televisions -15.9% open to competition Toys -7.7% open to competition Computer software -7.5% open to competition Cell phone service -4.8% open to competition Clothing, cars, furniture -2.6% Housing -0.2% Food and drink +0.0% Childcare +0.5% tracks wages Medical care +0.8% tracks wages College textbooks +1.4% ABOVE wages College tuition +1.7% ABOVE wages: subsidised Hospital services +2.1% ABOVE wages: third-party paid Dashed line = wage growth. Services that merely track wages sit on it. Only the subsidised ones run past it.
Each category against inflation, and against wages. Services that merely track wages sit on the dashed line. Only the subsidised ones run past it.

The split is brutally clean. Televisions fell 15.9% a year in real terms, hospital services rose 2.1%.

My first read of this was that machines get cheaper and human labour does not. That is wrong, or at least badly incomplete.

Test it against wages, not inflation

If the expensive side were just "a person's time", those services should rise in line with wages. Wages rose 127.4% over the same period, against 89.9% inflation.29

So the real test is which services outran wages, not which outran prices.

Servicevs inflationvs wagesWhat that means
Haircuts, laundry, restaurants, car repairrose~0.0%Tracked wages almost exactly32
Childcare+0.52%−0.23%Wages explain all of it
Medical care services+0.80%+0.05%Wages explain all of it
College tuition+1.69%+0.93%Excess. Something else is going on
Hospital services+2.08%+1.31%Excess. Something else is going on

There it is. Most services just track wages, which is the ordinary Baumol effect and no scandal at all.

Only two run meaningfully past wages, for 24 years straight. They happen to be the two most subsidised and most third-party-paid things on the list.

You were right about the cause

The New York Fed found tuition rises about 60 pence for every extra pound of subsidised loans, and 55p per pound of grant aid.31 An NBER study found for-profit colleges eligible for federal aid charged 78% more than ineligible ones.

So the chart is not machines against people. It is competition against protection. Things exposed to open competition collapse in price. Things propped up by subsidy, third-party payment or restricted supply do the opposite.

And housing sits on the expensive side for a reason this page has already argued at length: it is carrying a monetary premium. It is not just a house, it is a savings account with a roof.

Which gives every assumption in this model a source

So I stopped guessing. Each item below takes the rate its own category actually delivered.

ItemReal change a yearWhere that rate comes from2066 in today's money2066 ticket price
A mid-range car−2.6%What new cars actually did. Not the autonomy case£6,972£32,007
A year of groceries0.0%Food tracked inflation almost exactly£5,000£22,953
A flat white+0.5%A service. The chart's service side ran +0.5% to +2.1%£5£22
A year's rent+1.0%A claim on the same land, so it moves with housing£23,226£106,477
A UK house+1.0%US housing did −0.2%. UK land has done better, so this is the honest worst case£431,770£1,982,052
Where I am assuming more than history

My earlier version had the car at −5% a year on the strength of ARK's autonomy work.21 History says new cars did −2.6%.

I have used history. If electric drivetrains and autonomy do deliver −5%, a car costs £2,570 rather than £6,972, and every number below improves. That is upside I have left out rather than banked.

Why there is no laptop in that list

I had one, at £53 in 2066. It was wrong, and the reason is worth understanding because it applies to a lot of technology.

The CPI computer index is hedonically adjusted. BLS's own worked example: a laptop goes from $1,000 to $1,200 with a better processor and screen, and that is recorded as a price fall, because you got more machine per dollar.30

Which means the index measures the wrong thing for this question

Your dad's £4,000 laptop and a £4,000 laptop today cost the same. The second one is vastly better. The price of the category did not fall, the quality per pound exploded.

So "a laptop will cost £53" is what the index implies and is not a lived prediction. Nobody will buy a 2026-spec machine in 2066, any more than you would buy a 1986 one now.

The honest version of the same arithmetic: the same £1,200 buys roughly 23 times the machine. For fast-improving technology, the gain arrives as a better thing, not a cheaper one. It may not even be called a laptop.

So the model now only lists things where "the same thing" still exists in forty years: a car, a food shop, a coffee, rent, a house.

That is a narrower claim and a more defensible one. The technology gain is real, it just does not show up as a number you can put in a shopping list.

And "4% of a house" means your £20 covers 4% of one. Whether that sounds good depends entirely on what £20 buys you today, which is five coffees.

Model it yourself

Drag the share and everything moves with it, including the map. The whole argument on this page is one number, so it is worth playing with.

Bitcoin's share of the world in 2066 7.0%

of the $2,967T of stores of value the world will hold, in today's money

0%10%20%30%40%50%60%

Price per Bitcoin$9.9Min today's money
Headline price$45.4Mwith 40 years of inflation in it
£20 today becomes£2,440£2,929 allowing for lost coins
One satoshi7.4p$0.099
Return needed12.8%a year, real, for 40 years

What your £20 would buy in 2066 · in today’s money, with the 2066 ticket price beside it

Rates come from the Chart of the Century29, not from guesswork. Only categories that will still exist in 2066 are listed. The house moves with the slider: if Bitcoin absorbs the world’s savings, property loses the monetary premium it is carrying, with a floor at its shelter value.

