Bitcoin’s Hashrate Shrinks Again as Q3 Redraws the Map

Bitcoin closed the third quarter of 2026 with a statistic that would have read like a typo two years ago: its total computing power, the hashrate, finished the quarter lower than it started, and lower than the quarter before that. Luxor’s Hashrate Index put the network at roughly 940 exahashes per second (EH/s) in its Q3 2026 hashrate heatmap, down about 6.3% from the second quarter and around 12% below the record of roughly 1,066 EH/s set in December 2025.

For most of Bitcoin’s life the only interesting direction for this number was up. The network crossed 100 EH/s in January 2020 and passed its first zettahash (1,000 EH/s, on a seven-day average) around mid-September 2025, per TFTC’s growth timeline. The idea that it could fall for two quarters in a row, while the price sat near $83,000 and no government had banned anything, does not fit the script the last cycle wrote.

That is the quarter in one line. Bitcoin’s security engine is getting smaller, and for the first time it is doing so on purpose.

The number that closed the quarter

A quarter-end snapshot frames the rest of this piece. Every figure below is a moving target (hashprice and hashrate especially), so treat them as a photograph taken on the last day of September, not a fixed constant.

What hashrate actually measures, and why it is a guess

Hashrate counts guesses. Bitcoin mining is a lottery in which specialized machines (ASICs) run the SHA-256 hash function over and over, each hash a ticket in a draw that happens roughly every ten minutes. Hashrate is how many of those tickets the whole network buys per second, quoted these days in exahashes (10 to the 18th) and zettahashes (10 to the 21st).

The catch, and it matters for every headline about hashrate rising or falling, is that nobody can measure it directly. There is no meter on the network. Hashrate is inferred from two things the blockchain does record: the difficulty setting and how fast blocks are actually arriving. The standard estimate is difficulty multiplied by 2 to the 32nd, divided by the average block time in seconds, which is why a difficulty of 132.76T and roughly ten-minute blocks imply something close to 950 EH/s.

Because block discovery is random (a Poisson process), short windows are noisy: a quiet afternoon can print a spot reading near 1 ZH/s while a slow morning shows 850 EH/s, with nothing real changing underneath. Serious trackers smooth the estimate over seven or thirty days, and two reputable dashboards can disagree by 10% at any given instant. When this article says the network sits near 940 EH/s, it means the multi-day average the Hashrate Index heatmap uses, not any single tick. Keep that margin of error in mind: a quarterly move of 6% is real, a daily swing of the same size usually is not.

Two quarters of shrinking, and why that is new

The shape of the decline matters more than the number. Bitcoin’s hashrate peaked near 1.1 to 1.16 ZH/s in late October 2025, when the price was printing its own record above $126,000. From there it drifted lower: the Hashrate Index heatmap measured about 1,004 EH/s in the second quarter, down 5.8% from the first-quarter base, then roughly 940 in the third, another 6.3% lower. Two consecutive quarterly declines in the same series, and a network that has been sliding off its December high for most of the year.

Bitcoin has seen hashrate fall before. It has never fallen quite like this.

The 2026 row is the odd one out. Difficulty has now fallen year on year, only the second time that has happened in Bitcoin’s history (the first was the 2021 China exodus), yet there is no exodus and no crash, as CoinDesk documented in August. JPMorgan’s Nikolaos Panigirtzoglou described the mechanism cleanly earlier this year: “When bitcoin trades below its production cost, higher-cost miners power down, the hashrate declines, and difficulty adjusts lower,” a line he offered when the bank pegged the average all-in production cost near $78,000 (via TFTC). For stretches of 2026 the price did sit below that, so the least efficient rigs went dark. But the deeper driver is newer than price: the same megawatts are now worth more doing something else.

The Q3 leaderboard: a map being redrawn

If the aggregate number is shrinking, the map underneath it is being redrawn. The Q3 heatmap keeps the top of the table familiar and churns the bottom.

The United States, Russia, and China still hold roughly two-thirds of the world’s hashrate between them, a concentration that has barely moved in a year. But the edges tell the real story. Norway entered the top 10 for the first time. Kazakhstan, which briefly commanded something like 18% of the network after the 2021 China ban, fell out of the top 10 for the first time in years, down to about 15 EH/s. And the fastest growth on the board is happening in places that were rounding errors twelve months ago.

