The Hashrate That Stayed Flat: Miners, the Missed Rally, and the Price of a Relabeled Business

The Hashrate That Stayed Flat: Miners, the Missed Rally, and the Price of a Relabeled Business

Hook: Four Series, One Broken

Three series moved in the fourteen sessions I logged from the desk. Spot BTC moved. Listed exchange equity moved harder. Aggregate stablecoin float moved harder still. A fourth series did not. Total network hashrate held inside a band I could bracket with a single order, and the publicly listed miner complex printed flat to negative against BTC beta.

That is the anomaly worth auditing. Not the price. Price does what price does, and it does not need my commentary. The anomaly is that the production layer of an asset declined to participate in the appreciation of the asset it produces. Production layers are supposed to be the highest-beta expression of the underlying. When they stop being that, either the production business changed, or the market changed its mind about who owns the upside.

My dashboard, indexed to 100 at the start of the window:

• Spot BTC: up double digits.
• Listed exchange equity: up more.
• Aggregate stablecoin float: up more again.
• Public miner equity versus BTC beta: negative spread.
• Hashprice: pinned inside the forty to fifty dollars per petahash per day band.
• Network hashrate: range-bound inside a five percent envelope.

Six lines. One contradicts the other five. I have seen this exact contradiction in a different venue, and the resolution was never sentimental. In January 2024, hours after the spot ETF approval, I watched a fifteen dollar gap open between the fund net asset value and the underlying BTC on Coinbase Pro. Everyone with a terminal could see the number. Almost nobody could execute against it, because execution required rails that settled in minutes rather than days. The gap did not close because people agreed it should. It closed because somebody built the pipe.

That is how I read miner underperformance now. Not as a verdict on mining. As a plumbing problem with a label stapled to it.

Context: What A Miner Actually Owns

Strip the branding and a bitcoin miner owns four things: a power contract, land with an interconnection agreement, a fleet of application-specific integrated circuits, and a balance sheet denominated in a volatile asset. Three of the four have nothing to do with bitcoin.

The subsidy is the revenue line. After the April 2024 halving, the block subsidy fell to 3.125 BTC. Transaction fees, which briefly spiked during congestion episodes, remain a volatile and historically minor share of miner revenue - loud enough in bursts to matter for a week, too thin and too unpredictable to underwrite a ten-year capital plan. The revenue side of mining is therefore a decaying annuity with a lottery ticket stapled to it.

The cost side is where the business actually lives. A modern fleet runs somewhere between seventeen and twenty-five joules per terahash depending on vintage. Previous-generation S19-class machines sit near the high end. Current-generation units sit near the low end. Multiply joules per terahash by the industrial power rate, multiply by 86,400 seconds, divide by the joule-to-kilowatt-hour conversion factor, and you get a dollar cost per terahash per day. Subtract that from revenue per terahash per day and you have gross margin. Everything else - the treasury strategy, the shareholder letters, the keynote - is decoration on that one subtraction.

I learned this lesson the hard way in 2020, while finishing a master's degree and spending nights reading contract code. I found an integer overflow in an early governance module of a lending protocol and shipped a standardized bug report to the repository rather than posting it to a forum. Five thousand dollars and a formal acknowledgement later, the takeaway was not that open source is generous. The takeaway was that the only durable edge is verifying the logic before you trust the label. My entire reading of the mining sector is an application of that habit.

And the label has changed. The sector no longer describes itself as mining. It describes itself as energy infrastructure, high-performance computing, AI hosting, or, in the most aggressive framing, compute. That relabeling is not cosmetic. It is an attempt to change the comp set, and the comp set determines the multiple, and the multiple determines the cost of capital, and the cost of capital determines whether the business survives the next subsidy halving.

The Core: A Subtraction, Not A Story

The Unit Economics of Erosion

Here is the arithmetic the sector avoids presenting on a single page.

Take a fleet running twenty-two joules per terahash at a six cent industrial power rate. Electricity works out to roughly three point two cents per terahash per day once you account for the 86,400 seconds in a day and the unit conversion. At a hashprice near forty-five dollars per petahash per day - which is four point five cents of revenue per terahash per day - the power bill consumes roughly seventy percent of gross revenue. Gross margin lands near thirty percent.