The $846 trillion that nobody owes

You will see a much bigger number quoted: $846 trillion of derivatives,8 or even "$2.3 quadrillion" once someone adds it to the debt.

Do not use it. That figure is the notional, and the notional is a multiplier, not money.

The worked example

Two banks agree an interest rate swap on £100 million notional. One pays a fixed 4%, the other pays whatever the floating rate is.

The £100 million never moves. It exists only to work out the interest, and what actually changes hands is the difference between the two rates.

That might be £500,000 a year. Nobody is £100 million in debt.

Scale it up and the BIS publishes three numbers, not one. The gap between the first and the last is the whole story.

MeasureFigureWhat it means
Notional outstanding$846T8The size of the machine. Every trade counted twice, because both dealers report it.
Gross market value$21.8T8What replacing every contract today would cost. Just 2.6% of the notional.
Gross credit exposureabout $3.3T9The real number, after netting. The BIS says this runs at 14 to 16% of market value.

So the headline figure is roughly 260 times the thing people think it measures. Use $348 trillion of real debt instead, and nobody can touch you.

But do not dismiss it either

Notional tells you the size and the interconnection of the machine. Gross market value tells you what is at stake if nothing breaks.

AIG's swap book looked modest right up to September 2008, when the netting assumptions stopped holding and the US government put $182 billion into one insurer. It is a measure of fragility, not a measure of what is owed.

$846 trillion of derivatives, and what is actually at risk Notional outstanding the number people quote $846T Gross market value 2.6% of it $21.8T Gross credit exposure after netting. the real one $3.3T Drawn to scale. The bottom bar is 1/256th of the top one, and it is the only one anybody owes.
The three numbers the BIS actually publishes, drawn to the same scale.

So how big is Bitcoin in all this?

Bitcoin is 0.18% of the world's stores of value.25 Gold is about 3.4%.

That comparison is the cleanest one available, because gold is a pure store of value with no yield and no counterparty. But there is a catch that most people skip.

First, which 3% do you mean?

Gold's share depends entirely on what you put underneath it, and three different answers are all correct. Every figure below uses the same $31 trillion of gold, and the same $1.63 trillion of Bitcoin. Only the denominator changes.

Measured againstSizeGoldBitcoin
Everything people store wealth in, property and art included$910T3.4%0.18%
Financial assets only: equities, bonds, money$488T6.4%0.33%
Equities and bonds only, and investment gold only24$312T2.7%0.52%

The first row is the one this page uses, and it is the one to quote. It is the widest and the hardest to argue with, because it counts the house as well as the share portfolio. Every chart below is on that row.

Gold has no normal share

Gold has been worth anywhere from 0.62% to 4.4% of everything since 1960. It is 3.5% today, and the 1980 peak was 4.4%.39

The useful comparison is not 1980. It is 2000. Gold was 0.62% of the world's wealth then and is 3.5% now, five and a half times the share, inside twenty-six years. The 1980 peak was only about a third above today, because property was already the biggest thing in the pie back then and it swamps the difference.

Take property out and the swings get violent. Against financial assets alone gold went from 13.5% in 1980 to 1.3% in 1999 and back to 6.5%.38 Against a year of world output it ran at 23% in 1900, fell to 3.7% by 1970 while the price was still pinned at $35, and is about 27% today.37

You will also see 0.19% quoted for the year 2000. That counts only privately held investment gold, not the whole above-ground stock, so it is about seven times lower than everything else here. It is not comparable and it is not used.23

So there is no number to call gold's rightful share. A monetary asset gets whatever the world decides to give it that decade, and every way of measuring it moves by multiples inside one working lifetime.

Project the $910 trillion forward at 3% real growth and it becomes $2,967 trillion by 2066, in today's money. What Bitcoin is worth then depends on one assumption only.

If Bitcoin ends up as...SharePer coinReturn needed
A niche asset that survived1%$1.4M7.4% a year
What gold holds today3.2%$4.5M10.6% a year
Saylor's published base case137%$9.9M12.8% a year
The primary global reserve asset15%$21.2M14.9% a year
Saylor's published bull case1322%$31.1M16.0% a year
"Half of everything"1450%$70.7M18.4% a year

Every price there is in today's money and assumes 21 million coins. Between 3 and 4 million are lost for good,20 so the real figures are roughly 20% higher.

The same world in 2066, if Bitcoin reaches 7% · $2,967 trillion in today’s money Real estate $1,195T 40.3% Bonds $488T 16.5% Equities $462T 15.6% Broad money $438T 14.7% Bitcoin $208T Gold $95T Art & colle… $82T Saylor’s base case. Bitcoin is now bigger than gold, art and collectibles put together.
The same picture forty years on, if Bitcoin reaches Saylor's base case. Compare the orange block with the one at the top of this page.

What 12.8% a year actually means

Saylor's base case needs Bitcoin to return 12.8% a year after inflation for forty years. The obvious move is to compare that to the stock market, and the obvious move is wrong.

