Have you noticed how often people talk about Bitcoin price and almost never talk about the machines that keep the network honest? That gap matters right now. Computing power on the network has stayed below its late 2025 record for an unusually long stretch, and a growing number of operators are asking a blunt question before they buy another rack of chips: should this megawatt go to Bitcoin, or should it go to artificial intelligence? I’ve found that this is the kind of market turn people only recognize after the fact.
Why This Hashrate Slump Feels Different
Industry executives have started calling the current stretch Bitcoin’s first hashrate bear market. The phrase is informal. It is not an official network label. Still, it captures something real. Estimated computing power climbed toward 1.3 zettahashes per second late last year, then slipped and refused to reclaim that high. Presentation materials circulating after a late August industry talk put the drawdown in a rough band of 22% to 24% from the peak.
That is not a one-week blip. Daily readings jump around because Bitcoin never publishes a neat machine census. Analysts infer hashrate from block times and difficulty. One day looks strong. The next day looks weak. Longer averages tell a clearer story, and those averages have been soft for months.
In my experience, the market loves a dramatic crash more than a slow grind. A sudden outage is easy to narrate. A quiet reallocation of capital is harder to see and, frankly, more important. This cycle looks like the second kind.
The 2021 Shock Was Relocation, Not Reassessment
The last famous hashrate collapse came after China forced miners out in 2021. Facilities went dark fast. Machines were boxed, shipped, and plugged in elsewhere. North America, Central Asia, and a handful of other regions absorbed the hardware. Hashrate recovered because the same business model simply changed address.
This time the story is slower and more structural. Operators are not only moving boxes. They are questioning whether new power contracts, new buildings, and new cooling plants should be dedicated to Bitcoin at all. That is a different animal. Relocation keeps the fleet intact. Reassessment shrinks the fleet at the margin.
This has been the longest period that we’ve seen from an all-time high until recovery.
That line from a late August conference talk is the heart of the argument. Recovery used to be a logistics problem. Now it is a capital allocation problem. Logistics can be solved with ships and electricians. Capital allocation depends on what else those megawatts can earn.
What Hashrate Actually Measures
Hashrate is an estimate of the computing power miners throw at the network to win block rewards and secure transactions. More hashrate usually means more machines, newer machines, or both. Less hashrate usually means older units are unplugged, sites are delayed, or operators are sending power somewhere else.
On September 2, one widely watched estimate sat near 829 exahashes per second after several late August days that briefly pushed above one zettahash. Those daily prints are noisy. Treat them like weather, not climate. The climate reading is the multi-month failure to retake the prior record.
Earlier summer analysis already pointed to mining difficulty sitting nearly 20% below its November peak, with hashrate in a downward trend for close to 287 days. Difficulty is the network’s way of keeping block times near ten minutes. When machines leave, difficulty eventually follows them down. When machines flood in, difficulty climbs. The two series do not move in lockstep day by day, but they rhyme over weeks.
AI Is Competing For The Same Scarce Inputs
Bitcoin mines and AI campuses want many of the same things. Large power interconnects. Land with a path to transmission. Cooling. Buildings. Construction crews. Patient capital. Fiber helps too, especially if the site might host more than hashing.
Conversion is not a swap of one box for another. AI halls need different chips, different networking, different uptime standards, and often different customers. You cannot unplug an ASIC row on Tuesday and sell training capacity on Wednesday. Still, a site that already owns power and a building has a head start that a greenfield campus does not.
That option changes miner behavior. In earlier cycles, spare megawatts almost automatically became more Bitcoin machines. Today those megawatts can be pitched to a cloud buyer, a model trainer, or a hosting client who will pay for steadier compute. Perhaps the most interesting aspect is how quickly public-market language shifted from “more hash” to “more high performance computing.”
One executive put it harshly: look across listed miners and almost nobody is staying the course at scale. That is overstated if you take it word for word. Several large public names still run substantial Bitcoin fleets. The trend underneath the exaggeration is harder to dismiss. Capital budgets are splitting. New interconnects are being marketed as dual-use. Earnings calls spend more time on contracted AI megawatts than on the next ASIC generation.
