Bitcoin Mining Pools in 2026: The Leaderboard Blind Spot

Open any Bitcoin mining dashboard and you get a tidy picture: about a dozen pools, ranked by the share of blocks each one finds. As of late September 2026, Foundry USA sits on top with roughly 24 percent of the network, AntPool is closing fast at about 22 percent, and F2Pool holds third near 15 percent, out of an estimated 952 EH/s of total hashrate, according to mempool.space. It looks like a competitive market with ten independent players. The trouble is that a pool name is a brand, not a person, and counting logos is not the same as counting the operators who actually decide what goes inside a Bitcoin block. If you want to know who really secures the network, the leaderboard is where the answer starts, not where it ends.

What a mining pool actually is, and why one exists

Mining is a lottery, and the tickets are cheap only in bulk. A single modern ASIC such as a Bitmain S21, running at roughly 200 TH/s, is a rounding error against a network pushing past 950 EH/s. On its own it would expect to find a block once every few decades. Difficulty, sitting at about 132.76T after the latest adjustment (CoinWarz), keeps the average block time near ten minutes no matter how much hardware joins, so more competition simply makes any one machine’s odds longer.

A pool solves that by pooling. Thousands of miners point their machines at a single coordinator, the coordinator combines their work into one giant lottery entry, and when the pool wins a block it splits the reward among contributors in proportion to the work they submitted. Instead of a life-changing payout once a generation, a miner gets a steady drip that tracks its share of the pool. That is the service everyone talks about: variance smoothing, paid for with a fee.

How the split works is its own small industry. Under pay-per-share and its variants (PPS, FPPS, PPS+), the pool pays a fixed rate for every unit of valid work a miner submits, absorbing the luck of whether a block actually lands; the operator is underwriting variance, and the fee is the premium. Under pay-per-last-N-shares (PPLNS), a miner is paid only when the pool finds a block, so the variance sits with the miner but the fee is lower and pool-hopping is discouraged. Fees typically run from zero to about four percent. None of that changes the control question, but it explains why miners cluster into a handful of large, well-capitalized pools: only a big operator can smooth payouts reliably, and that smoothing is exactly what concentrates hashrate.

There is a second service that gets far less attention and matters far more here. Under Stratum V1, the mining protocol almost every pool still runs, the pool does not just collect and pay; it builds the block template, the ordered list of transactions the miners then hash. When you join a V1 pool, you hand it the right to decide what your hashrate confirms. The full block subsidy is 3.125 BTC until the 2028 halving, worth about $264,000 per block at a BTC price near $84,600 (CoinDesk), plus fees, and there are 144 blocks a day. Whoever builds those templates is steering a large, valuable machine.

The leaderboard, as the world reads it

Here is the picture the dashboards paint, using a rolling one-week sample of the blocks found (mempool.space, late September 2026):

Read straight down, it is a familiar story. The top three (Foundry, AntPool and F2Pool) command roughly 61 percent of blocks between them; the top four clear 70 percent. The usual shorthand for this is the Nakamoto coefficient, the smallest number of entities that together control more than half the hashrate. For Bitcoin mining that number is 3: you need Foundry, AntPool and F2Pool acting together to cross 50 percent, a point TFTC made in 2026 when it noted that three pool operators, acting together, can control the majority of Bitcoin’s block space today. Three is already an uncomfortable number for a system that sells itself as decentralized. But three assumes every name in the table is its own master. That is the assumption worth testing.

A word on how those numbers are built. Nobody can see hashrate directly; what is visible is blocks, each stamped with a coinbase tag that names the pool that found it. Trackers such as mempool.space and Hashrate Index count those tags over a rolling window (a day, a week) and infer each pool’s share from its block count. That means the figures wobble from sample to sample, a pool on a lucky streak looks bigger than it really is, and any operator that chooses not to tag its blocks, or tags them inconsistently, muddies the picture. The leaderboard is an estimate assembled from the chain, not a registry anyone maintains, which is part of why the deeper question of who controls what is so easy to lose.

Why counting logos overstates decentralization

A pool brand is a front end. Behind it sits stratum server infrastructure, a payout ledger and a template builder, and none of those has to be unique to the logo on the website. Two mechanisms quietly shrink the real number of independent operators below the number of names.