The comparison people reach forReal returnWhy it misleads
The S&P 500, 1926 to 2026187.29%Companies earning and reinvesting profits. A perpetual engine.
Warren Buffett, 1965 to 20241915.40%A skill outcome, and 93% of fund managers fail to beat a plain index over 20 years.27

Both figures are real, inflation-adjusted returns, so the arithmetic is sound.18 The problem is that neither describes what Bitcoin is doing.

Buffett picked companies for sixty years without getting it badly wrong. Bitcoin does not pick anything, so asking whether it can match Buffett is not a question about Bitcoin.

All the gold ever mined, against one year of world output Every ounce ever mined, at the London price, ÷ world GDP 0% 7% 14% 21% 28% 35% 1870 1900 1940 1980 2026 gold price fixed by law 18.7% in 1980 3.7% in 1970 3.9% in 2000 27% today Dashed before 1960: benchmark years only, and the band is the USGS and the World Gold Council disagreeing. Solid from 1960. Gold has swung between 3.7% and 27% of a year of world output.
Every ounce ever mined, at the London price, against one year of world output, against the latest full year measured. Gold is not cheap against what the world makes. It is the highest reading in the whole series. Before 1960 there are only benchmark years, and the band is real disagreement: the USGS and the World Gold Council are 9,470 tonnes apart on how much gold existed before 1950, which is 4% of today's stock and a third of 1900's. The line takes the World Gold Council's figure, the floor of the band is the USGS's.37
Gold and Bitcoin as a share of everything, every year Property + bonds + equities + broad money + gold + art + Bitcoin. The same $910T pie as the top of this page. 0% 2% 4% 6% 1960 1980 2000 2026 4.4% 1980 peak 0.62% in 2001 gold 3.5% Bitcoin 0.18% Banded and dashed before 2015: Savills measures world property from 2015, and everything earlier is reconstructed. Gold has moved between 0.62% and 4.4% of everything. Bitcoin is on 0.18%.
The top row of the table above, drawn back to 1960: the same $910T pie the treemap at the top of this page shows, with gold and Bitcoin measured against all of it. Gold has moved between 0.62% and 4.4% of everything, so its 1980 peak was about a third above where it sits today, not double. Savills only measures world property from 2015; before that it is reconstructed from Piketty and Zucman's housing-to-income ratios, which is what the band and the dashed line are saying.39

The right comparison is gold in the 1970s

Bitcoin's 12.8% is not a stock-picking result. It is a monetary asset being repriced, and there is exactly one clean modern precedent for that.

GoldPriceReal return a yearWhat happened
Aug 1971 to Jan 1980$35 to $850+35.5%Nixon closed the gold window and gold repriced 24-fold in eight years.26
Jan 1980 to 2000$850 to $279−9.1%Then it gave back 85% in real terms over twenty years.26
2000 to 2026$279 to $4,391+8.4%Twenty-six years to make the round trip worth having.6
Gold’s price in today’s money, 1960 to 2026 Log scale. Nominal price deflated by US CPI, so every point is comparable $250 $500 $1,000 $2,000 $4,000 1960 1971 1980 1990 2000 2010 2020 2026 fixed at $35 1970: real low of the peg era 1980 peak, $2,473 2000: $537, down 78% 2026: $4,391, a new real high Pinned at $35 while inflation ate it, then freed in 1971. Anyone who bought the 1980 top waited 26 years to break even.
Gold in today's money, year by year. Pinned at $35 while inflation ate it, freed in 1971, and a 26-year wait for anyone who bought the 1980 top.

Your instinct about 1971 was right. Gold was fixed at $35 an ounce, so through the 1960s its real value fell 1.4% a year while the price never moved.35

Freeing it in 1971 released a decade of suppressed repricing all at once. That is what the spike actually is.

Read the first row again. The last time the world repriced its monetary asset, it ran at 35% a year for eight years.

Bitcoin's base case asks for 12.8%, and only 9.6 points of that is the repricing at all. Measured against the thing Bitcoin is actually doing, that is about a third of the pace gold managed.

So what is actually hard about it?

Not the rate. The duration. Gold cleared the rate by miles and then failed to hold it, losing most of the gain over the following twenty years.

Bitcoin's 12.8% has to be sustained for forty years with no twenty-year hole in the middle. That is the real risk, and it is about adoption stalling rather than about the arithmetic being heroic.

First, how easy it is to prove anything

Pick two dates and Bitcoin's return becomes whatever you want it to be. Here is the same asset, over broadly the same decade, measured seven ways.

Measured fromToHeldMultipleReal return a year
Dec 2018 bottom, $3,206Oct 2025 peak6.8y39.3×66.12%
Dec 2018 bottom, $3,206today7.8y25.3×46.85%
Dec 2017 peak, $19,783Oct 2025 peak7.8y6.4×22.86%
Dec 2017 peak, $19,783today8.8y4.1×13.83%
Dec 2017 peak, $19,783Jun 2026 low8.5y2.9×9.98%
Nov 2021 peak, $68,789today4.9y1.2×0.23%
Oct 2025 high, $126,073today1.0y0.6×−38.86%

Nought point two percent to sixty-six percent. Every row is true, sourced and correctly calculated.2836

Which is why anyone quoting a single pair of dates, in either direction, is telling you nothing.