Why the United States is loosening its grip

The US is still the center of gravity, and still number one by a wide margin. But it is also the biggest single loser of raw hashrate this year. Its share slipped from about 37.5% in the first quarter to 36.7% by late in the third, and Hashrate Index estimates the country shed roughly 55 EH/s of computing power over that stretch even as it kept the top spot, a slide tracked by Bitcoin.com News.

The reason is the same force pulling the aggregate down. American miners are sitting on exactly the asset the artificial-intelligence build-out wants: interconnected, energy-hungry real estate already wired into a grid. In Texas, where much of the US fleet lives, close to 90% of the grid interconnection queue is now AI and high-performance computing rather than crypto, per US Energy Information Administration data on the ERCOT system. A megawatt that earns a mining margin measured in single-digit cents can instead earn a fixed, long-dated lease from a data-center tenant. Riot Platforms leased 191 MW of a Texas campus to what it calls a leading frontier AI lab (reported by CNBC and others to be Anthropic) on a 20-year contract worth up to $16.1 billion; MARA is redirecting its own gigawatts toward the same demand. The hashrate those sites would have added simply never plugged in, or unplugged to make room.

There is a subtler wrinkle here. For most of 2025 the American mining and AI booms looked additive, both expanding at once; in 2026 they started to look competitive, with the AI queue crowding miners for the same interconnections and the same substations. The miners that thrive in that environment are the ones that lean into being flexible grid assets, earning demand-response and curtailment credits by powering down when the grid is stressed, a line of business Riot alone booked tens of millions of dollars from last year. Being a good grid citizen, not just a cheap one, is turning into a competitive edge.

The frontier is where the growth went

While the US shed exahashes, the growth showed up on the frontier, in countries chasing what miners have always chased: the cheapest stranded electricity on earth. Year over year, the heatmap clocked Pakistan up 733% (from a rounding error to about 2.5 EH/s), Bolivia up 300%, Kyrgyzstan up 167%, and Ethiopia up 35% to roughly 23 EH/s on the back of hydropower from the Grand Ethiopian Renaissance Dam.

Paraguay is the clearest winner of the reshuffle. At roughly 44 EH/s and 4.7% of the network it is now the fourth-largest mining country on earth, up more than 50% year on year, powered by surplus hydro from the Itaipu and Yacyreta dams; operators such as HIVE Digital have been expanding a single Paraguayan campus toward 400 MW. It is the exception that tests the rule, cheap power that has so far proven durable because the dams are enormous and the grid has few competing buyers.

These are real megawatts, and they are fragile ones. Cheap power is usually cheap for a structural reason (a subsidy, a stranded asset, a grid with no other buyer), and that reason can change fast. Iran is the cautionary tale of the quarter: its hashrate fell 71% year on year, from about 7 EH/s to 2, as regional conflict knocked capacity offline. Kazakhstan is the cautionary tale of the whole cycle: the same country that absorbed the 2021 China diaspora and briefly ran close to 18% of the network throttled miners with grid caps and a state-run electricity marketplace, and has now dropped out of the top 10 entirely. Frontier hashrate migrates in, and it migrates out.

Norway in, Kazakhstan out

The single most telling swap on the Q3 board is Norway replacing Kazakhstan. It is telling precisely because Norway did not surge. It held flat, at about 16 EH/s, while the network around it shrank, and holding flat was enough to lift it into the top 10.

Luxor research analyst Kaan Farahani framed the dynamic directly in the heatmap report: “In a shrinking network, holding flat is a win, it usually signals stable, low-cost operations that don’t need to curtail.” Norway’s hydro-heavy grid, cold climate, and political stability are the opposite of a frontier bet. Kazakhstan’s cheap coal power came bundled with a grid that could not carry it and a government that eventually clamped down. In a growing network, the frontier’s cheap-but-fragile megawatts win the marginal race. In a shrinking one, the boring, durable megawatts are the ones left standing.

That inversion is the quiet thesis of the quarter, and it runs against the reflex that mining always flows to the lowest price per kilowatt-hour. When the whole network is contracting, resilience beats raw cheapness.

