There is a number circulating through Bitcoin research circles that deserves far more skepticism than it is currently receiving: more than half of all circulating BTC sits in addresses with what analysts classify as "unknown ownership." The figure, attributed to Galaxy Digital's head of research Alex Thorn, has been quietly absorbed into the market's collective mental model as evidence that Bitcoin remains fundamentally decentralized — that the whales have not won, that self-custody is alive, that the ETF era has not swallowed the asset whole.
I want to be precise about what that number is and what it is not, because I have spent enough hours inside block explorers and raw node data to know that "unknown" is not a synonym for "mysterious." It is, far more often, a synonym for "unlabeled" — and unlabeled is a statement about the tooling, not the chain. The distinction sounds semantic. It is not. It determines whether we read that sentence as a decentralization victory or as an admission of measurement failure. Pattern recognition precedes prediction, and the first pattern to recognize here is that a bucket defined by the absence of a label cannot tell us who is inside it.
Let me reconstruct the evidence chain, step by step, the way I would reconstruct any attribution claim — starting with what the chain actually records, then moving to what the analysts actually do, and ending with what the market is actually pricing.
Context: How On-Chain Attribution Actually Works
Bitcoin does not have accounts. It has Unspent Transaction Outputs — UTXOs. When you receive bitcoin, you are not credited in a ledger that bears your name; you receive the right to spend a specific output, and you spend it by consuming it and creating new outputs. A single person, running a single wallet, can accumulate thousands of these outputs across thousands of addresses. There is no registry that maps a human being to an address. The chain records keys and amounts. It does not record identity.
This is the foundational constraint that every ownership statistic inherits. To say that "an address holds X bitcoin" is trivially true and analytically useless, because the unit of economic interest is not the address — it is the entity. An entity is a person, a fund, an exchange, a custodian, a miner, a lost key. The entire discipline of on-chain analytics exists to bridge the gap between the two, and that bridge is built from heuristics, not from facts.
The most important heuristic is Common-Input Ownership. The logic runs like this: if a single transaction consumes several UTXOs as inputs, then whoever signed that transaction controlled all of those UTXOs simultaneously. Therefore all the addresses contributing to those inputs almost certainly belong to the same entity. Chain the logic forward — if address A and address B are co-spent, and B is later co-spent with C, then A, B, and C are clustered — and you can grow entities from individual addresses into sprawling clusters that occasionally contain hundreds of thousands of outputs.
This is the standard method. It is also, by the admission of everyone who uses it, imprecise. It fails whenever an entity deliberately avoids co-spending, which is exactly what sophisticated custodians and privacy-conscious holders do. It fails when a single transaction legitimately combines inputs from multiple parties — as in CoinJoin, or in a custodial sweep. It fails at the boundary of every exchange, where thousands of user deposits are commingled into hot wallets that the exchange then spends from on behalf of everyone. An exchange hot wallet is not one entity. It is a crowd wearing one mask.
On top of clustering sits the label database. Glassnode, Chainalysis, Arkham, Nansen, and Galaxy Research each maintain proprietary maps of which clusters belong to which known actors. An address gets labeled when it can be tied — through a deposit, a published address, a KYC disclosure, a subpoena, a leak, or a heuristic of sufficient confidence — to a named counterparty. Everything that cannot be tied stays in the residual bucket. That residual bucket is what the market is now calling "unknown ownership."
Alex Thorn's claim, as reported, is that this residual bucket contains more than half of circulating BTC, and that the majority of it is "likely" self-custodied by individuals. Note the hedging. The first half of the sentence is an output of a database query. The second half is an interpretation of that output, and the word "likely" is doing an enormous amount of unexamined work. Galaxy Digital is a serious institution; Alex Thorn is a respected researcher; Galaxy Research produces high-quality work. None of that changes the fact that we are looking at a single-source, methodology-opaque claim, and I want to be careful to treat it as such.
Core: The Anatomy of the Unknown Bucket
The residual is not one thing — it is four things wearing the same coat
When an analyst tells you that half of Bitcoin is "unknown," you are not being told about a homogeneous group of shadowy holders. You are being told that a database has failed to resolve a residual into its components. The residual is, in practice, a mixture of at least four populations, and each carries a completely different market meaning.
The first population is genuine individual self-custody: the hardware wallet in a drawer, the single-key paper backup, the multisig vault, the sovereign holder who moved coins off an exchange and never looked back. This is the population Thorn's interpretation privileges, and it is real. It is also the population that has grown measurably since the 2022 collapses, when a generation of holders learned — expensively — that counterparty risk is not theoretical.