Second, why the early years have to go

The usual answer is to measure from the start. That is worse, not better, and the reason is the market cap rather than the price.

YearPriceMarket capWhat Bitcoin was, as an asset
2012$13.50$0.1bnSmaller than one branch of a supermarket chain
2013$947$11.6bnA mid-cap company
2016$970$15.6bnStill a mid-cap
2017$14,156$238bnFirst year it was large enough to be investable at scale
2021$46,266$874bnA large-cap
2026$81,092$1.63TAn asset class

Going from $11 billion to $238 billion is a 21-fold move on something smaller than Greggs.28 That cannot be repeated from $1.6 trillion.

So measuring from 2013, or worse from under $1,000, imports returns that were only available because the thing was tiny. 2017 is the first honest starting line.

Third, the one number worth carrying

Take the day-to-day price out entirely and measure the four-year moving average, which is one full halving cycle and has no start date to argue about.

Every honest way of measuring it, in one place base case needs 12.8% 2018 bottom → 2025 peak the flattering pair 66.1% 2018 bottom → today good start, ordinary finish 46.9% 4-year average since 2017 every window, no dates chosen 33.0% 2017 peak → 2025 peak bad start, good finish 22.9% Power law to 2066 the fitted model, 40 years out 15.5% 2017 peak → today the worst entry ever 13.8% 2021 peak → today the unflattering pair 0.2% Orange is the cherry-pick-proof number: every four-year window averaged. Green is the fitted power law.
Seven honest answers. Orange is the cherry-pick-proof one; green is the fitted power law.
The number to quote, and the only one

Bitcoin's four-year moving average has grown 33% a year in real terms since 2017.

Not 66%, which is two cherry-picked ends. Not 0.2%, which is the other cherry-pick. 33%, from a line nobody gets to choose the start of, beginning the first year Bitcoin was big enough to count.

For context, the base case that gets Bitcoin to 7% of the world needs 12.76%. The moving average has run at two and a half times that.

That does not make forty more years a certainty, and section 12 lists ten reasons to be careful. It does mean the required rate is not the thing standing in the way.

The four-year window, which nobody gets to cherry-pick

Bitcoin's supply halves roughly every four years, so four years is one full cycle. Measure every four-year window from year-end to year-end and the date-picking disappears.

Starting at the 2017 peak also means starting late enough that the sub-$1,000 era cannot flatter the numbers.

Every four-year window since the 2017 peak, no date picking what the base case needs, 12.8% 2017–2021 3.27× over four years 30.3% 2018–2022 4.42× over four years 40.5% 2019–2023 5.88× over four years 50.9% 2020–2024 3.22× over four years 29.8% 2021–2025 1.89× over four years 13.6% Year-end to year-end, so nobody chooses the start or the stop. The weakest window still clears the bar.
Rolling four-year real returns, year-end to year-end. The dashed line is what Saylor's base case needs for forty years.

Five windows, five answers, and all five clear the bar. But five year-ends is still five choices.

So I ran it daily instead. Throw a dart at a calendar, buy that day, sell four years later to the day. I did that for every single day since the 2017 peak, which is 1,738 throws.34

Not one of them lost money. The worst came out at 2.4% a year, the middling one at 30%, and 87% beat the 12.8% the base case needs.

That is the honest version. The floor is far lower than five year-ends suggest, and the bar still gets cleared closer to nine times in ten than ten.

The start date is pinned to the 2017 peak on purpose, so no throw is allowed to begin in the years when Bitcoin was small enough to move on nothing.

The line to watch: Bitcoin’s 4-year moving average Daily price in grey. The 1,461-day trailing average in orange. Log scale. $2k $5k $10k $25k $50k $100k 2019 2020 2021 2022 2023 2024 2025 2026 price −77%, average still rising average $64k The orange line has not fallen on a single day since it began. The grey one lost 77% at its worst.
Daily price in grey, the 1,461-day trailing average in orange. The orange line has never had a down day.

This is the line worth watching. Bitcoin's four-year moving average has not fallen on a single day since it began.34

Over the same period the daily price lost 77% at its worst. Same asset, same days.

Where the rate is now, and where it goes

Now: about 33% a year after inflation. That is the average of every four-year window since the 2017 peak, and the four-year moving average has grown at much the same rate over the same period.

Later: a lot less, and that is what the model says should happen. The power law prices Bitcoin as time since the genesis block raised to 5.69, so the annual rate is 5.69 divided by the number of years elapsed. It falls every year by construction.33

If the power law holdsReal return a yearAgainst the 12.8% the base case needs
Now to 203048.7%clears it four times over
2030s19.4%still clears it
2040s12.5%about level, and crossing below around 2044
2050 to 20668.0%below it
Whole forty years15.5%clears it

So the honest shape is not a flat 33% for forty years. It is a high rate now that decays to single digits, and the forty-year average still lands above what the base case needs.

The first leg looks the steepest because Bitcoin is currently sitting below the power-law line, so part of that 48.7% is catching up rather than growing.

What this means for you, rather than for the world

None of these numbers tell you what to do. The one that matters is how much Bitcoin you would need, at a rate you actually believe, to stop working.