Not Every Miner Is Walking Away
It helps to separate rhetoric from the actual fleet. Companies such as MARA, CleanSpark, Riot, and Bitdeer still operate large Bitcoin operations even while they test adjacent compute. The pivot is further along at names that have already booked meaningful high performance computing revenue or signed long-dated data center contracts.
Some operators have already seen AI or hosting revenue overtake mining in a single quarter. Others have lined up revolving credit specifically to fund long-term AI halls. That does not mean Bitcoin mining is finished. It means the industry’s growth capex is no longer a one-way street into ASICs.
- Some firms keep hashing at scale and add AI only at the margin.
- Some split campuses so one hall hashes and another hosts GPUs.
- Some treat Bitcoin as a flexible load while they build contracted compute.
- A smaller group is trying to leave mining as the core story altogether.
I’ve found that investors get sloppy here. They hear “miners are going into AI” and assume the entire hash supply vanishes next month. Reality is messier. Machines already paid for will keep running if power is cheap enough. The squeeze shows up in new deployments, delayed expansions, and older generations that no longer clear the cost curve.
Why Low Cost Operators Can Still Win
Mining is not automatically a bad business. It is a cost-curve business. An operator with efficient machines, cheap power, and a sane balance sheet can stay profitable in conditions that force a high-cost rival to shut down. Debt maturity can matter as much as watts per terahash. A site that “works” on paper can still fail if the loan officer arrives first.
Hash price — expected miner revenue per unit of computing power — remains low versus older boom periods. That punishes last-generation boxes and expensive electricity. It also explains why some boards would rather sell power or lease halls than keep stretching for a thinner Bitcoin margin.
Here is the twist that keeps veterans in the game. When hashrate leaves, difficulty eventually drops. Remaining miners then control a larger slice of the same block reward without buying extra machines. That is the quiet consolation prize of a hashrate bear market.
The beautiful thing about Bitcoin mining being in a bear market of hashrate is that, for those that stay around, they naturally get a higher share of the market.
Share is not the same thing as profit. Revenue still depends on Bitcoin’s price, fees, power cost, machine efficiency, and how many competitors remain. A bigger slice of a shrinking pie can still taste bad. A bigger slice of a rising pie can look brilliant. The network does not promise either outcome. It only rebalances the contest every 2,016 blocks, or roughly every two weeks.
Price Has To Outrun Hashrate Growth
One useful rule of thumb keeps coming back in operator conversations. Mining has the best chance of beating spot Bitcoin when the asset’s price rises faster than network hashrate. If Bitcoin jumps 50% and computing power stays flat, each active machine can earn more without facing a flood of new rivals. If hashrate grows faster than price, revenue per machine gets diluted even while the coin looks “healthy” on a chart.
That is why some treasury-minded executives still tell small allocators to buy the coin first. A tiny account does not need construction risk, transformer delays, or a machine that becomes scrap after one efficiency jump. Larger, diversified books can mix spot exposure with operating leverage. The operating piece has to earn its extra risk. If it cannot credibly aim to outperform Bitcoin itself, the simpler product wins.
If you only have $1, buy Bitcoin first. I think that’s the best way to express your view.
I tend to agree with the spirit of that advice, even if the dollar figure is theatrical. Mining is a business with schedule risk, counterparty risk, and hardware risk. Holding Bitcoin in a straightforward vehicle avoids most of that. The interesting debate starts only after an investor already has the core exposure and wants torque.
Energy Flexibility Is Still Mining’s Best Card
Critics like to say Bitcoin wastes electricity. Operators answer with a different frame. Mining is interruptible demand. ASICs can shut down and restart faster than a steel mill or a continuous industrial line. In grids with jumpy renewable output, that flexibility can be sold as a service. Miners curtail when the system is tight and spin up when power would otherwise be stranded.
Results depend on the fuel mix and the contract. A hydro site in a wet year does not look like a congested gas node in August. Still, the optionality is real. AI halls generally need steadier power because customer workloads hate interruption. That difference may keep Bitcoin in the mix at sites where electricity is abundant, cheap, or hard to transmit, but not reliable enough for a hyperscale tenant.
Think of four kinds of optionality that miners still claim:
- Flexible energy demand that can follow the grid.
- A larger network share when weaker competitors leave.
- Direct economic proximity to Bitcoin’s issuance schedule.
