A Year After the Zettahash: Bitcoin’s Hashrate Stall

Twelve months ago, Bitcoin’s mining machines crossed a line with a round number attached to it. In early September 2025, the seven-day average hashrate held above one zettahash per second for the first time: a thousand exahashes, a one trailed by twenty-one zeroes of guesses every second. Glassnode data reported by CoinDesk put the moment at the start of that month, noting the figure had briefly touched a zettahash earlier in the year but had never before stayed there on a smoothed average.The trajectory looked one-directional. Hashrate had climbed from roughly 800 EH/s at the start of 2025 to a zettahash by September, and analysts were already drawing the line to two. Instead the curve bent back. The network peaked near 1,162 EH/s on 19 October 2025, slid through the winter, and on 17 January 2026 dropped below the zettahash for the first time since the milestone, according to Yellow, which described the end of a four-month stretch above the symbolic mark. It has spent the eight months since stuck in a band well beneath it.So a year after the zettahash, Bitcoin’s hashrate sits lower than the day it first arrived there. That has happened before, but only when a government banned mining or a price crash bankrupted the people running the machines. This time no ban was passed and no crash forced the issue. The difference between this plateau and every earlier one is the whole story, and it says as much about artificial intelligence as it does about Bitcoin.

The twelve months that bent the curve

For most of Bitcoin’s history the hashrate chart has had one shape: up and to the right, steeper after every price rally, with brief dips that filled back in within weeks. The run to a zettahash fit that pattern. The seven-day average went from about 800 EH/s in January 2025 to 1,000 EH/s in September, a 25% gain in nine months, and the jump was steep enough that CoinDesk flagged an imminent difficulty adjustment of more than 7%, at the time the second-largest upward move of the year.Then it stopped fitting the pattern. The peak came a few weeks later, near 1,162 EH/s in mid-October 2025. From there the line drifted down rather than paused: below the zettahash in January, a first-quarter decline that broke a six-year run of quarterly growth, and a summer spent oscillating between roughly 850 and 950 EH/s depending on which tracker and which averaging window you trust. What makes 2026 unusual is not the size of the dip, which is modest by historical standards, but its shape. Earlier declines were sharp V’s that snapped back. This one is a long, shallow plateau that will not lift. The table below marks the stations on that round trip.

Date	Hashrate (7-day average)	What it meant
Jan 2025	~800 EH/s	Starting line for the zettahash run
Early Sep 2025	~1,000 EH/s (1 ZH/s)	First time the 7-day average held a zettahash
19 Oct 2025	~1,162 EH/s	The actual peak to date (about 1.16 ZH/s)
17 Jan 2026	~988 EH/s	First drop back below 1 ZH/s, ending a four-month streak
Q1 2026	~1,000 EH/s	First quarterly decline in six years
Sep 2026	~915 to 940 EH/s	Stalled well below the mark it first crossed a year earlier

Reading the number: an estimate, not a measurement

Before going further, a caveat that matters more here than usual: nobody measures Bitcoin’s hashrate directly. Miners do not report how many machines they run, and most would rather not. The figure quoted everywhere is inferred from how fast blocks are actually found compared with how hard the network has made them, using the relationship that hashrate is roughly difficulty times 2^32 divided by the average block time. Hashrate Index, which publishes one of the more-watched estimates, puts it plainly: it is not possible to pinpoint the exact hashrate, only to produce a ballpark from a moving average of recent blocks.Because block discovery is random (a Poisson process, in the jargon), short windows are noisy and long windows lag. That is why two reputable trackers can disagree by tens of exahashes on the same afternoon, and why a single fast run of blocks can print a spot reading near a zettahash while the smoothed seven-day figure sits closer to 915 EH/s. It is also why the peak is contested. Smoothed data from Glassnode, Hashrate Index and CoinShares all place the October 2025 high around 1.1 to 1.16 ZH/s, yet some public dashboards, including CoinWarz, still label a 1.44 ZH/s reading from 20 September 2025 as the all-time high. That number does not survive contact with any averaged series; it is best read as a noisy single-interval spike, not a level the network ever actually reached. When a chart tells you the network is running at 70% of its record, it is worth checking whether the record is real. Units help keep the scale straight.

