Somewhere between late summer and early autumn, something shifted on the Bitcoin chain that the data vendors decided to call a turning point. Miner wallets stopped bleeding. The 100-to-1,000 BTC cohort, which had shed roughly a fifth of its balance over the preceding quarter, flattened out. Satoshi-era addresses โ the oldest hands in the network โ cut their distribution by 70%. Daily miner revenue climbed from $27 million to $48 million. Hashrate recovered from a drawdown of 18% to a shallower 13%, printing above 960 EH/s. The Puell-style sustainability gauge flipped from "extremely underpaid" to "reasonably compensated."
On its face, that is a clean bullish set. Supply pressure removed, capital expenditure returning, the strongest cohort of holders refusing to sell. The narrative writes itself, and the vendor that supplied the numbers handed it to the market in a single-source weekly report, packaged for a wire service and pushed to every terminal that pays for the feed.
But the audit trail does not support the story the headline sells. Every one of those improvements traces back to a single variable that has nothing to do with miner conviction: the price of Bitcoin rose roughly 45% over the same window. Revenue up 78%? That is price elasticity, not demand. Miners "no longer obligated to sell"? That is a description of a miner whose marginal cost now sits below spot โ not a miner who has decided to believe.
The miners did not stop selling because they became believers. They stopped selling because the price rose above their cost of production. Those two statements look identical on a chart and are opposites in meaning. That distinction is the entire article.
Context: Three Numbers, and the One the Report Left Out
To read this correctly you have to hold three figures in your head at once, and the report only gives you two of them cleanly.
The first is the block subsidy. Since the April 2024 halving, every block mints 3.125 BTC to whoever finds it. That is the fixed, programmed, non-negotiable half of miner revenue, and it decays by half roughly every four years on a schedule no committee can vote to change. The second is the transaction fee pool. The third โ the one the report mentions in a single passing clause โ is the ratio between them.
Bitcoin's security budget was designed around a handoff. As the subsidy decays toward zero across successive halvings, fees are supposed to grow to replace it. That handoff has not happened. In the current data, the seven-day average fee revenue sits near $275,000 per day, up 41% from $195,000, but still a fraction of the peaks recorded during the inscription and ordinals cycles of prior years. Against a total daily revenue of roughly $48 million, fees account for approximately 0.57% of miner income. The other 99.4% is subsidy plus price.
That single ratio reframes everything downstream. A miner whose revenue is 0.57% demand-driven and 99.4% price-driven is not a business responding to network usage. It is a leveraged long position on the spot price with an electricity bill attached. When such a miner "stops selling," the honest translation is not "conviction" โ it is "the position is temporarily in the money."
The second piece of context is where the numbers come from. This is a miner-flow analytics dataset โ wallet-clustering of addresses believed to belong to mining entities, layered by holding size, with a proprietary sustainability oscillator bolted on top. It is a mature, commercialized method, and that is precisely the problem. Glassnode and CryptoQuant and IntoTheBlock all run versions of it. The clustering algorithms are trade secrets. Nobody publishes the address lists. Nobody discloses whether exchange mining-pool wallets are included, whether internal transfers between a miner's own wallets are netted out, or whether custodial addresses are counted as "miner wallets" at all. The output is presented as objective on-chain truth. It is a model, and the model's assumptions are invisible.
I spent six weeks in a Solidity bootcamp during the 2020 DeFi Summer โ not to become a developer, but because I wanted to audit contracts rather than trust the dashboards describing them. That instinct is the same one I bring here. A dashboard is a claim. The claim is only as good as the assumptions you cannot see.
And the assumptions here matter, because the report itself flags that in August, miners moved coins into NYDIG, a regulated custody provider. That single fact can contaminate the entire "miner balance" series, and I will return to it with a scalpel.
Core: The Arithmetic of a Reflex
The 78% That Isn't Demand
The headline number is that miner daily revenue rose from $27 million to $48 million, a 78% increase. Read it in isolation and it looks like a business inflecting. Read it against the fee line and it collapses into something simpler.