- Buildings and interconnects that can later host other compute.
Whether those options pay will show up in two places. Upcoming difficulty prints will reveal how much hash is truly gone. Public filings will reveal how much cash is still being reserved for new ASICs versus AI construction. I would watch both before declaring a permanent regime change.
How Difficulty Adjustments Change The Scoreboard
Bitcoin targets a ten-minute block on average. If blocks come in fast, difficulty rises. If blocks lag, difficulty falls. That mechanism is boring until you live inside it. Then it becomes the whole game.
A falling difficulty period can feel like oxygen for a low-cost site. The same machine finds blocks more often. Revenue per unit of hash can improve even if the coin price is sleepy. The catch is obvious. If price also slumps, or if power prices spike, the oxygen is not enough. Difficulty is a relative gift, not a profit guarantee.
Operators who understand this stop treating hashrate as a vanity metric. More hash is only useful if it is cheap hash. A company that grows its fleet into a rising difficulty tape can look busy and still dilute itself. A company that holds a lean fleet through a hashrate winter can look inactive and still gain share.
Public Markets Are Pricing A Split Identity
Listed miners used to be leveraged Bitcoin proxies with a power story attached. Many still are. The market is now trying to decide which names are becoming data center developers that happen to own ASICs. Valuation language is shifting with that identity fight. A pure miner is judged on hash, efficiency, and treasury policy. A hybrid is judged on contracted megawatts, counterparty quality, and delivery dates.
That split creates messy comparisons. Two companies can own similar power and print very different stories. One leans into coin accumulation. The other leans into multi-year compute leases. Neither path is automatically smarter. The first path lives and dies with Bitcoin. The second path lives and dies with construction execution and customer credit.
In my view, the danger is narrative whiplash. Boards that talk like AI developers still have mining depreciation on the books. Boards that talk like Bitcoin maximalists still shop GPU partners. Readers should ignore the slogan and follow the capex line. Money does not lie as easily as a keynote slide.
| Operator posture | What they optimize | Main risk |
| Core Bitcoin miner | Cheap hash and coin inventory | Hash price and power cost |
| Hybrid campus | Shared interconnect and mixed revenue | Execution and split focus |
| AI-first converter | Long contracts and uptime | Build delays and tenant quality |
What A Prolonged Hashrate Winter Means For Security Talk
Every time hashrate falls, someone asks whether the network is less secure. The honest answer is layered. Absolute hashrate is still vast by historical standards even after a 20% plus drawdown from the peak. Security is also about the cost of assembling enough competing power to attack the chain, not about whether the last record still stands.
A gradual decline caused by economics is not the same as a sudden political ban that concentrates hardware in one remaining region. The 2021 event was a geographic shock. The current event looks more like a return-on-capital shock. Geography still matters. So do machine ownership, pool concentration, and who controls the cheapest stranded power. Those are the better security questions.
I get uneasy when commentary treats any dip from an all-time high as an existential warning. Networks do not need a new record every quarter to remain robust. They need a competitive market of miners who can still profit at the margin. If that market thins too far, the conversation changes. We are not obviously there yet.
Small Capital Versus Operating Leverage
This is where the article stops being industry gossip and becomes an allocation problem. If you are expressing a simple view that Bitcoin should be worth more in five years, owning the asset is clean. You do not need to guess which transformer gets delivered on time.
If you want operating leverage, mining can amplify a bull tape in which price outruns hashrate and costs stay contained. It can also punish you when the opposite happens. AI conversion adds another layer. You might be underwriting a construction project with a Bitcoin ticker. That can work. It is not the same bet as holding coins.
A practical way to think about it:
- Small capital usually belongs in the coin, not the plant.
- Larger books can add miners after they accept project risk.
- Hybrid names need extra work on contract quality and delivery.
- Balance sheet stress can erase an otherwise decent site.
None of that is romantic. Good. Mining stopped being a cottage hobby years ago. It is industrial finance with a protocol wrapped around it.
Why The Current Cycle Feels Longer
Previous recoveries had a simple script. Price heals. Used machines get cheap. Power comes back online. Hashrate prints a new high. Investors celebrate the “network is stronger than ever” headline and move on.
