Nobody Knows Bitcoin’s Real Hashrate: How It’s Estimated in 2026

On September 19, Bitcoin’s mining difficulty is on track for its biggest upward jump in months, an estimated increase of roughly 4 to 5 percent that would lift the number above 130 trillion for the first time since the summer. Headlines will translate that into a story about hashrate. Some will say Bitcoin’s computing power “surged.” Others, pointing at a plateau that has now lasted most of the year, will say it “stalled.” Both camps will state their figure to the nearest exahash. Almost none of them will mention that the number is a guess.

Bitcoin’s hashrate, the total rate at which miners around the world are racing to solve the next block, is the most-cited health metric of the largest proof-of-work network on earth, and it cannot be measured. There is no meter. Miners are anonymous, they report to no central registry, and no Bitcoin node counts the hashes being burned across the planet. Every hashrate figure you have ever read, including the ones in this publication, is an estimate reverse-engineered from two things the network actually publishes: the difficulty, and the timestamps on blocks.

That gap between the number’s authority and its real precision matters more in 2026 than it used to. This is the year hashrate stopped setting records, a plateau we covered in detail in our look at the hashrate stall a year after the first zettahash, and the year miners began quietly rerouting megawatts to artificial-intelligence tenants, which makes the estimate harder to read rather than easier. This is a guide to how the number is really produced, why reputable trackers can disagree about it by nearly a third on the same afternoon, how a phantom all-time high keeps circulating, and how to read the metric without being fooled.

There Is No Hashometer

Proof-of-work mining is a brute-force guessing game. Each machine repeatedly hashes a candidate block header, hoping to stumble on an output below the network’s target. The overwhelming majority of those attempts fail and vanish; only the winning hash, roughly one every ten minutes across the whole planet, ever becomes a block. A Bitcoin node sees the winners. It never sees the trillions upon trillions of losing guesses, and it is precisely those losing guesses that hashrate is trying to count.

Compounding the problem, the machines making those guesses are scattered across dozens of countries, sit inside private facilities, and belong to operators who have every incentive not to publish their exact capacity. Some hashrate runs through public mining pools that do disclose their share of blocks, but a meaningful slice is solo or private, and no one aggregates a trustworthy planet-wide total. As Luxor’s Hashrate Index puts it plainly, “It is not possible to pinpoint the exact hashrate of the Bitcoin network; all the bitcoin miners across the world do not publicly disclose this information, and there’s no way to directly gauge the metric using a Bitcoin node.” You can read their full explainer on measuring hashrate for the mechanics.

The cleanest analogy is a lottery. Imagine trying to work out how many tickets a city bought this week when the only things you can observe are how hard the jackpot was to hit and how quickly someone won it. You never see the losing tickets, only the winners and the difficulty of the draw. From those two facts you can make a very good statistical estimate of total ticket-buying, but you cannot count the tickets, and on any given week luck will throw your estimate off. That is exactly the position every hashrate tracker is in.

How the Number Is Actually Estimated

The estimate rests on the two public quantities. The first is difficulty, an exact integer written into consensus that encodes how much work, on average, a single block requires. The second is the pace of blocks, read off their timestamps. If you know the work per block and how fast blocks are arriving, you can back out the work per second, and work per second is what we call hashrate.

The standard formula is compact: network hashrate is roughly difficulty multiplied by 2^32, divided by 600. The 2^32 term (about 4.29 billion) is the average number of hashes needed to find one block at difficulty 1; multiplying by the current difficulty scales that up to the real target, and dividing by the 600-second target block interval converts work-per-block into work-per-second. At today’s difficulty of 127.45 trillion, that arithmetic produces about 912 exahashes per second. A second common form leans on observed blocks instead of assuming the target pace: take the blocks found over some window, divide by the number you would expect at ten minutes each, and multiply that ratio into the same difficulty term, a recipe spelled out on the Bitcoin wiki difficulty page.

Before going further it helps to fix the units, because Bitcoin’s hashrate has climbed through so many orders of magnitude that the vocabulary changes every few years. The network is now quoted in exahashes, and in 2025 it brushed the next rung up, the zettahash.

The catch is buried in the formula’s assumption. It treats blocks as if they arrive exactly on the ten-minute schedule difficulty is tuned for. Over a two-week retarget window that is close enough. Over a single day it is not, and that gap is where all the noise comes from.