Now rewind eighteen months. Hashprice was closer to one hundred dollars per petahash per day. Power cost was identical. Gross margin was closer to sixty-eight percent. The machines did not change. The power contract did not change. The only variable that moved was the ratio between network hashrate and the dollar value of the subsidy, and that ratio is set by strangers.

This is the structural fact that a bull market in the underlying asset conceals. Hashrate is a treadmill. Every dollar of hashprice above the marginal miner breakeven pulls new capital into machines. Those machines come online six to twelve months later. Hashprice falls back toward the marginal cost of production. The industry has spent a decade proving that it can convert an appreciating asset into a depreciating return on capital, and it keeps doing it because the option value of holding BTC on the balance sheet feels real.

The consequence is that miner equity is not a levered BTC proxy. It is a short position on hashprice dressed as a long position on BTC. Everyone who bought miner equity in 2021 expecting leverage got the short leg instead. The charts have said so for four years. The narrative still has not updated.

The Code That Tells You The Truth

I do not trust a sector thesis that cannot be expressed as a function. Below is the version I run against any miner disclosed fleet efficiency and power cost. It is the same class of tool I built for Solana RPC monitoring in late 2023, when transaction failures were eating fifteen percent of my bot fills and the fix was not a better view but a standardized measurement loop. Efficiency is the only honest validator.

def mine_margin(hashprice_usd_per_ph_day, j_per_th, power_usd_per_kwh):
    # kWh consumed per terahash per day
    kwh_per_th_day = (j_per_th * 86400.0) / 3.6e6
    power_cost_per_th_day = kwh_per_th_day * power_usd_per_kwh
    revenue_per_th_day = hashprice_usd_per_ph_day / 1000.0
    gross = revenue_per_th_day - power_cost_per_th_day
    return {
        'revenue_per_th_day': round(revenue_per_th_day, 5),
        'power_cost_per_th_day': round(power_cost_per_th_day, 5),
        'gross_margin_pct': round(100.0 * gross / revenue_per_th_day, 2),
        'cash_breakeven_power_price': round(revenue_per_th_day / kwh_per_th_day, 4),
        'cash_breakeven_hashprice': round(power_cost_per_th_day * 1000.0, 2),
    }

Run it at forty-five dollars per petahash per day, twenty-two joules per terahash, six cents per kilowatt hour, and the function returns a gross margin near thirty percent and a cash breakeven hashprice near thirty-two dollars. That last number is the one that matters. It means the marginal machine in the marginal jurisdiction is roughly two weeks of adverse compression away from shutting off. It also means every operator with a power contract above seven cents is already running at zero or negative gross margin, and is doing so only because shutting down forfeits the site.

This is not a forecast. It is a static reading of a dynamic book. But it explains something the price charts cannot: why miner treasuries were being sold into a rally.

The Conversion Arbitrage, Quantified

Now run the same hundred megawatts through the other business.

A hundred megawatt site populated with twenty-two joule machines produces roughly four and a half exahashes per second. At forty-five dollars per petahash per day, that is about two hundred and five thousand dollars of gross revenue per day. The power bill at six cents is roughly one hundred and forty-four thousand dollars per day. Gross margin: about sixty-one thousand dollars per day, call it twenty-two million dollars per year, before corporate overhead, interest, and depreciation on a fleet that is obsolete on a three-year clock.

The capital that bought you that outcome is brutal. Four and a half exahashes of current-generation hardware at a street price near twelve dollars per terahash is roughly fifty-four million dollars. The site itself - land, transformer, substation, buildings, cooling - is another hundred million or more depending on jurisdiction and interconnection queue position. Call it one hundred and sixty million dollars of capital for twenty-two million dollars of annual gross margin. Fourteen percent, before overhead. That is not a business. That is a hobby with a legal department.

Now convert the same hundred megawatts. Strip the ASICs, install high-density racks and liquid cooling, and sign a ten to twelve year colocation contract with a hyperscaler or an AI lab. Contracted pricing in the deals I have reviewed has been disclosed or inferred in the range of one hundred to one hundred and thirty dollars per kilowatt per month for powered shell, and higher for full-service hosting. Take the low end of that range. One hundred thousand kilowatts times one hundred and twenty dollars times twelve months is one hundred and forty-four million dollars of annual revenue. At a seventy percent margin, that is roughly one hundred million dollars of EBITDA.

