Bitcoin Exchange Reserves: What the Number Actually Measures

Exchange reserve charts are among the most quoted on-chain metrics there is. Reserves at a six-year low, coins leaving exchanges, supply shock incoming. The charts are usually built carefully by people who know what they are doing, and the underlying blockchain data is genuinely public and genuinely verifiable.

The part that is not verifiable is the bit that decides the answer. Bitcoin has no field marking an output as belonging to Coinbase or Binance. Every reserve number in existence is a sum over a list of addresses that somebody attributed by inference, and the list is not published. So we went and measured what that dependency is actually worth.

What is a Bitcoin exchange reserve?

It is the total balance held by the addresses attributed to a given exchange, computed from the public chain at a point in time. The chain supplies the balances exactly and without ambiguity. What it does not supply is the attribution: which addresses belong to which exchange. That mapping is built by watching deposit and withdrawal patterns, clustering addresses that co-spend, and matching against addresses exchanges have acknowledged. It is inference, and different providers infer differently.

This is why two reputable dashboards can show different reserve totals for the same exchange on the same day, and both be computed correctly. They are summing different lists.

How are Bitcoin exchange reserves calculated?

In three steps. First, assemble an address set for the exchange. Second, compute each address's balance from the indexed chain, which is funded outputs minus spent outputs. Third, sum them and plot the total over time. Steps two and three are arithmetic anyone can reproduce. Step one is the entire product, and it is the step that is never shown to you.

To put a number on how much step one matters, we pulled the current balances of five widely reported exchange cold addresses on 16 August 2026, when the chain tip was block 962,803. We are not claiming first-hand knowledge of who controls them. They are addresses that public lists and reporting commonly attribute to major exchanges, which is precisely the point being tested.

Two things fall straight out of that table.

The first is concentration. One address is 37 percent of the set, and the top four are 99.99 percent of it. A provider whose list includes that one address and a provider whose list does not will report reserve totals roughly 248,000 BTC apart. Not because either made an arithmetic error, but because attribution is a judgment call and they made it differently.

The second is staler than it looks. The fifth address has 3,329 transactions in its history and holds 0.0139 BTC. Its most recent transaction was in block 767,597, on 16 December 2022. It has been empty and untouched for over three and a half years. It still appears on public exchange address lists.

Why do exchange reserve charts disagree?

Because a reserve series has three moving parts and only one of them is the blockchain. Providers differ on which addresses are in the list, on when an address is added or removed, and on whether internal movements between an exchange's own wallets count as flow. Any of the three can move a chart without a single coin changing owner.

The last one deserves attention, because it is the same failure mode we measured in on-chain transaction volume, where 94 percent of a block's headline volume was change returning to the sender. Exchange flow metrics inherit the identical problem. When an exchange consolidates UTXOs or rotates cold storage, coins move from an address the provider knows to an address it may not know yet. The chain sees an outflow. Nothing left the exchange. If the destination address is later attributed correctly, the reserve series gets revised and the "outflow" quietly disappears.

That 2022-dormant address is the same trap running the other way. When it was drained, the coins went somewhere. If the receiving address was never attributed, that exchange's reserves appear to have permanently dropped by the amount, and the chart shows an exodus that was a wallet migration.

Do falling exchange reserves mean people are buying?

That is the interpretation usually attached to the chart, and it is not one we are going to make for you. What we can say is what the data does and does not support. A declining reserve series is consistent with coins moving to self-custody. It is equally consistent with an exchange rotating wallets, moving funds to a custodian, restructuring after an acquisition, or a provider quietly dropping an address from its list. The chart cannot distinguish between those, because all four look identical on-chain.

Anyone claiming otherwise is adding an interpretation the measurement does not contain. This is informational on-chain data, not investment or trading advice, and the honest position is that reserve movements raise a question rather than answer one.

Is proof of reserves the same as exchange reserves?

No, and the distinction is worth keeping straight because the names are close enough to blur. An exchange reserve estimate is built by outsiders inferring which addresses an exchange controls. A proof of reserves is built by the exchange itself, which proves control of specific addresses by signing messages with the corresponding keys, usually combined with a Merkle tree of customer balances so users can verify their own account is included.

Proof of reserves solves the attribution problem, because the exchange is telling you which addresses are theirs and proving it cryptographically. What it does not solve on its own is liabilities. Proving you hold 100,000 BTC says nothing about whether you owe 150,000. That is why serious attestations pair the asset proof with a liability commitment. We cover the mechanics and how to query the address side of it on the Bitcoin proof of reserves API page.

How to work with reserve data without being misled

Three habits cover most of it, and none require you to build your own labeling operation.

Ask any provider for the address count behind a series, not just the total. A reserve figure that moved 15 percent while the address set also changed is a different event from one where the address set was constant. If they cannot tell you whether the list changed, the series cannot be read as a clean measurement.

Check concentration before you trust a delta. As the table shows, a handful of addresses carry nearly all the balance, so a single large address being added, dropped or drained dominates everything else. It is the same reason a single self-transfer distorted a whole block's volume figure. Concentrated data is fragile data, and the instinct generalizes well past crypto: nobody sensible values a company off a headline revenue figure without seeing what the underlying numbers are built from.

Separate internal movement from real flow. If you can query the chain yourself, an exchange consolidation is usually visible: many inputs, few outputs, and the destination going on to behave like the source did. Flow metrics that do not net out internal transfers will report it as customers withdrawing.

Doing this yourself

Everything above was computed with public data. Balances came from a public Esplora index, one request per address, reading funded outputs minus spent outputs from the address summary. That is fine for five addresses. It stops being fine at the scale a real reserve series needs, which is thousands of addresses re-evaluated every block, plus the history to compute a series rather than a snapshot.

That is the work an indexed chain does for you. On BitcoinDatabase the address set is a query, not a crawl: you can pull balances for a list of addresses in one call, attach entity labels to see which are attributed and to whom, and rebuild the whole series in SQL against your own address list rather than somebody else's. Flow in and out of a labeled set is available directly through the exchange flow API, and if you need a balance as of a past date rather than today, that is what the address balance history API answers.

If you are choosing between providers for this specifically, the practical question is whether you can see and change the address list, or only read the finished chart. Our CryptoQuant alternative page compares that directly, including where API access sits in each plan structure.

All figures in this article were measured on 16 August 2026 at chain tip 962,803 from a public Bitcoin index, and every address is printed above so you can verify the balances against any block explorer. Address attributions are as commonly reported publicly, not claims of first-hand knowledge of who controls them. This is informational on-chain data only, not financial, investment or trading advice.