Unit	Hashes per second	Scale
MH/s (megahash)	1,000,000	Early CPU and GPU mining
GH/s (gigahash)	1,000,000,000	First-generation ASICs (2013)
TH/s (terahash)	10^12	One modern mining rig
PH/s (petahash)	10^15	A small farm; the unit hashprice is quoted in
EH/s (exahash)	10^18	The unit the whole network is measured in
ZH/s (zettahash)	10^21	1,000 EH/s; crossed for the first time in Sep 2025

The stall in numbers

Strip out the noise and the picture is consistent. On the difficulty side, which is the one number the protocol fixes exactly rather than estimates, the network sits at 127.45 trillion after an adjustment on 6 September 2026, according to mining coverage from Bitcoin.com News. That is roughly 18% below the all-time high near 156 trillion set in November 2025, and it is lower than where 2026 began. Through early September the year had recorded eight upward adjustments against ten cuts, a net decline of about 12% on the year, per the same report. Difficulty falling year-over-year has happened only a couple of times in Bitcoin’s history, and every prior instance coincided with a crackdown or a crash.Miner revenue, by contrast, has recovered. Hashprice, the daily revenue a miner earns per unit of hashing power, climbed to about $39.63 per petahash per day in early September, up more than 22% in a month as Bitcoin’s late-August rally lifted the dollar value of the block reward. That is the tension of the moment in one line: the pay for hashing went up, yet the amount of hashing barely moved. The next difficulty retarget, estimated for 19 September, is set to rise by roughly 5% to reflect a recent run of faster blocks, per CoinWarz, which would be the largest upward move in months and the clearest sign yet that the floor has been found.

Metric	Value (mid-September 2026)	Source
Network hashrate (7-day)	~915 to 940 EH/s	Hashrate Index / Bitcoin.com News
Difficulty	127.45T (about 18% below Nov 2025 ATH)	CoinWarz / Bitcoin.com News
Next retarget (est. 19 Sep)	about +5%	CoinWarz
Hashprice	~$39.63 / PH/day (+22% MoM)	Bitcoin.com News
BTC price	~$75,700	CoinGecko
2026 difficulty (net YTD)	roughly -12% (8 up, 10 down)	Bitcoin.com News

The hashrate vacuum: why growth stopped

Bitcoin’s hashrate has fallen three times in a serious way, and it is worth separating the causes. In 2018 and again in 2022 the driver was price: a bear market pushed the marginal miner below break-even, machines switched off, and hashrate fell until difficulty dropped far enough to make the survivors profitable again. In 2021 the driver was policy: China banned mining outright, knocking out more than half the network in weeks before it migrated abroad. Both were shocks that resolved themselves, and in both cases the recovery was faster than the fall.The 2026 plateau is neither. No country banned anything, and while Bitcoin trades well off its highs it is not in a 2018-style collapse. What changed is that the machines competing for a miner’s electricity are no longer only other miners. The same megawatts, the same substations, the same cheap-power sites that a mining company would have filled with ASICs are now worth far more filled with GPUs training and serving artificial-intelligence models. The industry has a name for the result, a hashrate vacuum, where capacity that would once have gone to Bitcoin gets pulled toward AI and high-performance computing instead (millionminer.com). A widely cited estimate tallies more than $70 billion in cumulative AI and HPC contracts signed by listed miners, and reckons AI could supply up to 70% of their revenue by the end of 2026, up from roughly 30% now, figures from CoinShares reported by Bitcoin.com News.The math behind the vacuum is not subtle. A megawatt pointed at Bitcoin earns whatever that megawatt can hash multiplied by hashprice, a figure that swings with the coin and that spent much of 2026 hovering near the cost of the electricity itself. The same megawatt leased to an AI tenant earns a contracted rate, often locked in for years, frequently at several times the revenue density and without the price risk. Faced with that spread, a miner’s board does not agonize; it signs the lease and points the guaranteed cash flow at whatever comes next. Every such decision quietly subtracts capacity that would otherwise have shown up as hashrate, which is how the network can be starved of growth even while the mining companies themselves report record revenue.Jamie Redman, who tracks the mining sector for Bitcoin.com News, framed the stall this way after the September difficulty rise: “The flat hashrate indicates that miners are evolving very carefully, and they might be facing hardware limits.” The marginal decision is no longer buy more rigs or switch some off; it is mine another coin or rent the rack to a hyperscaler. For most of the large public miners, the rack is winning, and the network-wide flatline is what that choice looks like when you add it up.