The engine of the decline: AI and the marginal megawatt

Zoom out and every thread on the map ties back to one competition: Bitcoin miners and AI data centers now bid for the same megawatts, and for the first time the AI bid is usually the higher one.

CoinShares, whose quarterly mining reports track the public fleet, estimates the sector has signed more than $70 billion in cumulative AI and HPC contracts, and that some listed miners will draw as much as 70% of revenue from AI by the end of 2026, up from roughly 30% today, per research head James Butterfill (via Bitcoin.com News). In its Q2 2026 report the firm flagged a threshold moment: at IREN, one of the larger operators, AI cloud revenue (about $70.5 million) overtook Bitcoin mining revenue (about $66.7 million) for the first time, and it called the AI transition substantially complete across parts of the sector by year-end.

The scale of the reallocation is hard to overstate. Core Scientific has signed multi-gigawatt colocation deals, and Hut 8, TeraWulf, and IREN have each contracted billions of dollars of high-performance-computing capacity. The common thread is that these are fixed, multi-year, investment-grade revenues that a volatile mining margin cannot match. For a public miner answerable to shareholders, converting a megawatt from mining to an AI lease is often the rational call even when Bitcoin’s price is climbing, which is a large part of why the network can keep shrinking straight into a rally.

The economics are not subtle. “The revenue per megawatt and EBITDA margins are far higher for HPC and AI colocation than for mining,” Needham analyst John Todaro has said. MARA chief executive Fred Thiel put the mining half in blunter terms: it “is a zero-sum game. As more people add capacity, it gets harder for everybody else. Margins compress, and the floor is your energy cost” (both via CoinGeek). The projects fighting hardest for those megawatts, from centralized hyperscalers to decentralized training networks and GPU marketplaces like Akash, are the reason a Texas megawatt can be worth more idle-for-AI than busy-for-Bitcoin.

Ethan Vera, Luxor’s chief operating officer, argues the miners that remain are being repriced rather than replaced: “Miners are worth more to the grid than people realize. They are flexible loads that can power down instantly when supply is tight, something the AI capacity replacing them can’t yet do.” That flexibility, the ability to shut off in a heartbeat and sell the power back, is exactly what a strained grid values and what an always-on AI campus cannot offer. It is also why the miners still running are increasingly the ones with the cheapest, most interruptible power arrangements.

The thermostat that keeps a shrinking network safe

A network that loses 12% of its computing power sounds like a network in trouble. It is not, because of the single most elegant piece of Bitcoin’s design: the difficulty adjustment.

Every 2,016 blocks, roughly every two weeks, the protocol measures how long those blocks actually took and resets the mining difficulty to steer the next batch back toward ten-minute blocks. Satoshi Nakamoto described it in the white paper as a moving average: “To compensate for increasing hardware speed and varying interest in running nodes over time, the proof-of-work difficulty is determined by a moving average targeting an average number of blocks per hour. If they’re generated too fast, the difficulty increases.” Run that in reverse and you have 2026’s whole story: fewer machines means slower blocks means lower difficulty, which lowers the cost of mining for whoever stays, which lets the network settle at a smaller size without breaking.

The 2026 record shows the thermostat working in both directions. Difficulty bottomed near 124.9T in mid-June, and after the summer’s price recovery pulled idled rigs back on, it climbed again: +1.31% on 5 September, then +4.16% on 19 September, the biggest single jump since June. It now sits at 132.76T, about 15% below its November 2025 record, with the next adjustment due around 3 October estimated at a slight +0.63%, per CoinWarz. The mechanism is the same one this network has leaned on since 2009. It is really a rolling auction for the cheapest power, a framing explored in Bitcoin’s Difficulty Is Really a Bidding War for Power.

Price leads, hashrate lags

The reason the map redrew in the third quarter, and not the first, is that price moved first. Bitcoin spent the summer below many miners’ cost of production, bottoming near $60,000, then rallied hard after the Federal Reserve’s surprise September rate hike, its first since 2023, and climbed back above $83,000, a move traced in Crypto’s FOMC Reaction: A Hawkish Fed and an 8-Month High. Miner revenue, expressed as hashprice (the dollars of daily revenue per petahash of capacity), followed: it recovered from summer lows near $27 to $32 per PH/day up to roughly $40, a gain of about 27% in a month, per Bitcoin.com News.