If fees moved from $195,000 to $275,000, that is a $80,000-per-day improvement. The total revenue improvement was $21 million per day. Demand-driven revenue contributed less than four-tenths of one percent of the gain. The other 99.6% came from the price of the asset miners are paid in. This is not a miner whose product got more valuable to the network. It is a miner whose paycheck got revalued because the currency it is denominated in went up.
I have made this mistake before in the other direction. In 2021, as an undergraduate, I spent four weeks modeling Shiba Inu's liquidity pools on Uniswap against Ethereum gas fees, trying to prove that meme-coin "demand" was a mirage. The report I published โ "The Illusion of Decentralization in Hyper-Speculative Assets" โ got mocked by the finance department and went viral in crypto circles, and it taught me the same lesson this miner data is teaching now: when a headline metric moves, your first job is to decompose it until only one variable is left standing. When I decomposed SHIB flows, the "demand" was a handful of wallets recycling liquidity. When you decompose miner revenue, the "growth" is the price of Bitcoin.
That does not make the revenue fake. It makes it reflexive. And reflexivity cuts both ways.
The Puell Reflex: Why "Reasonable Compensation" Is Mechanical
The report's most quotable line is that miners moved from "extremely underpaid" to "reasonably compensated" on a Puell-style sustainability metric. This is where the methodology deserves the most suspicion, because the indicator is doing something subtle and the report is presenting it as something profound.
The Puell Multiple, in its standard construction, divides miner daily dollar issuance by its own 365-day moving average. When the ratio is low, miners are earning less than their trailing norm โ "underpaid" โ and historically that condition precedes capitulation, because miners sell to cover costs. When the ratio recovers, the pressure eases.
Now trace the mechanical path. A 45% spot-price rally does not just nudge this indicator. It drags it upward almost arithmetically, because the numerator is denominated in dollars and the denominator is a backward-looking average that still carries the memory of the lower prices. You can flip a Puell-style gauge from "underpaid" to "reasonable" without a single miner improving their operations, simply by moving the price. The indicator does not measure miner health. It measures miner revenue against miner revenue's own history, and revenue is dominated by a price that moves faster than any cost structure.
So when the report says miners are now "reasonably compensated," the accurate reading is: the spot price rose fast enough that a mechanical oscillator crossed a threshold. That is a description of price action, wearing the costume of fundamental improvement.
This is the kind of slippage that gets people hurt. In 2022, when Luna collapsed and triggered a chain-wide liquidity crisis, I worked with three independent researchers to map stablecoin issuer reserves against traditional banking stress indicators โ a 50-page paper correlating USDT redemption rates with offshore non-deliverable forward markets. The whole exercise existed because the prevailing narrative said DeFi was dead, and the on-chain data said the stress was a plumbing problem, not an extinction event. The lesson I carried out of that period: an indicator is only as honest as the thing it refuses to measure. A Puell-style gauge that never separates subsidy from fees is refusing to measure the thing that actually determines whether a miner survives the next halving.
The Custody Mirage: How NYDIG Pollutes the Ledger
Here is the part of the audit trail that should stop a careful reader cold. In August, the report notes, miners transferred coins to NYDIG, a regulated custody provider. And yet the September data shows miner balances "stabilizing" and outflows "stopping."
Those two facts sit together uneasily.
When a miner moves BTC from a self-custodied wallet to a custodial account, the on-chain record shows coins leaving the miner's known address cluster. A naive flow model registers that as a sale. It is not a sale. It is a change of venue. The coins still belong to the miner; they have simply moved from an address the model calls "miner" to an address the model may or may not call "miner," depending on clustering rules nobody publishes.
This means the entire "miners stopped selling" claim may be, in part, an artifact of coins migrating from self-custody into custodial infrastructure โ a movement that looks like distribution on-chain while being nothing of the sort. The report acknowledges the August migration but never clarifies whether the September "stop" in outflows accounts for it. That omission is not a footnote. It is the load-bearing wall of the bullish thesis, and it has a crack running through it.