The script now has an extra scene. Before the used-machine bid appears, a data center team asks whether the same substation can support a higher-rent tenant. Before the expansion order goes out, a board asks whether ASIC lead times still make sense against GPU demand. Those questions add calendar time. That is why this slump can last even if Bitcoin’s price is not collapsing in a straight line.
There is also a measurement problem. Because estimates swing, commentators can claim a recovery after three strong days and a crisis after three weak ones. Resist that. Look at multi-week averages, difficulty epochs, and actual machine purchase commentary from operators. The tape of tweets is not a hashrate index.
The Cost Curve Is The Real Market
Forget the slogan that “all miners are leaving.” The better model is a sorting process. High-cost, high-debt, old-fleet operators feel the squeeze first. Efficient operators with long power and newer silicon keep running. Some of the departing power does not vanish from the economy. It gets re-priced into another compute market.
That sorting can be healthy for the survivors and still look ugly in the aggregate hashrate chart. Markets often confuse those two pictures. An industry can shrink at the top line and improve at the best operators. Equity investors need the second picture. Protocol watchers stare at the first.
I’ve sat with enough operator models to know the spreadsheet is unforgiving. Change the power price by a penny and the whole year moves. Change uptime by a few points and the debt covenant starts sweating. Change hash price and yesterday’s expansion plan becomes a museum piece. AI does not erase that math. It offers an alternative column.
What To Watch Over The Next Difficulty Windows
If you want a practical checklist instead of a vibe, start here. Watch whether difficulty keeps easing or stabilizes. Watch whether public miners still disclose ASIC purchase commitments. Watch whether AI revenue is contracted or merely advertised. Watch power markets in the regions that already host large fleets. Watch whether older machines reappear after a difficulty cut or stay in warehouses.
A true hashrate spring would show up as more than a couple of strong estimate days. It would show up as new energized capacity that is explicitly dedicated to Bitcoin, not dual-use slides. Until that appears, the cautious reading remains the better one. The network is in a long pause at the high end of installed compute.
Simple miner scoreboard: Price speed versus hashrate speed Power cost versus hash price Debt clock versus difficulty relief ASIC spend versus AI construction
A More Human Way To Read The Moment
People get mystical about Bitcoin infrastructure. They talk as if machines have loyalty. Machines do not have loyalty. Owners have hurdle rates. When another industry offers a cleaner spread on the same interconnect, some owners will take it. That is not betrayal. That is business.
The protocol still does what it was designed to do. It adjusts. It keeps blocks coming. It pays whoever remains. The romance is optional. The adjustment is not.
So where does that leave a reader who is not running a mine? It leaves you with a cleaner map. Bitcoin’s price path still dominates most non-operating portfolios. Mining exposure is a specialist overlay. AI conversion is a construction and contract overlay on top of that. Mixing the three without noticing is how people get surprised.
I do not think this is the end of Bitcoin mining. I think it is the end of the assumption that every new watt automatically becomes more hash. That assumption had a good run. It is meeting a rival use case with deeper pockets and pickier uptime needs. Some sites will stay with Bitcoin because flexibility still pays. Some will leave because contracted compute pays more. Most will hover in between and make the next two years feel uneven.
Uneven is not the same as broken. If you remember only one thing from this stretch, remember that. A hashrate bear market is a statement about capital at the margin. It is not a verdict on the asset, and it is not a promise that the next high will arrive on schedule. The next high, if it comes, will have to compete for electricity in a way the last cycle never did.
The Bottom Line Without The Slogan
Bitcoin’s estimated computing power remains well below its late 2025 record after a long, grinding retreat. The useful label for that retreat is a hashrate bear market, even if the phrase is unofficial. The cause is not a single ban or a single outage. It is a slower contest between two hungry users of power and industrial land.
Surviving miners can gain share when difficulty eases. They can still lose money if price, fees, and power refuse to cooperate. Investors can own the coin without owning the plant. Operators can chase AI without abandoning every ASIC they already paid for. All of those statements can be true at the same time. Markets hate that kind of sentence. It happens to be the accurate one.
Keep an eye on the next difficulty cuts, the next capex footnotes, and the next quarter in which a miner has to explain whether new megawatts were born for hashing or for rented compute. That is where this story stops being a conference line and becomes a set of numbers you can actually use.