The first is the proxy or white-label pool. A small pool can rent another operator’s stratum endpoint and resell it under its own name, or route its members’ work to a larger pool’s servers while keeping its own branding and support desk. To the miner it looks like a distinct service; to the network it is more hashrate flowing to the same template builder. White-labeling is a normal business arrangement in plenty of industries. In mining it has a side effect: it launders concentration into the appearance of choice.

The second is shared block templates. Even without a formal ownership link, if two pools consistently produce near-identical templates, block after block, they are making the same transaction-selection decisions. The independence the leaderboard implies, ten operators each choosing what to confirm, simply is not there. And unlike ownership, which hides behind corporate structures, template similarity is measurable on-chain by anyone willing to do the work. Someone has.

The evidence: pools that build the same blocks

0xB10C, an independent Bitcoin developer who monitors mining decentralization, compared the block templates different pools were broadcasting and found that several of them were not just similar but nearly identical. In an analysis published in September 2024, the overlap looked like this:

The method behind the finding is worth understanding, because it is what turns the claim from a hunch into a measurement. Every ten minutes or so, each pool broadcasts the template it wants its miners to work on. Two pools drawing from the same mempool will naturally overlap a lot, high-fee transactions are attractive to everyone, so a baseline similarity of, say, 90 percent is unremarkable. What is not normal is agreement in the high 90s sustained block after block, including the fiddly choices, the ordering and which marginal transactions to include or drop, where genuinely independent operators would diverge. That is the signal 0xB10C isolated, and it is why the pairs at 98 and 99 percent stand out from the crowd.

Templates that agree 98 or 99 percent of the time are not being built independently. In all, 0xB10C counted nine interconnected pools that together accounted for 37.6 percent of network hashrate over the measured window, with the AntPool, BTC.com and Poolin cluster alone making up about 26 percent. He was careful about what that does and does not prove, writing that although the templates “might be unusually similar” and some pools “might be engaging as proxy pools for others here and there,” it is “not proven that there is a single entity behind these nine pools.” The measurement is hard and the interpretation is contested. But the raw fact stands: the number of distinct template-builders is smaller than the number of logos.

Meet AntPool and friends

The template-similarity finding got a stranger sequel. In a follow-up published in March 2025, 0xB10C documented a series of incidents in which a group of pools, AntPool, Poolin, Binance Pool, Ultimus, CloverPool, Rawpool and, in some cases, Braiins, all broadcast the same invalid mining jobs during block races: empty jobs whose coinbase output tried to pay out more than the block subsidy allows. The pattern looked like cached coinbase values from a previous block being reused across several pools at once.

0xB10C read the glitch as a coinbase-code bug rather than an attack, noting the behavior did not look like a selfish-mining attempt. But identical bugs firing in lockstep across supposedly separate companies say something about the plumbing underneath. He agreed with a fellow developer, boerst, that the shared error “once more confirms that these pools are operated by the same entity,” and proposed a name for the group that has stuck in mining circles: “AntPool and friends.”

The ownership anchor for that cluster is not a mystery. AntPool is owned by Bitmain, the company that builds most of the world’s mining hardware, which hands one firm both the dominant ASIC supply line and a pool that shows up near the top of every leaderboard. That vertical reach is the same story driving the Bitmain versus MicroBT fight over who makes Bitcoin’s miners. And the invalid-coinbase incident is a useful reminder that Bitcoin’s consensus code carries real, live edge cases, the kind of low-level plumbing that also sits at the center of the fight to fix Bitcoin’s one genuine difficulty bug. None of this is proof of a single hidden owner; 0xB10C is consistent about that. It is proof that the leaderboard’s tidy rows overstate how many hands are really on the wheel.

There is a structural reason a manufacturer-linked cluster is sticky. When the firm that sells you the ASIC also runs a pool the machine ships ready to join, the path of least resistance for a new miner is to plug in and point at the default. Multiply that across years of hardware sales and the cluster accretes hashrate almost passively, without having to win anyone over on fees or features. That is not a conspiracy; it is a distribution advantage. But a distribution advantage that quietly routes a quarter of the network’s block-building through one set of servers is exactly the kind of concentration a leaderboard of separate logos is built to obscure.