The second population is lost coins: bitcoin whose private keys have been destroyed, forgotten, or buried with their owners. These coins will never move again. They are functionally burned. Their market meaning is the opposite of self-custody in one crucial respect — they do not represent latent supply that could return to the market, they represent permanent supply reduction. If the unknown bucket is full of lost coins, the conclusion is deflationary. If it is full of self-custody, the conclusion is neutral-to-tight. Same bucket, opposite implications.
The third population is unlabeled institutional cold storage: coins held by custodians, family offices, funds, and treasuries whose addresses have never been publicly tied to them. These coins are not lost and not sovereign. They are latent sell pressure with a signature that the label databases simply have not captured yet. A custodian that has never published an address, never been subpoenaed, and never leaked is invisible to every heuristic in the industry.
The fourth population is the historical residue: the coins attributed to Satoshi Nakamoto, which have never moved in more than fifteen years, and the early-era coins whose original miners are long gone. These are quasi-permanent. They belong to history more than to the market. History is written in blocks, not promises, and this slice of the chain is the clearest example of that principle.
When a headline collapses all four of these into "unknown ownership," it destroys the only information that mattered. The number "50%" is compatible with a Bitcoin that is overwhelmingly sovereign, with a Bitcoin that is catastrophically hoarded by invisible institutions, and with a Bitcoin where a fifth of the supply is simply gone. The statistic cannot distinguish between them. A statistic that cannot distinguish between opposite realities is not a finding. It is a prompt for further work.
The lost-coin estimate is the buried variable
The single most consequential sub-question inside the unknown bucket is how much of it is lost. Chainalysis and other research houses have periodically estimated the lost-coin stock at roughly 3.7 million BTC, with credible ranges stretching from 2 million to 6 million depending on the assumptions baked into the model. I want to flag the confidence level explicitly: this is an industry estimate, not a chain-verified fact. You cannot read a blockchain and see a lost key. You can only infer loss from dormancy, and dormancy is ambiguous — a coin that has not moved in a decade might be lost, or it might belong to a holder with diamond hands and a working memory.
Here is why the lost-coin variable matters so much. Bitcoin's nominal issuance is currently running at roughly 0.8% per year, post-halving, and trending lower. But the supply that is actually available to trade is not the nominal supply. It is the nominal supply minus the coins that will never move. If a meaningful fraction of the unknown bucket is genuinely lost, then Bitcoin's effective free float is materially smaller than the headline 19.7 million circulating coins suggests — and a smaller float means the same dollar of marginal demand moves the price further. This is a structurally bullish variable that almost never appears in price models, because it cannot be measured directly. It can only be estimated, and estimates do not fit cleanly into spreadsheets.
I learned to respect this class of ambiguity during a forensic exercise of my own. When I traced the on-chain flow of funds out of Anchor Protocol during the final seventy-two hours before Terra's collapse, I was struck by how much of the drama was invisible in real time and obvious only in retrospect. The chain recorded every transaction. What it did not record was intent. Distinguishing a panic withdrawal from a routine rebalance required me to build a timeline and hold two hypotheses simultaneously until one of them collapsed. The lost-coin question is the same kind of problem, scaled to fifteen years. The chain shows us dormancy. It does not show us death. We are inferring death from silence, and silence has many causes.
The coverage gap is the real headline
If I had to compress the entire Galaxy claim into one sentence, it would not be "half of Bitcoin is self-custodied." It would be this: the industry's label databases do not cover half of Bitcoin's supply. That is a statement about instrumentation, and it is far more defensible than the ownership inference layered on top of it.
Consider how a label database grows. It grows through exchange disclosures, through published cold-wallet addresses, through enforcement actions that reveal addresses in court filings, through voluntary tagging by large custodians, and through heuristics that flag behavioral patterns. It does not grow through the cooperation of private holders. A sovereign individual who never touches a KYC exchange after their initial purchase is, from the database's perspective, a permanent blank. Their coins sit in the unknown bucket forever, not because they are mysterious, but because they never had a reason to identify themselves.
This is why the coverage gap should be expected to be large, and why the "50%" figure is not surprising. It is roughly what you would predict from first principles. The surprise would be if the unknown bucket were small, because that would imply either mass identification of private holders — which does not happen — or a label database of implausible reach.
The more interesting question is whether different research houses agree on the size of the gap. They do not, and this divergence is itself the finding. Glassnode, Chainalysis, Arkham, and Galaxy each operate different heuristics, different label sources, and different confidence thresholds. When four shops produce four different estimates of the "unknown" share, the variance between them is a direct measure of how much the number depends on methodology rather than on the chain. A quantity that swings by ten percentage points depending on whose spreadsheet you open is not a fact about Bitcoin. It is a fact about the analyst.
What the ETF era actually did to the supply structure
To understand why this debate has heat right now, you have to understand what the last two years did to Bitcoin's holder composition. The approval of US spot ETFs created, for the first time, a large, transparent, exchange-listed wrapper for institutional exposure. The flows into those vehicles have been substantial and, importantly, disclosed daily. For the first time in Bitcoin's history, a meaningful slice of institutional demand is legible in real time.