Run it in the Financial Freedom Calculator. Put in your own monthly costs and your own assumed rate, 12.8% or 15.5% or 8%, and it gives you the number of Bitcoin and the year. It takes about thirty seconds.

And if the power law holds

There is a published model saying Bitcoin's price tracks time since launch raised to a power of about 5.69.33 I fitted it myself rather than take the number.

Seventeen year-end prices from 2010 to 2026, a straight line through log price against log time. My fit returns an exponent of 5.69 and an R² of 0.951, which reproduces the published figure.

The power law, fitted here on Bitcoin’s whole price history Log price against log time. Our fit: price ∝ days^5.69, R² = 0.951. Published: 5.69, R² 0.961 $1 $100 $10k $1M $100M 2010 2015 2020 2026 2035 2050 2066 extrapolation, not data $120M headline today, $81k Seventeen points, no curve fitting beyond a straight line. The grey dots are real year-end prices.
The fit, and what it implies if extended. The shaded half is extrapolation, not data.
Power law saysHeadline priceIn today's moneyShare of the world
2030$461,187$396,049
2040$3,980,878$2,336,316
2050$18,927,956$7,591,667
2066$120,039,909$26,184,72518.5%

That lands between Saylor's base case and his bull case, at an implied 15.5% a year real. Which is roughly Buffett's record.

Your £20 becomes £6,458 in today's money on that path.

Treat this one more carefully than the rest

A power law is a curve fitted to a log-log chart. It has a plausible mechanism behind it, network growth times Metcalfe-style pricing, but nothing guarantees a line drawn through 17 years keeps holding for another 40.

There is peer-reviewed work arguing the structure is weaker than the fit suggests, even where the forecasts perform well.33 I include it as a third independent method, not as a prediction.

Three methods, three answers, one neighbourhood

Gold's share says 3.2% to 5%. Saylor's capital-flow model says 7% base and 22% bull. The power law, fitted here from scratch, says 18.5%.

None of them agree. All of them clear the bear case by a wide margin, and that agreement matters more than the spread.

The same thing, measured as returns

Nearly nine in ten four-year holds since the 2017 peak beat the base case, and not one of them lost money. The power law then says that rate decays, and still averages above the bar across the forty years.

That is the strongest single fact in this research. No four-year hold since the 2017 peak has lost money in real terms, and nearly nine in ten beat what the base case needs.

It does not make forty years a certainty. It does mean the required rate is not the thing standing in the way.

Two things stop it being conclusive

Nine years is not forty. The 2021 row shows why: someone who bought that top has made 0.23% a year real over five years, and the base case has to absorb stretches like that and still average out.

And returns are decelerating by design. But notice that supports the case rather than undermining it: Saylor models a slowdown to 20% a year, which is about 16% real, and the base case only needs 12.8%.13

Which makes it a different kind of bet entirely

Break the 12.8% apart and it stops looking like a return. Three points of it is simply the world getting richer.

The other 9.6% is a re-rating, Bitcoin's share moving from 0.18% to 7% over forty years. That is not profit, it is other people deciding to hold it instead of something else.

The distinction that matters

The stock market's 7% is produced. Companies earn, reinvest and pay dividends, so a bad decade still compounds.

Bitcoin's re-rating is transferred, and needs a continuous net inflow of buyers with no earnings floor underneath it. Different engine, different failure mode.

So the claim is big, and smaller than it sounds. Bitcoin does not have to out-earn the world economy for forty years.

It has to be repriced once, from a rounding error to roughly what gold is worth, and then stay there.

Check this yourself

Every figure on this page links to its original source below. The two worth checking first are the ones that do the most work.

Source 4 is the 7% money growth figure, and source 2 is the bond market number that the popular chart gets wrong. Neither requires you to take anyone's word for anything.

The four-year figures were built twice, separately, and met in the middle. One pass fixed the buy date and looked forwards, the other fixed the sell date and looked back. Both landed on the same worst window: bought 14 April 2021, sold 14 April 2025. Two methods, one answer, is worth more than either on its own.

Keep reading

What an economist would say about all this

Everything above is built to be checkable, which means it should survive someone trying to knock it down. Here is my best attempt at doing that myself.

These are the objections I think are strongest, roughly in order of how much damage they do.

1. Market cap is not money invested, and this is the big one

The whole page is framed as savings "flowing into" Bitcoin, as if $208 trillion has to arrive for it to reach 7%. That is not how market capitalisation works.

Every purchase has a seller. A dollar of net inflow can lift market cap by many dollars, because the price is set by the thin margin of people actually trading, not by the whole stock.

This cuts both ways and I should say both. It makes the target far easier to reach than "find $208 trillion" implies. It also makes the valuation more fragile, because a number that can be bid up on light volume can be bid down the same way.

2. The returns are not risk-adjusted, and that flatters Bitcoin badly

I compare Bitcoin's required 12.8% to the S&P's 7% as if they were like for like. They are not. Bitcoin's volatility has run roughly four times the S&P's.

On a risk-adjusted basis, earning 12.8% at Bitcoin's volatility is a worse deal than 7% at the S&P's. An economist would say the comparison is close to meaningless without that adjustment, and they would be right.