Block Discovery Is a Coin Flip

Mining is a memoryless random search. Every hash has the same vanishingly small chance of winning, independent of every hash that came before it, so finding a block is like waiting for a rare coin-flip streak. The long-run average wait is ten minutes, but any individual block can land in twenty seconds or take a full hour, even when the true hashrate has not budged an inch. The network does not slow down or speed up between blocks; the dice simply fall unevenly.

Stretch that randomness over a short measurement window and the arithmetic misfires. In a 24-hour period the network expects about 144 blocks, but luck alone can deliver ten or fifteen percent more or fewer, and the naive formula reads that lucky streak as a ten to fifteen percent change in hashrate that never actually happened. The Bitcoin wiki’s warning about solo mining captures the underlying truth in one line: “Remember it’s just probability! There are no guarantees you will win every N days.” The same probability governs the aggregate.

You can watch it live. The most recent blocks this week have averaged about 9 minutes 35 seconds, comfortably faster than the ten-minute target, and a spot reading that trusts that pace will report a hashrate above the difficulty-implied baseline. A few slow blocks tomorrow would flip the sign. This is why serious analysts smooth the estimate over seven, fourteen, or thirty days rather than quoting the instantaneous value, and why Hashrate Index and every careful tracker treat the 24-hour figure as the least reliable of the lot.

The statistics have a name. Block discovery follows a Poisson process, the same family of math that describes radioactive decay or calls arriving at a switchboard: events that are individually random but collectively predictable over a long enough horizon. Across thousands of blocks the average pins down tightly, which is why the two-week difficulty window is dependable; across the 144 blocks of a single day the spread is wide, which is why a spot reading is not. Widening the window does not shrink the error in proportion, it shrinks it by roughly the square root, so the step from one day to seven days of smoothing removes most of the noise, and the further step to thirty days clears most of what remains.

The Same Network, Several Different Numbers

Here is what the major trackers said Bitcoin’s hashrate was on the afternoon this piece was written. They are looking at the same blockchain, running variations of the same formula, and they do not agree.

Three of those four readings cluster near 912 to 915 exahashes today, which is a happy accident: the spot figure has drifted back toward the smoothed trend, so everything lines up. That is not always the case. Over the past month the same CoinWarz spot reading has printed values hundreds of exahashes apart, swinging from the low 800s of EH/s up past 1 zettahash on the noisiest days, while the difficulty-implied baseline barely moved. The divergence is not a flaw in any one site. It is four different answers to four slightly different questions: what is the rate right now, over the last week, over the last month, and over the current retarget epoch.

The disagreement even reaches the forecasts. For the September 19 retarget, CoinWarz projected roughly a 4.3 percent difficulty increase while mempool.space showed something closer to 4.6 or 4.7 percent, and mempool’s own estimate shifted between page loads minutes apart. These are estimates of an estimate, and they will only converge as the epoch’s final blocks land.

None of this is a reason to distrust the trackers; it is a reason to read them correctly. A trading desk watching for a large miner going offline wants the spot feed precisely because it reacts within hours. A researcher charting the security trend wants the thirty-day line precisely because it ignores a noisy afternoon. The error is not that the numbers differ, it is quoting one window as though it answered every question. When a story pins a single figure to three decimal places and calls it the hashrate, that is the tell that the window was never specified in the first place.

A Case Study in a Bad Number: The Phantom 1.44 Zettahash

One row in that table is simply wrong, and it is worth dwelling on because it refuses to die. CoinWarz’s hashrate page has for months carried an all-time high of 1.44 ZH/s dated September 20, 2025, and the figure surfaced again on the chart the day this was written, presenting the current network as roughly 63 percent of that supposed record. Search engines and automated summaries dutifully repeat it. It is a ghost.

No serious source corroborates 1.44 ZH/s. The best-documented true peak is about 1,162 EH/s, or 1.16 zettahashes, reached on October 19, 2025, per data reported by yellow.com. The seven-day average first held above the zettahash line in early September 2025, according to CoinDesk, and it fell back below 1 ZH/s on January 17, 2026, at 988 EH/s. Nothing in that record approaches 1.44.