That legibility has a side effect. It has trained the market to look at ETF holdings as the proxy for institutional adoption, and to assume that what is visible is what exists. But the ETF is a single, highly visible custody channel. It is not the whole of institutional exposure. Family offices, corporate treasuries, sovereign wealth funds, and private funds hold bitcoin through channels that do not publish daily flow data. Some of them are labeled in the databases. Many are not. The ETF era did not eliminate the unknown bucket; it simply created a bright, well-lit island next to a still-dark ocean, and the market has been staring at the island.
There is a real tension here that the bull case tends to smooth over. The ETF channel concentrates custody. A large share of US spot ETF bitcoin sits with a small number of qualified custodians, and that concentration is visible, regulated, and — from a systemic-risk perspective — exactly the kind of thing that draws scrutiny. Meanwhile, the "half is self-custodied" claim suggests the opposite: dispersion. These two facts can be true at the same time. ETF custody is concentrated and growing. Individual self-custody is dispersed and, according to Thorn, still dominant. The coexistence of concentrated institutional wrappers and dispersed sovereign holdings is the actual state of Bitcoin in 2024-2025, and neither narrative alone captures it.
When I built my ETF inflow correlation model in 2024, I spent 180 days watching the relationship between ETF purchase volume and exchange reserves. The inverse correlation between long-term holder supply and ETF accumulation was one of the cleanest signals I have ever worked with — as ETFs bought, coins left exchanges, and long-term holder supply rose. But clean correlation is not the same as clean causation, and it is not the same as knowing who holds the coins that left. A coin that leaves an exchange for a cold wallet could be going to a sovereign holder, a custodian, or a fund. The chain shows the movement. It does not show the destination's name. My model could tell you that supply was tightening. It could not tell you who was doing the tightening. That gap is precisely the gap at the heart of the Galaxy claim.
Self-custody infrastructure as the corroborating signal
If the "majority self-custodied" interpretation were true, we would expect to see it in the adoption data for self-custody tooling. This is where the claim becomes partially testable, and I want to walk through what I would actually look at, because this is the kind of verification the headline number skips.
The first signal is hardware wallet shipment volume. Hardware wallets are the physical embodiment of sovereign holding. Sustained growth in shipment volume — and, more importantly, in the number of active devices — is direct evidence that individuals are choosing self-custody at scale. A single shipment figure is noise; a multi-year trend is signal. In the noise, the signal remains silent, and this is one place where the signal is worth listening for.
The second signal is the activity of non-custodial wallet addresses. If self-custody is growing, the number of addresses that interact with the chain without routing through a custodial intermediary should grow. The complication is that address counts are inflated by exchange internal accounting and by users who generate fresh addresses constantly. The better metric is the number of distinct clusters exhibiting sovereign behavior — long dormancy, no exchange interaction, occasional consolidation. This is exactly the kind of measurement that the label databases are bad at, which is why the evidence for self-custody dominance tends to rest on survey data and device shipments rather than on chain data. That is a soft foundation for a hard claim.
The third signal is the growth of secondary infrastructure built for sovereign holders: multisig services, collaborative custody, and Bitcoin Layer 2 systems. When I look at the Layer 2 landscape, I see dozens of projects chasing the same small population of users who actually custody their own coins. That is not scaling. That is slicing already-scarce liquidity into fragments. The proliferation of self-custody-adjacent infrastructure is real, but it is a bet on a future in which self-custody is dominant, not proof that it already is.
None of these three signals is conclusive on its own. Together, they point in a direction — toward a real and growing base of sovereign holders — without ever telling you the exact share. That is the honest state of the evidence, and it is why the "50%" headline should be treated as a hypothesis with supporting infrastructure data, not as a measurement.
The methodology opacity problem
Here is the part of the Galaxy claim that I find most difficult to accept at face value, and it is not the conclusion — it is the silence around the method. The reported finding does not disclose the sample, the label source, the confidence threshold, or the date of the underlying data. For a claim of this magnitude, those four omissions are serious.
On-chain attribution is not a peer-reviewed science. It is a proprietary craft practiced by a handful of firms whose label databases are their commercial moats. There is no standard, no public benchmark, and no third party that can audit a rival's clustering accuracy. When Firm A says the unknown bucket is 50% and Firm B says it is 35%, there is no arbiter. Each firm's number is internally consistent and externally unverifiable. This is the analytical equivalent of an unaudited financial statement: it may be accurate, but you have no basis to conclude that it is, other than the reputation of the issuer.