The conversion capital is the variable. Retrofits of existing substations and buildings have been running in the range of one and a half to three million dollars per megawatt depending on how much of the original shell survives. Call it two hundred million dollars for the hundred megawatt site. That is one hundred million dollars of EBITDA on two hundred million of incremental capital, against twenty-two million of gross margin on the alternative.

The multiple on capital is not twice as good. It is four to five times as good, and the revenue is contracted rather than spot. When a mining executive tells you the company is pivoting to AI, that arithmetic is what they mean. They are not chasing a narrative. They are arbitraging a spot commodity against a long-duration lease, and the lease wins.

The problem is execution, and the problem is that the market can see the arithmetic too.

Why The Equity Did Not Follow BTC

Three mechanical reasons explain why miner stocks tracked hashrate instead of spot.

First, the financing structure. Converted sites require capital, and these companies raise it the way they have always raised it: at-the-market share issuance, convertible notes, and equipment financing. Share count growth converts operating leverage into dilution. A shareholder who is right about the pivot can still be wrong about the per-share outcome, and the market prices per-share outcomes.

Second, the reflexivity of the treasury. Miners pay for capex partly in BTC. That makes them structurally forced sellers into strength. Every rally of sufficient magnitude produces a wave of miner treasury liquidation, which caps the upside of the very move that created the incentive. This is the only asset class I know that systematically sells its own inventory into its own highest bid, and it sells because the capex clock does not care about your conviction.

Third, the wrapper. Between 2024 and now, institutions gained a clean, audited, low-fee, physically backed way to own BTC without owning a single share of a company with a power contract and a transformer. The ETF is the pure expression. Miner equity is the impure expression. When the clean instrument exists, the impure instrument must trade at a discount to its theoretical beta, and that discount is not a mispricing. It is the market correctly pricing the wrapper.

I documented the plumbing on the ETF gap in January 2024 because the lesson generalizes: institutional entry creates predictable, rule-based opportunities for anyone who can execute faster than a legacy desk. The flip side of that sentence is the part nobody writes down. Institutional entry also destroys the premium that retail used to earn for holding inconvenient wrappers. Miner equity was an inconvenience premium. The inconvenience is gone.

The Other Side of the Pipe

Meanwhile, exchange equity and stablecoin float both expanded. The mechanism is not mysterious. A rising BTC price raises on-chain collateral values, which raises borrowing demand, which raises stablecoin velocity, which raises the float, which raises the fee revenue of every venue that touches the rail.

The stablecoin leg deserves its own note. I have argued for two years that the launch of a regulated, bank-adjacent stablecoin by a major payments company was not primarily a product decision but a regulatory hedge - the rational move for an incumbent that would rather be inside the rulebook than in front of it. That logic is now visible in the float data. Regulated issuance is capturing share precisely because compliance has turned into distribution. Compliance is currently the cheapest form of efficiency at the settlement layer.

So the capital rotation in this window had a direction: out of the production layer, into the settlement layer. Out of the thing that burns electricity to create the asset, and into the thing that moves the asset and charges for movement. Mining is upstream, capital-intensive, and cyclical. Settlement is midstream, fee-based, and reflexive. In a sideways-to-up tape, midstream wins. Every time.

Contrarian: The Pivot Is A Relabeling, And The Comp Set Is Wrong

Here is where I part company with the consensus read.

The consensus read is that miners are pivoting to AI because they are agile and forward-looking. The audit read is that miners are relabeling a power-and-land arbitrage because the multiple on bitcoin mining has permanently compressed, and the only way to refinance a hundred-million-dollar site is to change the story attached to it.

Strip the AI language and what remains is a business that owns interconnection queue positions in places where power is cheap and latency to a fiber route is acceptable. That business is not a technology company. It is a real estate and power arbitrage operation with a hardware tenant. The correct comparables are data center REITs and independent power producers, not software companies, and definitely not bitcoin. The moment you accept the correct comp set, the entire valuation conversation changes. So the comp set is the last thing the sector will voluntarily accept, and it is the first thing any honest auditor should impose.