The miner scoreboard: follow the megawatts

If growth is now decided in miner boardrooms rather than out on the open network, the boardrooms are worth watching. The largest public miners split into two camps: pure-play accumulators still pointing megawatts at Bitcoin, and hybrids increasingly leasing capacity to AI tenants. The most striking convert is CleanSpark, long one of the most single-minded Bitcoin miners, which spent 2026 pivoting toward AI and HPC after its chief executive had earlier dismissed the business as complex and uncertain; the company went on to win an AI power bid in Wyoming and to line up further capacity in Texas. Riot Platforms went further, agreeing to convert part of its Texas power to a long-term lease with the AI developer Anthropic. Even the pure-plays feel the pull: when a megawatt of mining earns a fraction of what the same megawatt earns hosting AI, the pressure to switch is structural, not sentimental.Others are still racing. MARA runs the largest listed fleet and is still mining-first, though it too has begun exploring AI uses for its power; American Bitcoin, the Hut 8 and Trump-family venture, mines a few percent of the network as an unapologetic accumulator; and HIVE Digital is building hydro-powered capacity in Paraguay while adding HPC and AI growth of its own. The direction of travel is clearest in the scoreboard below, where the right-hand column, not the fleet size, is doing the talking.

Miner	Role in 2026	AI / HPC direction
MARA	Largest listed miner by fleet	Mining-first, exploring AI uses for power
Riot Platforms	Major Texas operator	Converting power to a long-term Anthropic AI lease
CleanSpark	Former pure-play miner	Pivoting to AI and HPC
American Bitcoin	Hut 8 and Trump-family venture	Pure-play Bitcoin accumulator
HIVE Digital	Hydro-powered operator	Paraguay build-out plus HPC and AI

Sources: CoinShares via Bitcoin.com News for the AI-pivot scale; company disclosures for posture. Fleet sizes shift quarter to quarter and are treated qualitatively here.

Efficiency, the engine that never stalled

There is one growth engine the plateau did not touch, and it explains why hashrate can rise even when miners add no new megawatts at all. Each generation of application-specific chip does more hashing per watt than the last, so the same power bill buys more terahashes every cycle. The improvement has been relentless. Bitmain’s Antminer S9, the workhorse of 2016, managed about 98 joules per terahash; the S19 series in 2020 cut that to the low 30s; the S21 XP in 2024 reached roughly 13.5 J/TH, per Bitmain; and the S23 Hydro shipping in 2026 is the first mainstream miner rated below 10 J/TH, at about 9.5, per reporting from CoinDesk. That is close to a tenfold efficiency gain in a decade.The practical effect is that a site can retire old machines, keep its power draw flat, and still raise its hashrate simply by swapping in newer hardware. So the 2026 stall is not a story of stagnant technology; the chips keep getting better on schedule. It is a story about where the electricity goes. When efficiency gains are the only growth lever still being pulled, and the cheapest new power is being bid away by AI, the curve flattens. It also means any restart could be quick: a wave of hardware upgrades, or a block of power freed up when an AI contract lapses, can lift hashrate without waiting for whole new data centers to be built.