I have a specific professional sensitivity to this. Years ago, while still an undergraduate, I spent eight weeks manually tracing over five hundred token swaps through the constant product formula of an early automated market maker, and I found a rounding error that affected small-cap assets. I compiled the analysis and submitted it. The anomaly was acknowledged, but the fix was deprioritized in favor of stability. What that episode taught me was not that the developers were careless — they were not. It taught me that even a precisely specified, fully reproducible finding, submitted to the people with the power to verify it, can sit unresolved for reasons that have nothing to do with its correctness. If a reproducible finding can stall, an unreproducible one has no chance. The Galaxy claim is unreproducible by construction, and that is a structural weakness, not a criticism of any individual.
The absence of a timestamp compounds the problem. Ownership structure is a moving target. If the underlying data predates the major ETF accumulation of 2024, the "unknown" share it describes has almost certainly already shrunk, because every coin that moved into a labeled ETF custody address left the unknown bucket and entered the known one. A claim about a static residual, published without a date, is a claim about a distribution that may no longer exist. The truth is buried in the timestamp, and here the timestamp is missing.
Contrarian: Correlation Is Not Ownership, and the Researcher Is a Participant
Let me now say the thing that the bullish reading of this claim does not want to hear. The leap from "this address is unlabeled" to "this address is self-custodied by an individual" is not a small inferential step. It is the entire argument, and it is unsupported by the data being cited.
What the data supports is narrow: a residual exists, it is large, and it has not been resolved by the current generation of label databases. What the interpretation adds is a claim about the composition of that residual — that it skews toward sovereign individuals rather than toward lost coins or invisible institutions. The chain cannot distinguish between a hardware wallet in Zurich and a key that fell into a landfill in Wales. Both are dormant. Both are unlabeled. Both sit in the same bucket, and the bucket does not sort them.
This is the correlation-versus-causation trap in its purest form. Dormancy correlates with self-custody. It also correlates with loss and with unlabeled institutional holding. Choosing self-custody as the primary explanation is a modeling choice, not a finding, and it happens to be the modeling choice that produces the most flattering narrative for the asset.
And that flattering narrative has an interested party. Galaxy Digital is not a neutral observer. It is a market participant — an asset manager, an investment bank, a trading desk — with positions and client relationships that are, in aggregate, long the asset class. This does not make its research wrong. It does make the framing worth examining, because there is a coherent story in which "Bitcoin remains majority self-custodied" serves the interests of an institution that holds Bitcoin and wants the market to believe the asset is decentralized, resilient, and not yet captured by Wall Street.
I do not want to overstate this. The conflict is structural, not personal, and it applies to every research desk that also runs a book. But it does mean that when a claim about ownership structure arrives from an interested party, without a disclosed method, without a date, and without a second independent source, the correct response is not to absorb it as fact. The correct response is to hold it as a hypothesis and wait for the cross-verification that has not yet arrived.
There is also a deeper institutional dynamic at work, and it deserves naming. The question of "who owns Bitcoin" is not just an analytical question. It is a question of institutional power. The firms that maintain the label databases — Glassnode, Chainalysis, Galaxy, Arkham — effectively decide what the public believes about Bitcoin's ownership, because no one else can see the map. The label database is the moat, and the research is the marketing that keeps the moat visible. This is not a conspiracy. It is simply how proprietary-data businesses work. But it means that every "ownership structure" finding is simultaneously an analytical claim and a competitive move, and the market should read it as both.
Finally, I want to name the blind spot that the bullish reading creates. If the market accepts "majority self-custodied" too readily, it may lower its guard on the concentration risk that genuinely exists — the custody concentration in the ETF channel, the small number of custodians holding a large share of institutional bitcoin, the operational fragility that comes with any concentrated point of trust. Volatility is the tax on unverified trust, and the flip side of that principle is that false confidence is a hidden liability. A market that believes it is decentralized when it is only unmeasured is a market that has not done its audit.
Takeaway: What to Watch Before You Believe the Number
The value of this claim is not that it settles the ownership question. It is that it exposes how little the industry actually knows, and it gives us a set of verifiable signals to track over the next quarter. Watch the cumulative holdings of US spot ETFs; if they push past roughly 1.5 million BTC, the "unknown" share must fall mechanically, and the claim weakens on its own terms. Watch for a second independent estimate of the unknown share from a rival data house; a wide divergence will tell you the number is a methodology artifact rather than a fact. Watch the annual lost-coin revisions; an upward revision strengthens the deflationary reading and undercuts the self-custody one. And watch hardware wallet shipments and non-custodial wallet activity, because if sovereign holding is truly dominant, the infrastructure that serves it should be growing in the same direction.
The chain records what happened. It does not record who did it, or why. Until the industry can close that gap with a transparent, reproducible method, the most honest thing to say about the ownership of half of Bitcoin is not "it belongs to individuals." It is "we have not yet labeled it." Those two sentences describe the same addresses. They do not describe the same world.