The pivot is not a strategy pivot. It is a multiple pivot, and the multiple is the product.

Second contrarian point: the framing that miners missed the rally flatters the market and insults the arithmetic. Miners did not miss anything. They were structurally excluded from the clean expression of the trade the moment the ETF listed, and they were structurally obligated to sell into strength because their capex clock runs on a three-year hardware depreciation cycle. The underperformance was not a mistake. It was pre-ordained by the wrapper and the financing stack. Calling it a mistake implies the outcome was avoidable, which implies someone should have done something differently. The people who did something differently bought the spot ETF and never opened a miner chart.

Third: the AI narrative around miners is presently un-earned. I can find conversion announcements, megawatt commitments, and letters of intent. I cannot yet find a cohort where contracted AI or HPC revenue is more than a footnote in the income statement. I have seen this movie before, and I lost money to it, and so did everyone who bought a whitepaper instead of an audited contract. In 2020 I read the logic of a governance module line by line before I accepted a single claim about it, and I got paid for the reading, not the believing. Audit the logic before you trust the label. The label currently says AI. The logic currently says power contract with no signed counterparty at scale.

Fourth, and this is the point almost nobody wants to hear: the AI pivot, if it succeeds, is neutral-to-bullish for the remaining miners and quietly bearish for the narrative. Every megawatt that leaves mining reduces hashrate growth, which is the single variable that has destroyed miner margins for a decade. Fewer machines chasing the same subsidy means higher hashprice for the machines that remain. The operator who converts to hosting is not a competitor to the operator who stays. He is a gift, and most of the sector has not priced the gift.

Red candles do not negotiate with hope, and neither does a thirty percent gross margin.

Risk Register And Kill Switches

I keep a written kill-switch list for every thesis, a habit that started the week of the Terra collapse in May 2022, when I executed a pre-defined liquidation of forty percent of my stablecoin holdings into bitcoin inside forty-eight hours and preserved six figures while people I knew were on video calls at four in the morning deciding whether to be brave. Discipline is not a personality trait. It is a pre-committed rule set that executes whether or not you feel like it.

For this thesis, the rules are:

• Exit the mining-equity thesis if hashprice does not reclaim sixty dollars per petahash per day within two quarters. Below that level, the marginal fleet cannot fund replacement capex.
• Exit if share count growth at the companies I track exceeds fifteen percent annualized for two consecutive quarters. Operating leverage cannot outrun dilution.
• Exit if miner BTC treasury balances fall for three consecutive months while hashrate rises. That is the forced-seller signature, and it means the sector is financing capex out of the asset.
• Exit if no operator discloses contracted AI or HPC revenue above thirty percent of total revenue within four quarters. Without that disclosure, the pivot is a press release.
• Size the position so that a fifty percent drawdown costs less than one percent of portfolio equity. Leverage magnifies character, not just capital.

Takeaway: The Lines I Am Watching

The question is no longer whether miners can participate in a crypto rally. They demonstrated in this window that they cannot, and the reason is structural rather than situational. The question is whether the AI compute business converts a spot-margin treadmill into a contracted annuity before the next halving compresses the revenue line again.

Three lines tell me the answer, and I check them weekly.

Hashprice. If it sits above sixty, the mining leg funds itself. If it sits below thirty-five, a meaningful share of the global fleet is cash-negative, consolidation is coming, and the survivors collect the gift of reduced hashrate growth.

Contracted revenue share. The first operator to put a number above thirty percent in an audited statement ends the debate, and the multiple re-rates within days. Until then, every announcement is a letter of intent wearing a quarterly report.

Miner treasury flows. Watch BTC leaving miner wallets and arriving at exchanges. Volume is the forced seller. The absence of volume is the signal that the capex cycle has actually funded itself from operations rather than from inventory.

Somewhere in the middle of the chop, one of these three lines breaks first. I do not need to predict which one. I need to be positioned on the other side of it, with the measurement loop running and the kill switch written before the trade. Fear is a bad indicator. Data is a leader. The hashrate stayed flat because the money left the table. The question worth carrying into next week is simpler than it looks: if the pivot works, does anyone still want to be a miner, and if nobody wants to be a miner, who is left to sell the next rally?