3. The moving average never falling is partly an artefact

I made a lot of "the four-year average has never had a down day". A trailing average of any series that trended up over the sample will mechanically rise.

It is a statement about a period in which Bitcoin went up about 25 times, not a property of the asset. If Bitcoin fell 80% and stayed there, the average would fall too, just with a lag of up to four years. It removes noise. It does not remove risk.

4. Forty years is longer than Bitcoin has existed, twice over

Bitcoin is 17 years old. Projecting 40 years out is extrapolating 2.3 times beyond the entire life of the thing.

No asset class has ever been forecast that far with useful accuracy. The base rate for "new monetary technology wins within four decades" is not low, it is unknown, because the sample size is about one.

5. The power law fits, but so would almost anything

An R² of 0.951 on a log-log chart sounds decisive and is not. Log-log compresses everything, so any strongly trending series fits a straight line well.

The peer-reviewed work I cited is titled "weak structure, strong forecasts" for exactly this reason.33 I have included the model as a third opinion, and it should not be read as more than that.

6. The monetary premium numbers are mine, not measured

The Booth calculation puts 40% of housing, 30% of equities and 75% of bonds down as monetary premium. Those are judgements, not data, and the 53.6% headline is highly sensitive to them.

Halve the housing assumption to 20% and the pool drops from $488 trillion to $410 trillion, which is 45% rather than 53.6%. The argument survives. The precision does not.

7. American data, British conclusions

The Chart of the Century, the CPI series, the wage data and the housing comparison are all US. UK inflation, UK wages and especially UK housing have behaved differently.

I flag it on housing and nowhere else. Anyone applying these rates to a British life should treat them as directionally right and numerically approximate.

8. You have to sell it to spend it

Every "your £20 buys X" figure assumes you can liquidate at the model price, in size, with a willing buyer, and ignores capital gains tax, which on these numbers would be the single largest deduction.

A 40-year hold in the UK with a gain of that size is a very large tax event, and none of the figures on this page are after tax.

9. Two assumptions in the denominator that could easily be wrong

The pie grows at 3% real forever, which is roughly past global growth. Growth is decelerating almost everywhere as populations age, so that may be generous.

It also assumes assets stay a constant multiple of output. That multiple has been rising for forty years, which is the financialisation story. If it mean-reverts, the denominator shrinks and every share on this page gets easier to reach for the wrong reason.

10. I compare Bitcoin's forecast to gold's history

Gold's 3.4% share is what it actually achieved, including a 20-year stretch where it lost 84% in real terms. Bitcoin's 7% is a projection.

Comparing a hoped-for number to a realised one always favours the hope. The fair version is the one in the gold chart above: the realised path had a 26-year hole in it.

What survives all of that

Three things, I think. The money supply really has grown 7% a year while output grew 3%, and that is measured, not modelled.

Bitcoin really is 0.18% of global stores of value against gold's 3.4%, so the gap is real whatever you think it will close to. And a price target really is two claims in one, so most arguments about "$10 million" are arguments about the pound, not about Bitcoin.

What does not survive is precision. Treat every number past 2030 on this page as an illustration of a mechanism, not a forecast.

Sources

Every bracketed number above jumps here. Each entry says which figure it supports, so a single claim can be checked without reading the whole page.

