Loss evidence and its limits

How Much Bitcoin Is Actually Lost?

The 21 million supply ceiling is not a measurement of spendable or liquid coins. Unspendable scripts, inferred burn addresses, lost keys and dormant balances require different evidence. This page compares those distinctions, links the burn-address study, and flags historical estimates whose original sources have not yet been verified. Coins that have not moved are not necessarily lost.

Short answer: there is no verified total of all lost Bitcoin. One study reports 3,197.61 BTC at addresses classified as burn addresses through block 840,682 (April 24, 2024). That is an address-based inference, not a mathematical proof or a count of all unspendable scripts. Broader lost-key estimates and dormant-supply figures measure different things; inactivity does not prove loss.

Bitcoin loss evidence: lost keys are not directly observable; dormant does not mean lost; estimates are not proof. Embed this badge on your site →

Last updated: 2026-09-03. This is not a live supply feed. The linked study has its own block-height cutoff; the separately labeled historical claims remain unverified.

Why there is no verified supply split here

Scheduled issuance Spendable supply Liquid supply

Conceptual distinctions only; not a quantitative chart. Subtracting an uncertain loss estimate from a supply figure measured at a different date does not establish spendable supply. Even spendable coins may be held off the market, so the remainder cannot be labeled liquid.

A linked study and historical claims awaiting verification

El Khatib & Legout (arXiv:2503.14057v1)— through block 840,682 (Apr 24, 2024); published Mar 2025 ~3,197.61 BTC

The authors use manual review and machine learning to classify handmade-looking addresses as burn addresses. Section 2.1.2 cautions that address unspendability is an extremely strong probability claim, not a mathematical proof. This reported total is not an OP_RETURN census or a count of all lost Bitcoin; this site has not independently reproduced it.

Attributed to Chainalysis— original source and date unverified Unverified: ~2.78M–3.79M BTC

Not verified for citation. This range appeared in an earlier version of this page without an original-source link. Its attribution, date and methodology have not been checked against the original research. It is not a current measurement or an input to a supply calculation here.

Glassnode ("Inert Supply")— metric definition verified; historical amount and date unverified Unverified: ~5.25M BTC

Not verified for citation. This warning applies to the historical amount and observation date, not the metric definition. Glassnode's Adjusted MVRV documentation defines Inert Supply as the sum of coins last active 7–10 years ago and more than 10 years ago. This is a dormancy classification; its interpretation allows both lost coins and long-term holding. Its HODL Waves guide explains that spending resets coin age, which does not necessarily mean a change in ownership. The approximately 5.25M BTC value inherited by this page has not been checked against a dated original dataset and is not used in a supply calculation here. Inactivity does not prove that keys are inaccessible.

Attributed to CZ (Changpeng Zhao)— original statement and date unverified Unverified: ~2.0M–4.0M BTC

Not verified for citation. The earlier page included this attribution without a link to the original statement. Neither its wording nor its supporting calculation has been verified. It should not be treated as an independent loss study or a current mined-supply measurement.

Coins attributed to Satoshi Nakamoto— original analyses and dates unverified Unverified: ~700K–1.13M BTC

Not verified for citation. The earlier page combined several early-miner estimates without linking their original analyses. This range has not been reconciled against those sources. Miner attribution, ownership and loss are separate questions; this page has not verified that all coins in any such grouping remain unmoved or inaccessible.

Why the range is this wide

Measuring a balance or its age is not the same as proving its keys are lost. Researchers infer loss from behavior such as long inactivity; the burn-address paper instead infers deliberate construction from address patterns. A long-term holder and someone who lost a key can leave the same on-chain evidence. These categories may overlap and should not be added together.

Addresses and scripts require different evidence. The paper's 3,197.61 BTC figure is based on address classification, not script-level proof. By contrast, Bitcoin Core v29.0's CScript::IsUnspendable() recognizes a scriptPubKey beginning with OP_RETURN or exceeding MAX_SCRIPT_SIZE as guaranteed to fail. Merely finding an OP_RETURN byte somewhere in a script is not that test. This page provides no verified chain-wide total of script-unspendable outputs.

What this means for "21 million"

Want to check the issuance arithmetic separately? Reproduce the scheduled subsidy ceiling with a CSV and dependency-free Python. That resource calculates the rules, not a lost-coin or spendable-supply total.

The site is called Twenty-One Million because of the issuance ceiling explained in Chapter 1. Lost coins do not get reissued or alter the subsidy schedule. They reduce spendable supply, but subtracting a dormancy figure or uncertain loss estimate from the cap does not establish how many coins will ever be usable, or how many are liquid today.

Source coverage and limitations

The burn-address study and Bitcoin Core rule are linked above. The four historical estimate rows are explicitly unverified: they are retained to identify claims needing original-source checks, not to endorse them or provide a complete survey. No live loss measurement, automatic refresh or independent reproduction is provided. Consult the linked paper for its methods and limitations, and do not cite the unsupported rows as established results. For an explanation of ways coins can become inaccessible, see Chapter 11.

Common questions, answered directly

How much Bitcoin has been lost forever?

There is no verified total of all lost Bitcoin. El Khatib and Legout report 3,197.61 BTC at addresses classified as burn addresses through block 840,682 (April 24, 2024), not a mathematical proof of loss. Lost-key estimates and dormant-supply measurements describe different, potentially overlapping populations; inactivity alone does not establish that keys are gone.

How much BTC is provably lost versus just probably lost?

A scriptPubKey beginning with OP_RETURN is a script-level example of a provably unspendable output. This page does not provide a verified chain-wide total of such outputs. The paper's 3,197.61 BTC figure instead uses manual review and machine learning to classify handmade-looking addresses. The authors caution that address unspendability is a probability claim, not a mathematical proof.

Were millions of bitcoin deliberately burned?

The cited burn-address study reports 3,197.61 BTC in its classified addresses, not millions. It does not establish a complete total of every deliberate burn. Multi-million-BTC loss or dormancy estimates are not evidence that millions were deliberately burned.

Does lost Bitcoin change the 21 million supply cap?

No. Lost coins do not get reissued or change the subsidy schedule. Loss affects how much issued Bitcoin remains spendable, but neither dormant balances nor inferred loss estimates establish an exact present or future spendable supply.