Miner (year)	Hashrate	Efficiency (J/TH)
Antminer S9 (2016)	13.5 TH/s	~98
Antminer S19 (2020)	95 TH/s	~34
Antminer S21 XP (2024)	270 TH/s	~13.5
Antminer S23 Hydro (2026)	~560 TH/s	~9.5

Price leads, hashrate lags, and the Fed

Miners do not respond to news; they respond to the dollar value of a block, and that is set by price. This is why hashrate is a lagging indicator: a rally lifts hashprice and, months later, funds new machines, while a sell-off squeezes margins and, eventually, switches machines off. Right now the price signal is pointing down. Bitcoin traded around $75,700 in mid-September, roughly 40% below its October 2025 record near $126,080, according to CoinGecko. That is a long way from the euphoria that carried the network to a zettahash in the first place.The macro backdrop is the reason. The Federal Reserve, now chaired by Kevin Warsh, is expected to raise its benchmark rate by a quarter point later today, to a range of 3.75% to 4.00%, its first hike since 2023, with an energy-driven inflation scare and a July hold having, as JPMorgan Chase’s own analysts put it, lowered the bar for a move. Going into the meeting, prediction markets and bank desks treated a hike as the overwhelmingly likely outcome. Higher rates pull money toward yield and away from non-yielding assets like Bitcoin, which is how the mining industry, of all things, ends up hostage to a rate decision; the same countdown of macro and regulatory catalysts we mapped in our look at crypto’s September zero hour runs straight through the hashrate chart. For readers curious how those odds get priced in the first place, the mechanics sit in our piece on prediction markets in 2026.

The climb to two zettahashes that did not happen

A year ago, two zettahashes looked like a matter of time. The path from 800 EH/s to 1,000 in nine months implied the second thousand would arrive on a similar clock. CoinShares still forecasts it, but with a condition attached. Its Q1 2026 mining report projected hashrate reaching 1.8 ZH/s by the end of 2026 and 2 ZH/s by the end of March 2027, explicitly contingent on Bitcoin recovering toward $100,000. The report’s scenarios spell out the dependence: hashprice near $37 per petahash per day if Bitcoin reaches about $100,000, closer to $59 if it retests its record, and a further decline in hashrate if the price stays below $80,000.With Bitcoin at $75,700 and the Fed tightening, the sub-$80,000 scenario is the one playing out. James Butterfill, head of research at CoinShares, has called this one of the most challenging periods the industry has faced, with a Q4 2025 all-in cash cost to mine one Bitcoin that his team pegged near $79,995, uncomfortably close to the coin’s current price. When the cost to produce a Bitcoin sits right on top of its market value, the incentive to add hashing power evaporates, and the climb to two zettahashes waits for a better price rather than a better chip.

Difficulty, the shock absorber

What keeps a stalled network running smoothly is the difficulty adjustment, Bitcoin’s automatic thermostat. Every 2,016 blocks, roughly every two weeks, the protocol measures how fast blocks were found and resets the target so the next stretch averages ten minutes. When hashrate falls, blocks slow, and the next adjustment cuts difficulty to speed them back up; when hashrate rises, difficulty climbs to slow them down. It is the mechanism that has let the network shed and regain a hundred exahashes over 2026 without any user noticing a difference in how fast their transactions confirmed.The adjustment has guardrails. However far blocks drift from the ten-minute target, a single retarget can move difficulty by at most a factor of four in either direction, a clamp that stops a sudden loss or influx of hashrate from whipsawing the schedule. In practice the moves are far smaller: a stretch where blocks arrive a little quickly nudges difficulty up a few percent, a slow stretch nudges it down. The elegant part is that no one decides any of this. There is no committee and no vote; the rule is arithmetic baked into the software every node runs, which is why the network kept a steady ten-minute heartbeat through every switch-off and switch-on of 2026.Nikolaos Panigirtzoglou, a managing director at JPMorgan, describes the loop in economic terms: when Bitcoin trades below its production cost, higher-cost miners power down, the hashrate declines, and difficulty adjusts lower. His team estimates a difficulty-to-price sensitivity, a beta, of about 0.62, meaning difficulty tends to move a little more than half as much as price over time. That self-correction is visible across 2026: ten cuts against eight increases, difficulty net lower on the year, and yet the network never once stalled or raced in a way an ordinary user would notice. The 19 September retarget, estimated near a 5% increase, is the thermostat leaning the other way for once, responding to a brief run of faster blocks rather than to any surge in new machines.