Recount the leaderboard, this time by operator

Now redo the arithmetic. Take the live shares and, instead of reading them as separate companies, group the AntPool-correlated names together. The tight core, AntPool plus BTC.com plus Poolin, the three pools whose templates matched 98 to 99 percent, comes to about 23.3 percent of blocks. Fold in Binance Pool and Ultimus, both of which turned up in the shared-infrastructure incidents, and the cluster reaches roughly 25 percent.

On the broad reading, the ranking flips: AntPool and friends, not Foundry, becomes the single largest decider of what goes into Bitcoin blocks, and the effective Nakamoto coefficient drops from 3 to 2. Two entities, a Bitmain-linked cluster and a US corporate pool, would together control close to half of all block space. That is a materially worse picture than the leaderboard’s headline, and it does not require anyone to have lied; it only requires the smaller logos to not be as independent as they look.

The honest caveats matter. Where you draw the cluster boundary changes the answer; Braiins in particular is widely regarded as an independent, decentralization-minded operator despite showing up in the invalid-jobs data, and 0xB10C has never claimed courtroom proof of single ownership. Drop every disputed name and you are still left with a template-builder count in the low single digits. The leaderboard says ten-ish. The block chain says fewer.

Foundry, DCG and the other kind of concentration

If the AntPool cluster is concentration hiding in plain sight, Foundry is concentration that never hid at all. It is a subsidiary of Digital Currency Group, based in the United States, and it climbed to the top of the leaderboard the old-fashioned way: by being nearly free. Foundry ran without fees from 2019 into 2023 and still charges effectively zero on its full-pay-per-share tier for large institutional miners, treating the pool as a loss-leader monetized across DCG’s wider stack of mining-services, financing and data businesses.

That model tracked the institutionalization of Bitcoin mining. As publicly traded miners scaled and as spot exposure went mainstream through the wave of crypto ETF approvals mapped across 2026, hashrate concentrated with the operators that could offer compliant, US-based, zero-fee service at institutional scale. Foundry won that race. The result is a different flavor of centralization risk from the AntPool cluster: not a hidden web of correlated logos, but a single, named, US-jurisdiction corporate entity sitting on roughly a quarter of the network.

Put the two poles side by side and Bitcoin’s block production looks like a rough United States versus China axis, a Bitmain-linked cluster on one side and a DCG-owned pool on the other, the same geopolitical fault line that runs through the hardware supply chain. Neither pole is a villain. Both are single points where legal, commercial or political pressure could reach a large slice of the network through one door.

What a dominant operator can and cannot do

It helps to separate the horror movie from the real risk. A pool with a majority of hashrate cannot steal coins out of wallets, forge anyone’s signature or mint more bitcoin than the schedule allows; the cryptography protecting balances does not care how much hashrate you have, as Chainlink’s primer on the subject lays out. What majority hashrate buys is power over recent history: the ability to reorder or reverse the last few blocks (a double-spend against a counterparty such as an exchange) and to orphan rival blocks.

That kind of attack is real but self-defeating for anyone with money in the game: it burns the attacker’s own capital and sends miners fleeing, which is why serious analysts treat a Foundry or AntPool double-spend as economically irrational rather than technically impossible. The cheaper, quieter power is the one that does not need 51 percent at all: transaction selection. Under Stratum V1, the operator picks the contents of every block its miners find. A pool with 24 percent of hashrate is not casting a 24 percent vote on what Bitcoin confirms; it is unilaterally deciding the contents of roughly a quarter of all blocks. Cluster the correlated pools and one operator may be deciding a quarter to a third of them.

The scale of what would have to be burned is the real deterrent. Bitcoin pays out 450 new BTC a day in block subsidies (3.125 BTC across 144 blocks), worth on the order of $38 million a day at recent prices, and an attacker would need enough hardware and power to out-hash everyone else while forfeiting that revenue and torching the value of any coins and rigs it holds. That is why the credible threat from concentration is not a spectacular reorg; it is the quiet, cheap, deniable act of leaving certain transactions out.

The censorship precedent is not hypothetical

Bitcoin already has a worked example. In November 2023, F2Pool was caught filtering transactions from addresses on the US OFAC sanctions list; the missing transactions first surfaced through 0xB10C’s monitoring. F2Pool co-founder Chun Wang initially defended the practice, then reversed course under community pressure, arguing that a censorship-resistant system “must be designed to resist censorship at the protocol level, rather than relying on each participant to act conscientiously” and disabling the filter until the community reached a broader consensus. 0xB10C has since documented further stretches of missing sanctioned transactions.