I have a professional reason to care about custody mechanics. In 2024, after the spot Bitcoin ETF approvals, I traveled to Dubai and Singapore interviewing compliance officers at fintech startups, mapping the gaps in AML regimes that crypto firms could exploit โ work that became a series on regulatory arbitrage as a market-making force. What that reporting taught me is that custody is where the visible ledger and the real ownership diverge most sharply. A regulated custodian like NYDIG sits at the seam between the two. It lets miners access institutional lending and hedging โ and in doing so, it moves their selling off-chain, into collateral and derivatives, where the on-chain flow model cannot see it.
So the ledger says miners stopped selling. The plumbing says miners may have simply relocated where they sell from, and in some cases converted outright selling into borrowing against the same coins. Both statements can be true. The dashboard only shows one.
The Hashrate Signal: The One Number That Earns Its Keep
Buried under the softer claims is a genuinely harder signal, and I want to give it its due because a fair audit credits what holds up.
Hashrate recovered from a drawdown of 18% to a shallower 13%, crossing above 960 EH/s. The report frames this as "capacity returning, not capitulation," and unlike the revenue and Puell claims, this one survives scrutiny.
The reason is economic. Hashrate is a capital-expenditure commitment. Bringing rigs online means paying for electricity, cooling, and hardware โ a bet that spot price will justify the marginal cost of production over months, not days. A miner does not spin up machines on a 45% reflex bounce unless they believe the move has legs, because the machines cost money the moment they run.
This is the closest thing in the dataset to a leading indicator, precisely because it is the most expensive signal to fake. A wallet balance can stabilize for custodial reasons. A Puell gauge can flip on price alone. But hashrate rising requires someone to have signed an electricity contract and turned on hardware. That is revealed preference with a bill attached.
It is also the signal that connects to my current research. In 2026 I launched a project modeling decentralized compute markets as a new liquidity layer, building a predictive model for AI-token valuations against compute supply elasticity. That work sharpened a question that applies directly here: when you have two competing buyers for the same megawatt โ an AI datacenter and a Bitcoin miner โ the miner's cost of production is no longer set in isolation. Energy is becoming a contested input, and a hashrate recovery is a statement about the miner's willingness to win that contest. That is a more interesting fact than any wallet balance.
The Satoshi-Era Cohort: Strong Signal, Tiny Sample
The report's most emotionally resonant claim is that Satoshi-era miners โ the earliest participants, sitting on holdings of roughly 590,000 BTC โ reduced their selling by 70%. The framing is that the network's original believers have stopped distributing, which reads as a long-term vote of confidence.
I am sympathetic to the signal and suspicious of the sample.
Sympathetic, because these addresses have the lowest cost basis in existence โ effectively zero โ which means their selling behavior is the least price-sensitive and the most conviction-driven data in the entire set. When the cohort with no cost pressure decides to hold, that is information.
Suspicious, because the September figure involved only around 600 BTC. A 70% reduction in distribution, measured on a base that small, is statistical noise wearing a suit. The report itself notes that Patoshi-pattern addresses โ the non-standard mining footprint attributed to Satoshi's own early mining โ were excluded to avoid contamination, which is methodologically correct and also a reminder of how delicate this cohort is to define. You are drawing conclusions about the market's most legendary holders from a handful of coins and a clustering heuristic.
So I file this claim under "directionally interesting, statistically fragile." It belongs in the watchlist, not in the thesis.
The Cost Line: Where the Reflex Reverses
The most useful thing you can extract from this dataset is not the headline โ it is the implied breakeven.
The report tells us miners were "extremely underpaid" when spot sat around $58,000 and "reasonably compensated" when it reached roughly $83,000. Back out the transition and you land on a rough breakeven band of $65,000 to $75,000. That is where the average miner's marginal cost of production meets the market.
This number is the article's real payload, because it converts a static snapshot into a dynamic trigger. If spot falls back into the $65,000-$75,000 zone, the "stop selling" condition mechanically reverses into "resume selling" โ not because sentiment changed, but because the position goes underwater again and the electricity bill does not care about narratives. The bullish signal everyone is celebrating is, in structural terms, a conditional state that holds only while price stays above a line you can estimate.