  1. Savills. World's real estate worth $393.3 trillion and is the world's largest store of wealth January 2025. Residential $286.9T, commercial $58.5T, agricultural land $47.9T. https://www.savills.com/insight-and-opinion/savills-news/381209/world-s-real-estate-worth-$393.3-trillion-and-is-the-world-s-largest-store-of-wealth
  2. SIFMA. 2026 Capital Markets Fact Book Global fixed income securities outstanding $160.7T in 2025, up 10.6% year on year. https://www.sifma.org/news/blog/2026-capital-markets-fact-book-key-findings
  3. World Federation of Exchanges. Public markets post strong growth for 2025 Global equity market capitalisation $151.94T at end-2025, up 18.5%. https://www.world-exchanges.org/news/articles/new-wfe-data-public-markets-post-strong-growth-2025-despite-geopolitical-instability
  4. Voronoi / Econovis. Global broad money supply reaches $144 trillion in 2025 $26T in 2000 to $144T in December 2025, a 7.0% compound annual rate. 169 countries, 99% of world GDP. https://www.voronoiapp.com/money/-Global-Broad-Money-Supply-Reaches-144-Trillion-in-2025-5166
  5. World Gold Council. How much gold has been mined? Above-ground stock around 220,000 tonnes. https://www.gold.org/goldhub/data/how-much-gold
  6. Trading Economics. Gold spot price $4,391.42 per troy ounce, 20 September 2026. https://tradingeconomics.com/commodity/gold
  7. Institute of International Finance. Global Debt Monitor Total global debt $348T at end-2025, up $29T on the year. https://www.iif.com/Products/Global-Debt-Monitor
  8. Bank for International Settlements. OTC derivatives statistics at end-June 2025 Notional amounts outstanding $846T. Gross market value $21.8T. https://www.bis.org/publications/202512-commentary-otc-derivatives
  9. Bank for International Settlements. OTC derivatives statistics: definitions Gross credit exposure adjusts gross market value for netting, and has run at 14 to 16% of market values since end-2008. https://www.bis.org/statistics/derstats.htm
  10. McKinsey Global Institute. The Global Balance Sheet 2026 Household net worth $570T. Total global balance sheet near $1.8 quadrillion. https://www.mckinsey.com/mgi/our-research/the-global-balance-sheet-2026-imbalance-and-divergence
  11. UBS. Global Wealth Report 2026 Global personal wealth above $500T after rising 10.8% in 2025. https://www.ubs.com/global/en/media/display-page-ndp/en-20260630-gwr-2026.html
  12. Jesse Myers. Bitcoin's Full Potential Valuation The widely shared global asset treemap. His own caveat: it "undoubtedly overstates some categories and understates others". https://www.onceinaspecies.com/p/bitcoins-full-potential-valuation
  13. Michael Saylor via TheStreet. Bitcoin24: $13 million per Bitcoin by 2045 Base case $280T market cap and 7% of global wealth. Bear $3M at 2%, bull $49M at 22%. https://www.thestreet.com/crypto/markets/michael-saylor-predicts-13-million-for-bitcoin-by-2045
  14. Bitcoin Magazine. Michael Saylor at Bitcoin 2025 Saylor from the stage: Bitcoin would come to represent "half of the world’s value". https://bitcoinmagazine.com/news/michael-saylor-presents-the-21-ways-to-wealth-at-bitcoin-2025
  15. Jeff Booth. The Price of Tomorrow, and subsequent interviews "Under $900 trillion in assets sits $600 trillion in debt." Technology is deflationary and the debt system cannot tolerate deflation. https://www.jeffbooth.ca/
  16. Macrotrends. M2 money supply, 1959 to 2026 US M2 median growth since 1960 is 6.6% a year, roughly 2.0% after inflation. https://www.macrotrends.net/3005/m2-money-supply
  17. World Bank via Trading Economics. World GDP Global nominal GDP $118.35T in 2025, against roughly $47T in 2005. https://tradingeconomics.com/world/gdp
  18. Official Data Foundation. S&P 500 returns since 1926 10.47% a year nominal with dividends reinvested, 7.29% a year after inflation. https://www.officialdata.org/us/stocks/s-p-500/1926
  19. CNBC. Warren Buffett's return tally after 60 years Berkshire compounded 19.9% a year from 1965 to end-2024, roughly 15.4% after inflation. https://www.cnbc.com/2025/05/05/warren-buffetts-return-tally-after-60-years-5502284percent.html
  20. BitGo and Chainalysis. How many Bitcoins are lost? 3 to 4 million BTC believed permanently inaccessible, around 1 million of them Satoshi’s. https://www.bitgo.com/resources/blog/bitcoins-invisible-burn-lost-coins-outpace-new-supply/
  21. ARK Invest. ARK's autonomous taxi model Robotaxi costs approaching $0.25 a mile at scale under Wright’s Law, against roughly $1 a mile to run a private car. https://www.ark-invest.com/articles/analyst-research/autonomous-taxi-model
  22. Deloitte and ArtTactic. Art & Finance Report 2025 Ultra-high-net-worth art and collectible wealth $2.56T in 2024. https://www.deloitte.com/lu/en/services/consulting-financial/research/art-finance-report.html
  23. Gold Eagle. Gold as a percentage of global financial assets Around 5% in 1960, a peak near 5% in 1980, a bottom of 0.19% in 2000. NOT used for either chart on this page: that series counts privately held investment gold only, so it is roughly seven times lower than the whole above-ground stock and cannot share a line with the other figures here. https://www.gold-eagle.com/article/gold-percentage-global-financial-assets
  24. World Gold Council. Gold market primer: market size and structure Gold is about 3% of global financial assets excluding central bank reserves. https://www.gold.org/goldhub/research/market-primer/gold-market-primer-market-size-and-structure
  25. CoinDesk. Bitcoin price and market cap $81,092 per BTC and a $1.63T market cap on 21 September 2026. https://www.coindesk.com/price/bitcoin
  26. Gainesville Coins. 50-year gold price history, and the Nixon Shock Gold was fixed at $35/oz until 15 August 1971, peaked at $850 in January 1980, then fell to a $279 annual average by 2000. https://www.gainesvillecoins.com/blog/gold-price-history