The pressure is not purely theoretical. Back in 2021, the US miner Marathon briefly ran an avowedly OFAC-compliant pool that filtered sanctioned transactions, then abandoned it within a month after an outcry. The episode showed both that a public, US-listed operator will feel the pull of compliance and that community backlash can still reverse it, at least for now. The worry for the years ahead is that the pull gets stronger as more hashrate sits inside US-jurisdiction corporate pools, precisely the Foundry end of the concentration picture, where a subpoena or a sanctions designation lands on a named company with lawyers and shareholders rather than on an anonymous global collective.

Line that up with the concentration math and the stakes get concrete. If a single operator, whether Foundry alone or AntPool and friends, decides the contents of a quarter of blocks, then a compliance order served on that one company reaches a quarter of Bitcoin’s throughput without touching a line of protocol code. Concentration is what turns censorship from an irritant a user can route around into a lever a regulator can pull. The 2023 episode was resolved by one founder’s choice. The structural worry is that the choice exists at all, and that it sits with fewer hands every quarter.

The fix is to stop trusting the pool with the template

The durable answer is not to hope pools behave. It is to take away the power that makes their behavior matter: block construction. Two efforts do exactly that. Stratum V2, a full rewrite of the mining protocol, includes a mode called Job Declaration in which the miner, running its own Bitcoin node, builds its own block template and the pool is demoted to smoothing payouts and validating the coinbase. The Spark research breakdown describes a binary protocol roughly 70 percent smaller than V1’s plaintext JSON, with an encrypted miner-to-pool link that also closes off the hashrate-hijacking attacks V1 allowed.

Ocean’s DATUM takes the same idea and trades elegance for deployability: the miner runs a node and a gateway and builds templates locally, layered on top of existing V1 firmware so no ASIC upgrade is needed. Either way, the significance for this story is precise. If miners build their own templates, then two pools sharing a brand, an owner or a set of servers no longer means their blocks share contents. Job Declaration severs the link between pool-name concentration and block-content control, which is the exact link this whole piece is about.

The people building it say so plainly. Ocean co-founder and Bitcoin Core developer Luke Dashjr described the payoff on X: “Instead of two blocks made by the same pool, we get blocks made by two independent miners, exactly the same as if they weren’t using any pool at all.” When GoMining and the V2-native pool DMND mined the first production Job Declaration block in June 2026, DMND chief executive Alejandro De La Torre called it proof that “a miner just mined the first Stratum V2 block to power their own product end to end,” while GoMining’s Mark Zalan put the stakes in one line: “For years, mining pools have determined which transactions are included in Bitcoin blocks.”

Where deployment actually stands in 2026

The gap between the promise and the reality is the part worth watching. On 7 May 2026, seven pools representing roughly 75 percent of hashrate, Foundry, AntPool, F2Pool, SpiderPool, MARA, Block Inc and DMND, joined a Stratum V2 working group. AntPool chief executive Andy Zhou said the company was “proud to support the broader adoption of Stratum V2” as an open standard for efficiency, security and decentralization. The first production Job Declaration block followed on 25 June 2026.

But joining a working group is not the same as handing miners the template. As of mid-2026, only Braiins and DMND actually ran Job Declaration in production; the giants signed up to test and plan, not to deploy. Which means the pools whose concentration matters most, the ones at the top of both the logo leaderboard and the operator recount, are precisely the ones that have not yet given up template control. The reference implementation team projects that V2 could ship as the default firmware on new ASICs by the end of 2026, reaching perhaps 40 to 60 percent of hashrate, but that is a projection, not a deployment. Until the big operators actually turn Job Declaration on, the blind spot stays exactly where it is.

The regulator’s view, and the staking mirror

For US miners, one question is now settled. In March 2025 the SEC’s Division of Corporation Finance said that proof-of-work mining, including both solo mining and participation in a mining pool, is not a securities transaction, because the rewards come from a miner’s own computational work rather than the entrepreneurial efforts of others, as The Block reported. A broader interpretive release in March 2026 extended the no-securities view to mining, staking, wrapping and airdrops; SEC Chair Paul Atkins framed it as regulators doing their job, to “draw clear lines in clear terms.” Mining derivatives such as hashrate futures fall to the CFTC, and sanctions compliance is a Treasury and OFAC matter, so pool concentration is a market-structure and censorship problem, not a Howey problem.