Does a flat hashrate weaken Bitcoin’s security?

A plateau in hashrate is not the same as a plateau in security, but the two are related, and the question is fair. Hashrate is the wall an attacker has to climb to rewrite the chain: to run a 51% attack you need to out-hash the honest network, which means either building or renting more machines than everyone else combined. A flat hashrate means that wall stopped getting taller, not that it fell.And the wall is tall. Campbell Harvey, a finance professor at Duke, originally estimated the cost of a week-long majority attack at about $6 billion in hardware, construction and power, roughly a quarter of a percent of Bitcoin’s network value, as reported by crypto.news. He has since revised the figure upward to around $8 billion, or about 50 basis points of the value of bitcoin, while noting the economics have shifted because an attacker could now pair the assault with a large short position; as he puts it, “the difference today is the derivatives markets.” Not everyone is worried. Matt Prusak, president of American Bitcoin, told reporters that “economic feasibility kills the 51% thesis,” arguing that quietly amassing that much hardware would take years and trip every alarm along the way. Either way, Bitcoin’s main chain has never suffered a successful 51% attack, unlike smaller proof-of-work coins such as Ethereum Classic and Bitcoin Gold, which have.The deeper worry is the security budget, the total miner revenue that funds all of that hashing, currently on the order of $13 billion a year. That budget is what a proof-of-work chain buys with its emissions, and comparing it against the very different cost structure of proof-of-stake networks is a live debate; we walked through the staking side of that ledger in our look at validator economics in 2026. For now, a flat hashrate at nearly a zettahash still buys more security than any other public network on earth.

The map moved even as the total stood still

One reason the flat aggregate is misleading is that the map underneath it kept shifting. Even as global hashrate stalled, capacity drained out of some jurisdictions and pooled in others. Hashrate Index’s most recent country heatmap put the United States at about 37% of the network, Russia near 17%, and China, despite its 2021 ban, back around 12% as miners quietly returned to Sichuan and Xinjiang. The newer entrants are where the growth is: Paraguay near 4% on cheap hydropower, the Gulf states of the United Arab Emirates and Oman around 3% each as deliberate sovereign industrial policy, and Ethiopia about 2.5% on the output of its giant new dam.There is an irony worth stating plainly. China banned domestic mining but still manufactures the large majority of the world’s ASICs through Bitmain, MicroBT and Canaan, so it profits as the arms dealer of an industry it outlawed at home. And the redistribution is not costless: concentration remains high, with the top three countries accounting for roughly two-thirds of the network and the top three mining pools for a similar share, a centralization that matters more for censorship resistance than for raw attack cost. A stall in the total, in other words, can hide a great deal of churn in who holds the machines and where.

Country	Share of network	Note
United States	~37%	Texas and Wyoming grids, policy tailwinds
Russia	~17%	Legal nationally, sanctioned internationally
China	~12%	Banned in 2021, quietly resurged
Paraguay	~4%	Hydropower, HIVE build-out
UAE / Oman	~3% each	Sovereign industrial policy
Ethiopia	~2.5%	GERD hydropower, fragile economics

