Research · Measurement
How much Bitcoin is already exposed — and how we checked
On 13 September 2026 we measured how much Bitcoin sits behind cryptography that a sufficiently powerful quantum computer could one day break. This page records what we found, how we checked it, the mistake we made along the way and how we caught it. Everything here can be re-run against the public ledger by anyone with a Google Cloud account.
The short version
Two measurements, and a total
All figures as of block 959,193
| Measurement | Bitcoin |
|---|---|
Structural exposure The public key was written onto the ledger when the coins were created and has been visible for over a decade. Mostly early mining rewards. | 1,715,969 |
Operational exposure Coins at addresses that revealed their public key at some point by spending. | at least 3,115,944 |
Overlap between the two Counted once, not twice. | 469 |
Combined About 24.1% of all unspent Bitcoin at that height. | 4,831,444 |
None of this is a claim that any of these coins are at risk today. No quantum computer can do this now. Credible estimates of when one might are contested and range widely. We measure what is visible on the ledger; we do not forecast when it becomes dangerous.
The two kinds of exposure
In plain language
Structural
The key was never hidden. When these coins were created, the public key itself was written onto the ledger, and it has been sitting there in the open ever since — mostly from the earliest years of mining. There is no inference here, no clustering and no judgement call: you can read the key off the chain.
Operational
The key was revealed by spending. Spending from an address publishes the public key that authorised it. Where coins were left behind at that same address afterwards, the remainder now sits behind a key everybody has seen.
The floor, not the estimate
Why the operational figure says “at least”
For coins locked behind a script rather than a plain address, what a spend actually publishes depends on the script — and the script cannot always be read from the ledger alone. So the figure is a floor by construction rather than an estimate with error bars either side of it.
We count only the cases we can prove. We exclude every case we cannot. The true operational figure is higher than ours, and we do not know by how much.
How we checked ourselves
Two checks, in this order
Against an independent source
Glassnode published their own measurement on 20 May 2026.
| Glassnode | Theos | Difference | |
|---|---|---|---|
| Structural, on Glassnode’s definition | 1.92M | 1,939,733 | +1.0% |
| Operational | 4.12M | at least 3,115,944 | −24% |
The structural figure in that table is 1,939,733, not the 1,715,969 in our headline. That is deliberate. Glassnode’s definition is wider than ours — it also counts two other forms where a key, or a modified key, sits on the ledger — so to compare fairly we measured their definition as well as ours. The 1,715,969 is what we are prepared to claim; the 1,939,733 is what the comparison requires. On the same definition, two independent pipelines land within one percent of each other. On operational exposure we are 24% below them, and we are not going to explain that away: our measurement point is two months after theirs and operational exposure only grows, so the gap is definitional — they include categories we deliberately exclude.
Against a limit that cannot be crossed
Agreeing with somebody else is reassuring but weak: it only detects errors bigger than the distance between your definition and theirs. The stronger check is arithmetic that cannot be argued with — the total amount of Bitcoin we say is sitting unspent must be less than all the Bitcoin that has ever been created.
| Bitcoin | |
|---|---|
| Total unspent, by our measurement | 20,059,952.29 |
| Total ever mined, by the ledger’s own records | 20,378,230.81 |
| Headroom | 318,278.52 |
It passes.
The mistake
What we got wrong, and how we found it
Our first set of figures added up to 21,242,325 Bitcoin. That is more Bitcoin than can exist. The measurement was wrong before it was interesting.
The cause was not our software. Google’s public copy of the Bitcoin ledger — the dataset we were querying — is missing 1,986 blocks, between heights 959,194 and 966,545. Everything before that point is complete: 99.21% of the ledger. We fixed it by stopping every measurement at block 959,193, the last height before the gap begins. The supply check then passes. We published nothing while we were wrong.
Three consequences
- Anyone using this dataset inherits the same defect, silently.
- Comparing against another company's number would never have caught it — our error was about 4% and it hid comfortably inside an 11% definitional gap.
- The rule we adopted as a result: a measurement is not finished until it has been checked against a bound that cannot be exceeded — not merely against somebody else's number.
Reproducibility
Anyone can re-run this
Source: bigquery-public-data.crypto_bitcoin, Google’s maintained public copy of the Bitcoin ledger. Every query, what it scanned and what it cost is recorded in our repository. The whole exercise cost about US $22 in cloud query charges, roughly half of it spent on measurements that turned out to be wrong and the diagnostics needed to understand why. The two queries that located the defect cost a fraction of a cent between them.
What this is for
This is not a product
It is evidence about how we work. Our product inventories cryptography inside an organisation’s own systems. A prospective customer is not buying a number from us; they are buying the belief that when we tell them what is in their estate, it will be true — including the parts they would rather not hear. A larger, scarier number would be easier to produce and much easier to dismiss.
Refresh policy
When these numbers change
These figures describe the ledger at block 959,193 (mined in late July 2026). Structural exposure moves very slowly and will be refreshed quarterly. Operational exposure moves constantly; a refreshed figure will always name its height. Any figure on this page without a height next to it is out of date and should be treated as such.
Measured by Theos Quantum Technologies, 13 September 2026. Method and full query record: available on request.