Hashrate is the lagging variable in that chain. Rigs unplugged in July do not come back the day price recovers; leases have to be signed, machines re-energized, sometimes shipped across borders. JPMorgan pegged the sensitivity of difficulty to price at a beta of about 0.62, which means the network’s response is real but muted and slow. So the map you read at the end of a quarter is a photograph of where the cheapest power met the highest bidder two or three months earlier, not where it will be next week. The rally that lifted Bitcoin in September will show up in the hashrate map, if it holds, closer to the end of the year.

Efficiency is quietly doing the opposite of the headline

Here is the counterintuitive part. Even as raw hashrate falls, the network is arguably getting stronger per unit of energy, because the machines still running are far more efficient than the ones switching off. As Luxor’s analysts note, a shrinking network sheds its least efficient hardware first, while the capacity that stays online keeps migrating toward the cheapest power.

The hardware backs this up. In 2016 a flagship Antminer S9 needed about 98 joules per terahash. Today’s flagships are near 9.5 J/TH: Bitmain’s S23 Hydro broke the sub-10 barrier, and Bitdeer’s SEALMINER A4 edged it at about 9.45 J/TH. That is close to a tenfold efficiency gain in a decade. A network at 940 EH/s of 2026 silicon draws less power, and represents more durable, lower-cost security, than a network at 1,066 EH/s did when a meaningful chunk of it ran on older, marginal rigs. Cambridge’s index still puts Bitcoin’s total draw near 138 terawatt-hours a year, about 0.5% of global electricity, per the Cambridge Bitcoin Electricity Consumption Index, but each of those watt-hours now buys more hashes than ever before.

This is the nuance the headline number hides. Fewer exahashes does not mechanically mean less security, because the exahashes that left were the expensive, easy-to-outbid ones. What remains is leaner and harder to dislodge.

What a smaller network means for security

The blunt question a shrinking hashrate raises is whether Bitcoin is now easier to attack. On the numbers, barely.

A 51% attack requires assembling more hashrate than the honest network, and even at 940 EH/s that is an enormous ask. Duke finance professor Campbell Harvey modeled a one-week majority-hashrate attack last year at roughly $6 billion in hardware, data-center, and power costs (via crypto.news). He has since argued the real threat is cheaper, because an attacker could pair the hashrate with a large short position in Bitcoin derivatives and profit from the crash the attack itself causes. His refined framing: “The cost is about 50 basis points of the value of bitcoin,” and “the difference today is the derivatives markets” (via Bitcoin.com News). Even that is billions of dollars for an attack that would likely be spotted and frozen first. Matt Prusak, president of American Bitcoin, has pushed back that “economic feasibility kills the 51% thesis.”

It is worth being precise about what such an attack could and could not do. Even a majority miner cannot steal coins from wallets or lift the 21-million supply cap; the realistic damage is double-spending recent transactions and censoring others, which is why exchanges wait for confirmations before crediting large deposits. More hashrate makes that attack costlier, but it does nothing about the network’s one known theoretical quirk, the timewarp flaw in the difficulty rule, which is a bug fixed by code rather than by raw power. A smaller honest network changes the price of a 51% attack; it does not open a new class of them.

The more useful frame is the security budget: the total dollars miners earn per year, which is what an attacker has to out-compete. At roughly 450 BTC minted per day plus fees, times an $83,000 price, that budget runs near $13 billion a year. A 12% smaller hashrate does not cut it, because the budget is set by price and the block subsidy, not by machine count; difficulty just redistributes the same reward among fewer machines. This is the mirror image of proof-of-stake, where security comes from bonded capital rather than burned electricity, a contrast drawn out in Who Controls Your Staked ETH. Bitcoin has never suffered a successful 51% attack in more than fifteen years, and a two-quarter dip well within historical noise does not change that.

The clock behind the map: the 2028 halving

Every hashrate map is drawn under a countdown. Around 2028, at block 1,050,000, Bitcoin’s block subsidy halves again, from 3.125 BTC to 1.5625. At a constant price that halves the biggest component of the security budget overnight, and with transaction fees still running well under 1% of the block reward, there is no fee market yet ready to fill the gap.