And note the admission the report makes almost reluctantly: miners have not yet begun to accumulate. They stopped selling, and that is all. "Not selling" and "buying" are not adjacent positions on a spectrum โ they are separated by a balance-sheet decision that requires surplus cash. A miner who has just emerged from a period of being "extremely underpaid" is a miner who spent down reserves, not one sitting on dry powder. The absence of accumulation is the quiet tell that the cohort is still cautious, still watching, still one bad week from capitulation.
The Single-Source Problem
One more methodological flag before the contrarian turn. Everything above rests on a single vendor's weekly report, shared exclusively with a single media outlet. There is no cross-validation against Glassnode, IntoTheBlock, or any independent miner-flow series. The proprietary sustainability oscillator is undisclosed. The clustering algorithm is undisclosed.
I am not alleging fabrication. I am alleging something subtler and more common: a data vendor has an incentive to publish reports that generate attention, and attention flows toward clean, quotable, bullish narratives. "Miners stop selling" is a headline. "Miner revenue is 99.4% price-driven and their balances may be stabilizing for custodial reasons" is not. The selection pressure runs in one direction, and it runs toward the story you were just sold.
Contrarian: The Selling Data Is Going Blind
Here is the counter-intuitive angle, and it is the one I would bet on.
Everyone is debating whether miners have stopped selling. Almost nobody is asking whether "miner selling" is still a measurable thing at all.
The August migration to NYDIG is not an anomaly. It is the leading edge of a structural shift I would call miner financialization โ the migration of mining from self-custodied operators into the institutional lending, custody, and derivatives complex. Once a miner's coins sit with a custodian, their economic exposure can be reshaped without a single on-chain transaction. They can borrow against the coins instead of selling them. They can hedge with options, locking in a price without moving a satoshi. They can post the coins as collateral and let the custodian's own risk desk manage the delta.
Every one of those activities produces real selling pressure that the on-chain flow model cannot see. The model sees coins that did not move. It reports that as holding. The selling happened in the derivative, in the loan book, in the collateral ratio โ off the ledger, invisible to the dashboard.
This is the audit trail of a broken liquidity trap, and it is the same trap I documented in the meme-coin cycle from a different direction. In 2021, the liquidity looked decentralized because the addresses were many; the control was concentrated because the mechanics were opaque. Here, the selling looks absent because the wallets are quiet; the pressure is present because it moved somewhere the wallets do not record. The signal is not weakening because miners changed their minds. It is weakening because the market's plumbing evolved past the instruments we use to watch it.
The corollary is uncomfortable for everyone using this data as a trading input. If miner-flow analytics are increasingly blind to real supply pressure, then the entire "miners stopped selling, therefore supply shock, therefore bullish" chain of reasoning rests on an instrument that is quietly losing resolution. You are reading a thermometer that has been moved into the shade and concluding the fever broke.
And there is a reflexive irony worth naming. The more the market trusts the "miners stopped selling" signal, the more miners have an incentive to route their selling through custody and derivatives โ to preserve the appearance the market is paying for. The signal destroys itself by working.
Takeaway: Positioning for a Conditional Calm
So where does this leave the cycle?
What the data actually shows is a conditional calm, not a structural pivot. Miners stopped selling because price crossed above their cost line, not because their conviction deepened. Hashrate is recovering, which is the one hard, expensive, credible signal in the set. Fees remain under 1% of revenue, which means the network's long-term security budget is still hostage to price. And the balances that "stabilized" may have stabilized in a custodian's vault rather than in a miner's conviction.
The forward-looking question is not "did miners stop selling." It is "what happens to this calm when spot revisits $65,000 to $75,000?" If the reflex reverses there, the bullish narrative built on this dataset unwinds as fast as it was assembled. If miners start accumulating above that line โ the confirmation the report itself admits has not arrived โ then something real has changed. Until then, watch the breakeven, not the balance.
The most honest reading of the miner pause is the one the headline cannot sell: the supply pressure did not disappear. It went somewhere quieter, and the instruments we trust are still catching up.