The 2028 clock behind the plateau

Every stall has a deadline behind it. In the spring of 2028 the block subsidy halves again, from 3.125 BTC to 1.5625, cutting the coin-denominated revenue that funds hashing in half overnight. Historically, price has more than made up for each halving, but each cut sharpens the same long-run question: as the subsidy trends toward zero over the coming decades, transaction fees have to grow into the gap, or the security budget shrinks with it. Fees today are a rounding error, well under 1% of the block reward on a typical day.That is why the debate over what Bitcoin’s blocks should carry, and whether the base layer can generate meaningful fee demand without new functionality, is really a debate about who pays for security in the 2030s. It runs headlong into Bitcoin’s cultural resistance to change, a tension we explored in our piece on Bitcoin’s upgrade crisis. A stalled hashrate today is a manageable problem, self-correcting through difficulty and reversible the moment price or power frees up. A stalled fee market at the next-but-one halving would be a structural one, and no thermostat fixes that.

What US regulators say, and what to watch next

For US miners the regulatory picture is, unusually, a tailwind. The SEC’s Division of Corporation Finance stated in March 2025 that proof-of-work mining, whether solo or through a pool, does not involve the offer or sale of securities, lifting a cloud that had hung over the industry, per The Block. That sits alongside a broader push to treat Bitcoin as a commodity rather than a security, the same classification question that governs which products can list and trade, which we unpacked in our guide to the commodity-or-security gate. The federal government even holds a Strategic Bitcoin Reserve, established by executive order in March 2025, that it has pledged not to sell. Whatever else pressures the industry, in the United States at least the legal ground under mining has rarely been firmer.So what turns the stall back into growth? The honest answer is price. Watch Bitcoin against the roughly $80,000 line CoinShares flags as the threshold below which hashrate keeps sliding; watch the 19 September difficulty retarget for whether the recent uptick sticks; watch whether the AI leasing boom cools enough to free power back up for mining; and watch the hashprice, because the day a new rig pays for itself faster than an AI rack does, the curve will start climbing again. Until one of those changes, a year after the zettahash, Bitcoin’s hashrate is likely to keep doing the one thing it almost never used to do: nothing.

Frequently Asked Questions

Is Bitcoin’s hashrate going down in 2026?

Broadly, yes. After first crossing one zettahash per second in September 2025 and peaking near 1,162 EH/s that October, the seven-day average fell back below the zettahash in January 2026 and has hovered between about 915 and 940 EH/s through mid-September, lower than where it first reached the milestone a year earlier.

Why has Bitcoin’s hashrate stopped growing?

The main reason is competition for electricity from artificial intelligence. The cheap power and data-center sites that miners would have filled with new machines are now more valuable leased to AI and high-performance computing tenants, so listed miners are diverting capital and megawatts away from mining. Unlike past declines, no government ban or price crash forced it.

What was Bitcoin’s peak hashrate?

On smoothed data from Glassnode, Hashrate Index and CoinShares, the peak was about 1,162 EH/s, roughly 1.16 zettahashes, on 19 October 2025. Some public dashboards display a higher 1.44 ZH/s figure from September 2025, but that reflects a noisy single-interval spike rather than a level the network sustained.

Does a flat hashrate make Bitcoin less secure?

Not materially in the near term. A stable hashrate means the cost of a 51% attack stopped rising, not that it fell; estimates still put that cost in the billions of dollars, and Bitcoin’s main chain has never been successfully attacked. The longer-term concern is the security budget after future halvings, which depends on transaction fees growing as the block subsidy shrinks.

When is the next Bitcoin difficulty adjustment?

The next retarget is estimated for around 19 September 2026, with an expected increase of roughly 5%, which would be the largest upward move in months. Difficulty adjusts every 2,016 blocks, about every two weeks, to keep the average block time near ten minutes.Marcus Okafor covers Bitcoin mining and network infrastructure for HOGE Wire.

More from this desk

Taproot Was the Last Fork: Bitcoin’s Upgrade Crisis in 2026

Sep 15, 2026

Stacks vs Rootstock: Bitcoin’s Oldest DeFi Bet in 2026

Sep 14, 2026

Runes, Ordinals, and BTCfi: Bitcoin’s Asset Economy in 2026

Sep 13, 2026