That is why the composition of the map, not just its total, matters more with each cycle. When the subsidy halves, the miners who survive are the ones with the lowest all-in cost per coin, which increasingly means the cheapest and most flexible power. CoinShares put the public fleet’s weighted-average cash cost near $75,500 per coin in its Q2 2026 report, with the most efficient operators far below that (IREN’s electricity cost ran around $25,942 per coin) and the least efficient already underwater. The 2026 map, with its durable Norwegian hydro rising and its fragile frontier plays churning, is a preview of the sorting the 2028 halving will accelerate. Thiel’s forecast for miners is characteristically stark: “By 2028, you’ll either be a power generator, be owned by one, or be partnered with one.”

How to read the next quarter’s map

Will the network keep shrinking? The honest answer is that it depends on the same two variables that redrew this quarter’s map: the Bitcoin price and the price of a megawatt.

CoinShares’ base case still has hashrate climbing back toward 1.8 ZH by the end of 2026 and 2 ZH by early 2027, but explicitly on the condition that Bitcoin recovers toward $100,000; below roughly $80,000, its model has hashprice falling, machines switching off, and the network shrinking further. Here is what to watch as the fourth quarter unfolds:

• Price versus production cost. Sustained trading above the ~$78,000 to $80,000 average production cost pulls idled rigs back on; sustained trading below it pushes more offline.
• The difficulty prints. Each retarget is a two-week readout of whether machines are switching on or off; a string of increases means the fleet is growing again.
• AI lease flow. Every new gigawatt leased to a data-center tenant is capacity that will not mine, so watch the miners’ contract announcements as closely as their hashrate.
• Frontier fragility. Watch whether Pakistan, Bolivia, and Kyrgyzstan hold their gains or follow Kazakhstan’s arc from boom to crackdown.

None of this changes Bitcoin’s regulatory footing in the United States, where the SEC’s Division of Corporation Finance stated in March 2025 that proof-of-work mining, whether solo or pooled, does not involve the offer or sale of a security (SEC statement). Mining remains a commodity business dressed as a technology one, and its map remains a live picture of where the world’s cheapest, most flexible electricity is and who is willing to outbid a Bitcoin miner for it. For one quarter, and probably for a second in a row, the answer has increasingly been an AI company.

Frequently Asked Questions

Why is Bitcoin’s hashrate falling in 2026?

Two forces are at work. Bitcoin traded below many miners’ cost of production for much of the year, so the least efficient machines switched off and the difficulty adjustment let the network stabilize at a smaller size. At the same time, AI and high-performance computing tenants now outbid miners for the same electricity, so a growing share of megawatts is leased to data centers instead of running ASICs. The result is the first multi-quarter decline that is not caused by a price crash or a government ban.

What is Bitcoin’s hashrate right now?

Luxor’s Hashrate Index put the network near 940 exahashes per second at the end of the third quarter of 2026, about 6.3% below the prior quarter and roughly 12% under the record of about 1,066 EH/s set in December 2025. Because hashrate is estimated from difficulty and block times rather than measured directly, single readings are noisy, so multi-day averages are the reliable figure.

Does a lower hashrate make Bitcoin less secure?

Only marginally. A 51% attack still requires out-computing the honest network, which even at 940 EH/s would cost billions of dollars, and Bitcoin’s security budget (the roughly $13 billion a year miners earn) is set by price and the block subsidy, not by machine count. The network has never suffered a successful 51% attack, and a two-quarter dip is within historical noise.

Which countries mine the most Bitcoin in 2026?

The United States leads with about 36.7% of hashrate, followed by Russia near 17.2% and China around 12.2%, and together they hold roughly two-thirds of the network. Paraguay, the UAE, Oman, Canada, Ethiopia, Indonesia, and Norway round out the top 10. Norway entered the top 10 in the third quarter while Kazakhstan fell out for the first time in years.

What happens to Bitcoin’s hashrate at the 2028 halving?

Around 2028 the block subsidy halves from 3.125 to 1.5625 BTC, cutting the biggest part of miner revenue at a constant price. Unless the price rises or transaction fees grow well beyond today’s sub-1% share of the reward, the least efficient miners will be squeezed out, concentrating the network further among operators with the cheapest, most flexible power.

By Marcus Okafor, Bitcoin and mining desk, HOGE Wire.