So how does a 1.44 appear at all? A single anomalously fast run of blocks, a cluster of luck lasting an hour or two, can send the instantaneous formula spiking far above the real rate. If a tracker logs that momentary spike as a discrete daily high and never revises it, the spike calcifies into an enshrined all-time high, even though the seven-day and thirty-day averages never came within a quarter of it. The lesson is not that one website is careless; it is that any hashrate figure quoted to three significant figures, and above all any single-day record, deserves a skeptical second look. Kaan Farahani, a research associate at Luxor Technology, made the point about difficulty forecasts in a Hashrate Index roundup, and it applies word for word to hashrate: “Early on, difficulty predictions are shaky because of short-term variance. As time (or blocks) pass, the noise fades and the signal sharpens.” The shorter the window, the more noise. That is why this publication quotes current hashrate as a range, never a point.

Difficulty Is the Only Hard Number

If hashrate is a guess, difficulty is a fact. It is a single value written into every block header, enforced identically by every node, with no room for interpretation. It changes on a fixed schedule: every 2,016 blocks, roughly two weeks, the protocol compares how long those blocks actually took against the 20,160-minute target and rescales difficulty by that ratio, bounded by a maximum 4x rise or 0.25x fall that has never once been hit.

This is why the anchor row in the divergence table is the difficulty-implied hashrate. It is the rate that would produce exactly the block pace difficulty is tuned to expect, which makes it the market’s single most defensible number for the network’s sustained hashing power. Everything else is a shorter-window read on the same underlying quantity. When you want one number you can stand behind, translate difficulty into hashes and stop there.

The upcoming retarget shows the mechanism at work. Blocks have run a touch fast this epoch, so difficulty is set to rise, with CoinWarz and mempool both projecting somewhere between 4.3 and 4.7 percent on September 19, the largest upward move in months and enough to push difficulty near 133 trillion. Even that would leave it about 15 percent below the record near 156 trillion set in November 2025. What makes 2026 unusual is how two-sided the year has been.

Through early September the network logged eight increases and ten decreases, a net decline of roughly 12 percent on the year, as tracked by Bitcoin.com News. That kind of sawing back and forth only appears when the network is not growing, and it is the difficulty record, not any hashrate chart, that tells the story most cleanly.

Reading the Windows: Spot, Seven-Day, Thirty-Day

Because there is no single true hashrate to report, choosing a window is really choosing a question. Each one is right for something and wrong for the rest.

• Spot, or 24-hour. Answers “what is the rate this moment?” The most responsive figure and the noisiest, which makes it the worst basis for a headline. It is genuinely useful for catching a real, sudden event, such as a large facility tripping offline, and close to useless for judging a trend.
• Seven-day. The workhorse. It smooths out block-time luck while still reacting to a genuine multi-day shift, and it is the figure most analysts and this publication cite by default.
• Thirty-day. The structural view, best for the question “is the network growing, flat, or shrinking?” It is slow to react, so it can lag a real turn by weeks, which is a feature when you want signal over noise.
• Difficulty-implied, or per-epoch. The cleanest anchor of all. It updates only every two weeks and is consistent with difficulty by construction, so it cannot be fooled by a lucky hour.

A simple rule ties them together. When two windows disagree sharply, either the network is in transition or one of the numbers is a spot artifact; when they converge, as they roughly do today near 910 to 915 EH/s, the reading is trustworthy. The convergence itself is the confidence signal.

What the Estimate Says in September 2026

Put all the windows together and the honest read is that Bitcoin’s hashrate is somewhere around 910 to 940 exahashes per second on a smoothed basis, below the roughly 1,162 EH/s peak of October 2025 and hovering near, but mostly just under, the 1 zettahash line it first crossed a year ago. The plateau is real in the narrow sense that every window agrees the network has not grown this year, which is itself unusual, but “flat” is a statement about the smoothed estimate and nothing more; the number of machines and the megawatts behind them are separate questions the hashrate figure cannot answer.

Jamie Redman of Bitcoin.com News framed the stall in terms of supply constraints rather than economics: “The flat hashrate indicates that miners are evolving very carefully, and they might be facing hardware limits.” The price backdrop reinforces the caution. Bitcoin trades near $76,300, down about 40 percent from its October 2025 record and softer on the week after this week’s Federal Reserve rate hike. Because mining revenue tracks the coin price, a soft market keeps a lid on new deployment, even as hashprice, miners’ revenue per unit of hashrate, has climbed to $39.63 per petahash per day, up about 22 percent on the month as August’s price recovery filtered through.