  27. S&P Global. SPIVA US Scorecard 92.89% of actively managed US large-cap funds underperformed the S&P 500 over the 20 years to end-2025. https://www.spglobal.com/spdji/en/spiva/article/spiva-us-year-end-2021
  28. CoinDesk and Bitbo. Bitcoin price history The 2017 all-time high was $19,783.21 on 17 December 2017. The bear market bottomed at $3,206 on 14 December 2018, an 84% drawdown. https://charts.bitbo.io/price/
  29. Mark Perry, American Enterprise Institute. Chart of the Century Price changes for 14 categories of US goods and services, January 2000 to December 2024, against overall CPI (+89.9%) and average hourly wages (+127.4%). TVs −97%, toys −72%, computer software −70.5%, cell phone service −41%, hospital services +211%, college tuition +183.8%. https://www.aei.org/carpe-diem/chart-of-the-day-or-century-8/
  30. US Bureau of Labor Statistics. Quality adjustment in the CPI BLS's own example: a laptop goes from $1,000 to $1,200 with a better processor and screen, and the hedonic adjustment records that as a price FALL. The computer index measures price per unit of quality, not what anyone pays. https://www.bls.gov/cpi/quality-adjustment/
  31. Federal Reserve Bank of New York. Credit supply and the rise in college tuition Staff Report 733. Tuition rises roughly 60 cents for every extra dollar of subsidised loans, and about 55 cents per dollar of Pell Grant aid. A 2014 NBER study found for-profit colleges eligible for federal aid charged 78% more than ineligible ones. https://www.newyorkfed.org/medialibrary/media/research/staff_reports/sr733.pdf
  32. Noah Smith, on the Baumol effect. Why do education, health care and child care cost so much? Haircuts, laundry, restaurants and car repair rose almost one-for-one with wages, which is what Baumol predicts. Health care, childcare and college rose much faster than wages, which Baumol alone does not explain. https://www.noahpinion.blog/p/why-do-education-health-care-and
  33. Santostasi and Perrenod. Bitcoin's power law Price proportional to time since the genesis block raised to about 5.69, R² = 0.961 across 17 years. A critique of the structure is at arXiv 2605.21316. https://www.newsbtc.com/bitcoin-news/when-bitcoin-could-reach-10m-power-law/
  34. Federal Reserve Bank of St Louis (FRED). CPIAUCSL and CBBTCUSD US CPI for all urban consumers, monthly from 1947, used to deflate every real figure here. Coinbase Bitcoin daily price from December 2014, used for the four-year moving average and for the 1,738 daily four-year holds measured from the 2017 peak onwards. https://fred.stlouisfed.org/series/CPIAUCSL
  35. Metalcharts. Gold price history, 1970 to 2026 Annual average gold price per troy ounce. Before August 1971 gold was fixed at $35 under Bretton Woods. https://metalcharts.org/gold-price-history
  36. Savills, Piketty & Zucman, Deloitte and the World Bank. The whole stores-of-value pie, 1960 to 2026 World property measured by Savills from 2015 ($393.3T as of the start of 2025; the 2017 report changed method, so 2015 and 2016 are restated onto the later basis using Savills’ own +6.2% for 2017). Before 2015 it is reconstructed from Piketty & Zucman’s measured housing-to-national-income ratios for 8 rich countries carried onto world GDP, which carries roughly −40%/+25% on the early years and is what the band shows. Bitcoin’s market cap is its price times a supply computed from the real halving blocks. ⚠ The $27T for art and collectibles is an aggregator figure with no published derivation; the only primary estimate, Deloitte & ArtTactic’s ultra-high-net-worth art and collectible wealth, is $2.564T for 2024. Art is 3% of the pie and moves gold’s share by under a tenth of a point. https://www.savills.com/impacts/market-trends/the-total-value-of-global-real-estate-property-remains-the-worlds-biggest-store-of-wealth.html
  37. USGS, LBMA, Maddison Project and the World Bank. Gold against world output, 1870 to 2026 Above-ground stock from USGS Data Series 140 and Mineral Commodity Summaries, back-cast to 1800 on US Bureau of the Mint world production; price from 14,686 LBMA daily fixes since 1968 and MeasuringWorth before that; world GDP from the World Bank (annual, 1960 on) and the Maddison Project Database 2023 (benchmark years only before 1950, with no 1913 and no 1929). The band before 1960 is the USGS and the World Gold Council disagreeing by 9,470 tonnes about the pre-1950 base. Every figure, and what is measured against what is reconstructed, is in src/data/gold-share/SOURCES.md. https://www.usgs.gov/centers/national-minerals-information-center/gold-statistics-and-information
  38. McKinsey Global Institute, SIFMA and the World Bank. Global financial assets, 1960 to 2026 MGI, Mapping Global Capital Markets: Fifth Annual Report (Oct 2008): "the total value of the world’s financial assets — including equities, private and government debt, and deposits — climbed in 2007 by $29 trillion in nominal terms to $196 trillion". 1980 = $12T on that same basis. ⚠ The widely quoted $54T for 1990 and $212T for 2010 come from the 2011 vintage, which counts loans and NO deposits, and must never be put on one line with the 1980 figure. The chart is built from component series (World Bank equity market cap and broad money, SIFMA fixed income) and reproduces MGI’s totals to within the broad-money-minus-deposits gap. https://www.sifma.org/resources/research/fact-book/
  39. CNBC and market data. Bitcoin's 2025 high and the 2026 correction All-time high $126,073.42 on 6 October 2025. Bitcoin fell to roughly $58,000 by June 2026 and traded at $81,092 on 21 September 2026. https://www.cnbc.com/2026/01/08/bitcoin-btc-price-predictions-for-2026.html

Figures priced 21 September 2026. Scenarios are conditional projections, not forecasts, and depend entirely on the share assumption chosen. Nothing here is financial advice.