Twenty-five things the chain says

Two out of every three bitcoin mined in 2009 have never been spent. 21,488 of that year’s 32,490 blocks still hold their 50 BTC reward exactly where it was created, seventeen years on: 1,074,402 BTC in all.

Nothing like it has happened since. Of everything mined in 2010, 16.9% has never moved. Of 2011, 1.5%. Of 2019, 0.011%.

CaveatNever moved is not the same as lost. The chain records movement, not whether anybody still has the keys.

Check it Supply aged over 10 years Supply by age

The longest any bitcoin has ever sat untouched is sixteen years. A 50 BTC mining reward from 2 July 2010 was spent on 30 July 2026, 5,871 days later. Nothing in Bitcoin’s history has gone longer between being created and being spent.

Block 63,853 to block 960,174, if you want to look it up.

CaveatThis record is alive. It gets broken roughly whenever another 2010 coin wakes up, so it is true of the chain as we measured it on 4 August 2026 and not for ever.

Check it Spent volume aged over 10 years

Bitcoin’s oldest coins almost never move. In the whole of 2025, out of 53,082 blocks, exactly eighteen contained a coin that had last moved in 2010 or earlier. You could list them by hand.

There are 1.84 million bitcoin sitting in that state. 465 of them moved in 2025. In 2024 it was 9,863 across 39 blocks, and in the first seven months of 2026, 2,452 across 13.

Check it Are old coins moving?

Bitcoin has two populations and almost nothing in between. The typical coin that gets spent had been sitting still for about six hours. The typical bitcoin in existence last moved 664 days ago.

Coins that move, move fast. Coins that sit, sit for years: a quarter of all bitcoin has not moved for more than seven years.

Check it Supply by age Dormancy

A third of all bitcoin has not moved in five years, and that is the most there has ever been. The share set its record of 33.63% on 29 July 2026 and stood at 33.56% when we measured, six days later.

The ten year line is doing the same thing: 17.7% of all bitcoin has now sat still for a decade, a record set on 3 August 2026. A coin can only join that group by sitting still for ten years, so for the record to keep moving, more coins must be crossing the ten year line than are being spent back out of it.

CaveatA live claim. Both shares move with every block, and by block 961,000 the ten year share had already slipped a hair under its own record. These are the values on 4 August 2026. Do not quote either without the date.

Check it Supply untouched for a year Long-term holder share Holder cohorts

Bitcoin is not made of accounts. It is made of separate unspent chunks of coin, called outputs, and there are fewer of them every month. The count peaked at 188.2 million on 21 December 2024 and was 165.9 million on 4 August 2026, down 11.9%, while the number of bitcoin kept going up.

It has happened before, and worse. The same count fell 24% through 2018, twice as far as this one has so far.

CaveatA live figure, measured at block 961,000. The shape is in the data. The cause is not, and finding it would need guesswork we have not built, so we do not offer one.

Check it UTXO count UTXO set change

More than half of all bitcoin sits in about 23,000 unspent chunks. Those 22,794 outputs hold 11,503,955 BTC between them, 57.3% of every bitcoin there is. An output is not a person: one owner can control thousands of them, and a single output can be an exchange balance belonging to millions of customers. This counts outputs and nothing else.

The largest single unspent output holds 130,010 BTC and was made in June 2025. Every one of the 226 larger outputs ever created has since been spent. At the other end, 148.8 million outputs hold under 0.01 BTC each, and 0.43% of all the bitcoin there is between them.

CaveatWe hold no address grouping and no exchange labels, on purpose. Nothing on this page counts people, and reading these figures as a split of wealth between owners would be wrong.

Not charted yet

“A block every ten minutes” is the most repeated fact about Bitcoin, and it is not what usually happens. Half of all gaps between blocks are shorter than 6 minutes 42 seconds, and one gap in ten is under a minute. Ten minutes is the long-run average, which is a different thing.

Blocks do not even agree on the time. 15,824 of them, one in sixty, are stamped earlier than the block they were built on, one of them by nearly two hours, and 1,724 of those are from 2020 or later.

CaveatA block’s timestamp is a clock reading typed in by whoever mined it. The rules only require it to beat the median of the previous eleven blocks, so the backwards ones are the system working as designed rather than a bug. Every measurement on this page that uses time inherits that.

Not charted yet

The fullest block Bitcoin has ever produced holds five transactions. One of them is a 3.97 MB video file, which paid €6,493 to sit on the chain for ever. The other four are ordinary and small.

Block 836,964, mined on 30 March 2024, 3,993,936 bytes. Everyone pictures a full block as thousands of payments. The fullest one ever is a video and four bystanders.