A Year of Milestones, All of Them Estimates

The growth curve everyone cites, the one that runs from a hobbyist’s laptop to more than a zettahash, is really a chain of estimates crossing round numbers. That does not make the milestones meaningless, but it does mean each one should be read as a threshold on a smoothed average rather than a precise measurement.

The reason the honest milestones are phrased as “first held above” on a seven-day basis, rather than “first touched,” is the same variance problem that spawned the phantom 1.44: a spot reading brushes a level long before the trend does. The Q1 2026 turn was itself a landmark, the first quarterly hashrate decline in six years per CoinDesk, and even that was a claim about a smoothed average. The leap from 100 EH/s to more than 1 ZH/s in under six years is a tenfold rise in estimated hashes; the machines producing them changed by even more.

It is worth pausing on how thin some of these records are. The gap between 988 EH/s in January and the roughly 1,162 EH/s peak the previous October is about 15 percent, well inside the range a few unlucky weeks of block timing plus a genuine round of miner capitulation can produce, and the entire zettahash era, from the first sustained crossing to the first drop back below the line, lasted barely four months. Round numbers make tidy headlines, but the network does not respect them; it drifts across and back over thresholds that exist only in our base-ten notation.

Efficiency Hides Inside the Number

Here is the caveat that trips up even seasoned readers: hashrate counts hashes, not machines and not watts. Because ASIC efficiency has improved roughly tenfold in a decade, the same hashrate today draws far less power, and comes from far fewer and far more powerful machines, than it did even three years ago.

The efficiency figures come from manufacturer specifications and industry reporting, including Bitmain’s own documentation and coverage of the S23 series, the first to break below 10 joules per terahash. The implication for measurement is subtle but important. A flat hashrate does not mean flat energy consumption, which can fall as old rigs are swapped for efficient ones, and it certainly does not mean a flat number of machines. So the metric that headlines treat as a proxy for how much is being spent to secure Bitcoin is only loosely coupled to megawatts and to dollars. When you read that hashrate is flat, the accurate translation is that the estimated aggregate hashing rate is flat, and nothing more.

Does the Fuzziness Break the Security Case?

A fair objection follows from all of this. If we cannot measure hashrate to better than plus or minus a third on a bad day, is the entire claim that Bitcoin is secured by so-and-so many exahashes built on sand? The answer is no, and the reason is that the security argument depends on orders of magnitude, not decimal places.

Whether the network is running at 912 or 940 or a clean 1,000 EH/s, the cost of assembling a majority of it is measured in billions of dollars, and that conclusion is robust to the estimate’s error bars. Campbell Harvey, a finance professor at Duke University’s Fuqua School of Business, has modeled a majority-hashrate attack at roughly $8 billion, which he framed, as reported by Bitcoin.com News, as “about 50 basis points of the value of bitcoin.” His sharper modern point is that “The difference today is the derivatives markets,” because an attacker could pair the assault with a large short position and profit from the very price crash the attack would cause. Whether the underlying hashrate estimate is off by three percent does not move that thesis.

The same holds for the security budget, the roughly $13 billion a year in miner revenue implied by multiplying hashprice by hashrate. It is an order-of-magnitude figure, and a small measurement error leaves the argument intact. This is where proof of work and proof of stake diverge in kind rather than degree, a contrast we drew out in our breakdown of validator economics: Bitcoin’s cost to attack is an external, physical spend on hardware and power, while a staking network’s is bonded capital inside the system. The fuzziness of the hashrate estimate is a problem for headline precision and for traders chasing spot spikes, not for the core security claim.

The AI Twist: Energized, Operating, or Actually Hashing?

2026 has added a fresh layer of ambiguity. Public miners are diverting power and capital to artificial-intelligence and high-performance-computing tenants, and a megawatt contracted to an AI customer is a megawatt not making hashes. That widens the gap between three capacity numbers that headlines routinely conflate.

• Energized capacity is the hashrate a fleet could produce if every installed rig ran flat out.
• Operating capacity is what is actually powered on right now.
• Realized hashrate is what actually reached the network, inferred from the blocks a miner won.

These can differ enormously. A company can report tens of exahashes of energized capacity while curtailing rigs for grid-services payments or repurposing entire halls for AI compute, so its self-reported terahashes and the hashrate that actually hit the chain are not the same thing. The network-level estimate, which only ever sees blocks, is the honest arbiter; a single firm’s headline EH/s should be read with care.