CaveatCalling it a stored file is a pattern match. The 3.97 MB of data and the label video/mp4 are literal bytes inside the block, so the claim holds even if you distrust our detector.

Check it Block size Witness share of size

People store files inside Bitcoin. The chain is 759 GB, and 35.6 GB of that is files written into it deliberately, 24.95 GB of it pictures. Counting the transactions that carry them: 100,249 hold a video, 1,927 an audio clip, 681 a PDF and 112 an e-book.

Almost everything else is tiny. 111 million transactions carry nothing but a scrap of plain text, usually a few dozen bytes.

CaveatThese files are found by matching a byte pattern, not by validating a protocol. The file type is read from inside that pattern and read once per transaction, so the figures above count transactions rather than files, and 257,798 transactions declare no file type at all and are left out of the breakdown.

Check it Inscription body bytes Inscription count

Bitcoin’s payment traffic has not grown since 2023. Every month for 43 months, the number of transactions that move money has stayed between 5.8 and 9.2 million. Over the same 43 months, transactions that carry data instead of money moved by a factor of 122.

It reframes every transaction-count chart in the industry: the line quoted as “Bitcoin usage” is largely a data line sitting on top of a flat one.

CaveatThe split between money and data uses our own shape labels and our own pattern detectors, and a transaction that carries data and also pays somebody is counted here as data. The flat band survives that: plain payments on their own ran between 5.3 and 8.6 million a month, the same narrow band.

Check it What blockspace was used for Transactions per block

A third of all the room Bitcoin has ever sold went on housekeeping rather than on paying anybody. Transactions that sweep five or more chunks of coin into one or two are only 3.7% of every transaction ever made, and they take 33% of all the space ever used.

CaveatThis is the most heuristic label on the page. Five in, two out catches genuine clean-ups and also any large payment funded from several coins, and we cannot tell them apart without owner labels we do not hold. The shape itself is not a guess: 3.7% of transactions really do use 33% of the space, whatever you decide to call them.

Check it Consolidation weight share

The story is that JPEGs priced people out of Bitcoin. Per byte of space, they have always paid less than payments do. Over all of history a payment paid 71 satoshis a byte and a stored file paid 25. A satoshi is a hundred millionth of a bitcoin. In 2026 a payment pays 3.8 and a file 0.4, a tenth as much. Files are patient, so they wait and fill the gaps.

The honest counterweight, from the same numbers: in 2023, during the first wave, the two paid the same rate. It is 2025 and 2026 where files fell to the bottom of the market.

CaveatThis depends on our classifier. The rate is one big sum divided by another, so busy blocks weigh heaviest, which if anything makes stored files look expensive rather than cheap.

Check it Inscription fee against weight share What blockspace was used for

The most an ordinary bitcoin payment has ever cost is €242.84, in one block on halving day in April 2024. That is the whole horror story. In nine blocks out of ten across Bitcoin’s priced history the same payment cost under €1, and in more than a quarter of them under one cent.

CaveatThis is what patient users actually paid in that block, the middle of the fees that got in, and not the price of jumping the queue. A standard payment is taken to be 141 bytes throughout, which is the real size of the commonest kind.

Check it What blockspace cost Median fee rate

Miners are paid in two ways: newly created coins, and fees. In seventeen years and 961,001 blocks, the fees have beaten the new coins exactly 351 times. That is one block in 2,700.

A third of them happened in a single day. On 20 April 2024, the halving, 120 of the day’s 130 blocks earned more from fees than from new coins, and €75.7 million of fees were paid inside 24 hours.

CaveatMeasured against the schedule of new coins rather than against what the miner actually claimed, because a small number of blocks claim less than they are owed. Counting it the other way also gives 351.

Check it Fee share of revenue Block fees

The loudest Bitcoin argument of 2025 was about a rule capping how much data you may attach to a transaction as a note. Data over that old 83 byte cap has never been more than 0.065% of the chain’s bytes in any month, before the rule changed or after it.

The other way of putting data on the chain was never limited at all, and nobody fought about it. Over the whole chain the fought-over notes hold 5.17 GB and the unlimited route holds 39.78 GB, roughly seven to one.

CaveatThat 0.065% counts the data itself. Counting the whole transactions carrying it, the same month reaches 0.12% of the chain’s bytes, which is the wider and equally true reading. The 83 byte line is our stand-in for the old rule, and we make no claim about why the numbers are flat.

Check it OP_RETURN payload against inscriptions Data on the chain

In October 2022 bitcoin was worth less than the market had paid for it, and at the same time worth more. It depends which currency you count in. In all 4,743 blocks of that month the price sat below the market’s dollar cost basis, and above its euro one. Same coins, same month, opposite answer.

“The market is underwater” was the most repeated onchain claim of that month, and for anybody counting in euro it was not true for a single block.

CaveatThis sits in the clean price years and uses euro order books throughout. It is not a photo finish either: at its closest, the euro price was still 1.4% above the euro cost basis.

Check it Realized price Am I in profit?

Bitcoin’s swings are getting smaller. Divide the price by what the market actually paid for its coins and you get one number: at the 2013 top it was 6.68, so the price was nearly seven times the money that had gone in. Every cycle since has peaked lower, at 4.78, then 4.04, then 2.82, and every low has come in higher, at 0.57, 0.69, 0.75, 1.10. Eight numbers, all moving the same way.

CaveatPartly mechanical rather than a change in behaviour: what the market has paid only ever goes up in cash terms, so each cycle starts from a bigger base and the same price move gives a smaller ratio. The 2024 to 2026 window is still open and its low can still move. 2011 is left out because 44.8% of that year’s blocks carry a price we mark as poor. This describes what happened, and says nothing about a fifth cycle.

Check it MVRV Cycle position

Every bitcoin crash has been shallower in euro than in dollars, six times out of six. The 2022 crash was 77.1% in dollars and 74.4% in euro. The 2014 to 2015 one was 84.8% against 82.3%. It has never gone the other way.

The reason is one sentence: the dollar goes up in exactly the frightened periods when bitcoin goes down, so a euro holder’s loss is cushioned.

CaveatEach currency is measured against its own record high. The 2011 crash, 93.6% against 93.0%, rests on poor prices and a thin euro market and is counted only for completeness. The 2025 to 2026 fall is still running.

Check it Price against its all-time high

Bitcoin’s price fell 77% in the 2022 bear market. The money actually paid for the coins fell 19% over the same stretch, and has never fallen more than 24% in the entire history. Money that goes into bitcoin does not come back out at anything like the speed the price does.

CaveatPartly true by definition, and worth saying rather than dressing up: the money-paid figure can only fall when somebody sells at a loss. The 24% record belongs to 2011, our weakest price year, so 19% is the figure to hold on to.

Check it Realized cap against market cap Market value

On 26 January 2014 a bitcoin on the Mt. Gox exchange cost 23% more than one anywhere else. Twenty-six days later it cost 79% less: $119 on Gox against $568 on the exchanges still working. The premium crossed zero on 7 February.

Everybody knows Gox collapsed. Almost nobody has the numbers, because the site that published its trade record is dead and the surviving public copy of the file is empty. We hold a checked mirror of all 8,295,809 trades.

CaveatOur cross-exchange price is built to throw Gox out during exactly this window, so the two lines are not independent. The honest reading is Gox against the exchanges that were still working, not Gox against the true price.

Check it Data quality Price quality

We find stored files by matching a byte pattern, not by asking an oracle, so here is that pattern sitting an exam. Run over all 961,001 blocks it fires zero times before block 767,430, and its first hit is inside block 767,430: the block that carried the first such file ever made.

The detector for runestones, a different kind of chain data, is less clean and we publish that too. 28,334 of its 184 million hits come from before runestones existed, which is a measured false alarm floor of 0.0154%.

Caveat0.0154% is a floor, not the rate. The true false alarm rate after the launch could be higher and cannot be measured this way. Everything on this page that counts stored files or runestones inherits both figures.

Check it Inscription count Runestone transactions

There is no usable bitcoin price for 2010, and our own data says so out loud. 99.95% of that year’s blocks carry a price we mark as poor, and the typical one is 83 minutes out of date, because for most of that year nobody traded a bitcoin at all. From 2015 on, every single block carries a full cross-exchange price.

It matters far less than it sounds, and we publish that too. Of all the money the market had paid for its coins by 2015, 0.56% had gone in at one of those poor prices. Today that share is two ten-thousandths of one percent.

CaveatPoor means the nearest real trade was a long way from the block in time. It is a diagnostic of our own pipeline, not a claim about the world. No finding on this page quotes a price level from before 2012, and the three that touch 2011 at all say so in their own caveat.

Check it Price quality Realized cap on degraded prices Data quality

1,555,963 bitcoin, 7.75% of all of them, were mined before bitcoin had a market price at all, and have never moved since. Every sum on this site that asks what the market paid counts them at exactly zero, which drags the average down. Published, the market paid $52,776 a coin. Leave those coins out and it is $57,213, 8.4% higher.

In euro: €46,327 published, €50,222 leaving them out. Almost no dashboard tells you how big this assumption is.

CaveatLeaving those early blocks unpriced rather than filling in a zero is a stated policy, and the same one Coin Metrics and Glassnode use. It is not a hole in the data. We publish the size of the assumption so you can judge it yourself.

Check it Supply with no cost basis Data quality