This reallocation is the reason hashrate has stalled even as hashprice jumped about 22 percent on the month: the marginal megawatt went to AI rather than to new rigs. CoinShares has documented more than $70 billion in cumulative AI and HPC contracts across the listed mining sector and forecast that some public miners could draw up to 70 percent of revenue from AI by year-end, a shift we explored alongside the broader race for verifiable compute in our coverage of AI-focused blockchains. The net effect on measurement is that the number is getting harder to interpret, because a flat hashrate now conceals an active reallocation of the very megawatts that used to grow it.

What It Means for Regulation and Reporting

Even a fuzzy number carries legal and disclosure weight. Listed miners report hashrate to investors, and the definitional slack between energized, operating, and realized capacity is exactly where a disclosure can mislead. The softness of the metric is therefore an investor-protection issue, not merely a data-nerd curiosity, and a reader evaluating a mining stock should interrogate a company’s headline EH/s the way this piece interrogates the network figure: ask which definition, over which window.

On the activity itself, the picture is clearer. 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, on the reasoning that miners rely on their own efforts rather than those of a third party, a position reported by The Block. That keeps mining outside the securities perimeter, in contrast to the murkier treatment of staking and the broader commodity-or-security debate we traced in our piece on the commodity-security gate. But listed miners remain SEC registrants for their equity, so how they define and disclose hashrate, and the AI pivot layered on top of it, is squarely a disclosure question the estimate’s ambiguity makes harder to police.

How to Read the Hashrate Without Getting Fooled

None of this means the number is useless. It means it is a disciplined guess, and difficulty keeps it honest. A short checklist covers most of the traps this piece has walked through.

• Prefer a seven-day or thirty-day average over a spot or 24-hour reading, and treat single-day figures as noise until a trend confirms them.
• Cross-check at least two independent trackers. If they diverge by more than a few percent, the network is either in transition or one number is a spot artifact.
• Anchor on difficulty. The difficulty-implied rate is the one figure nobody can fudge, and it updates every two weeks like clockwork.
• Distrust round-number records and any all-time high pinned to a single day. The real peak was a seven-day crossing, not a momentary spike.
• Quote current hashrate as a range, not a point. “About 910 to 940 EH/s” is more honest than a false-precision 912.
• For a company’s hashrate, always ask whether the figure is energized, operating, or realized, and over what period.

When the September 19 retarget prints its rise of more than four percent and the headlines announce that Bitcoin’s hashrate has surged, you will know what they are really reporting. Not a measurement, but the network’s own two-week verdict on how fast the blocks have been coming. That verdict is trustworthy, precise, and consensus-enforced. The exahash figure stapled next to it is an educated estimate, and knowing the difference is most of what it takes to read Bitcoin’s mining data like a professional.

Frequently Asked Questions

Can Bitcoin’s hashrate be measured exactly?

No. Bitcoin miners are anonymous, they do not report their output, and no Bitcoin node counts the hashes being tried worldwide. Every hashrate figure is an estimate reverse-engineered from the network difficulty and how fast blocks are being found.

How is Bitcoin’s hashrate calculated?

The standard estimate multiplies the network difficulty by 2^32 (about 4.29 billion, the average number of hashes per block at difficulty 1) and divides by the 600-second target block time. At a difficulty of 127.45 trillion that yields roughly 912 exahashes per second.

Why do different websites show different Bitcoin hashrate numbers?

Because they use different averaging windows. A spot or 24-hour reading is noisy due to random block-time variance, a 7-day average is smoother, and a 30-day average is smoother still, so the same network can show readings that differ by tens of percent on the same day.

What is Bitcoin’s hashrate right now in September 2026?

On a smoothed basis it is roughly 910 to 940 exahashes per second, below the record of about 1,162 EH/s set on October 19, 2025, and hovering just under the 1 zettahash line it first crossed in September 2025.

Is the 1.44 zettahash all-time high real?

No. That figure, dated September 20, 2025, is a data artifact produced by a momentary spot spike and is not corroborated by smoothed data. The true peak was about 1,162 EH/s, or 1.16 ZH/s, on October 19, 2025.

Marcus Okafor covers Bitcoin mining, network data, and market structure for HOGE Wire.