The same shape shows up one chain over. Ethereum stopped mining at the 2022 Merge, so its “pools” are staking pools now, and they concentrate the same way: one operator coordinating many delegators. Lido alone accounts for roughly 23 percent of all staked ETH and about 62 percent of the liquid-staking segment, according to Datawallet’s staking data. The crucial difference is the safety net. Proof-of-stake has slashing, a protocol penalty that lets the network burn a misbehaving operator’s capital directly. Proof-of-work has no such lever; a pool that censors or double-spends can only be punished by miners leaving and users losing trust. Bitcoin’s discipline is social, which is exactly why the count of independent operators matters so much.

So what does this mean if you mine?

For an individual miner, the takeaway is not to panic about the leaderboard but to treat pool choice as a governance decision, not just a fee comparison. Pointing your hashrate at the cheapest large pool is rational for your wallet and quietly corrosive for the network, because it feeds the same template-builders that already decide most blocks. Spreading hashrate across smaller independent pools, or choosing one that supports miner-built templates, costs a little in convenience and sometimes in fees, and buys back a little decentralization.

The more durable move is to insist on template control. A miner running Ocean with DATUM, or pointed at a pool that has enabled Stratum V2 Job Declaration, is building its own blocks regardless of the logo on the payout. That is the one choice that breaks the link this article is about: it makes your slice of hashrate count as an independent vote on Bitcoin’s contents rather than as more weight behind someone else’s decisions. Until the largest pools switch that on by default, the responsibility sits with the miners who bother to ask.

What to watch next

Three things will tell you which way this is heading. The first is the Foundry-AntPool gap. AntPool at about 22 percent and rising has pulled within roughly two points of Foundry, the closest the top two have been in a long while; if AntPool, or AntPool and friends on the operator recount, takes the top spot outright, expect the centralization debate to reignite in earnest. The second is deployment: whether the big pools actually switch on Job Declaration or keep it in the working-group waiting room. The third is whether 0xB10C’s clusters tighten or loosen, that is, whether more pools start sharing templates or the correlated ones diverge.

The larger lesson is about which number to trust. The logo count is marketing. The headline Nakamoto coefficient is better, but it still takes the logos at face value. The metric that actually captures Bitcoin’s block-production risk is the number of independent template-builders, and on today’s evidence that number is smaller, maybe meaningfully smaller, than the ten-odd names the dashboard shows. Until miners hold their own templates, the leaderboard will keep flattering the network it measures.

Frequently Asked Questions

How many Bitcoin mining pools are there, and which is the biggest?

Dashboards list about a dozen meaningful pools. As of late September 2026, Foundry USA leads with roughly 24 percent of blocks and AntPool is close behind near 22 percent, with F2Pool third at about 15 percent (mempool.space). Those are brand-level shares, though; research suggests several smaller names may share infrastructure with AntPool, which would make the real leader closer than the leaderboard shows.

What is the Nakamoto coefficient for Bitcoin mining?

It is the smallest number of entities that together control more than 50 percent of hashrate. By the public leaderboard it is 3 (Foundry, AntPool and F2Pool). If you cluster the pools that appear to share block templates, the effective coefficient could be as low as 2, meaning fewer independent operators than the raw ranking implies.

Can a mining pool censor or reverse Bitcoin transactions?

A pool with a large share can reorder recent blocks or, far more cheaply, refuse to include specific transactions. It cannot steal coins from wallets or forge signatures. F2Pool was caught filtering OFAC-sanctioned transactions in 2023, which remains the clearest real-world example of pool-level censorship.

What is Stratum V2, and does it fix pool concentration?

Stratum V2 is a rewrite of the mining protocol whose Job Declaration mode lets individual miners build their own block templates, so the pool only smooths payouts. It does not change who holds hashrate, but it breaks the link between a pool’s size and its control over block contents. As of 2026 only a few pools run it in production.

Is joining a Bitcoin mining pool regulated by the SEC?

No. The SEC’s Division of Corporation Finance said in March 2025 that proof-of-work mining, including pool participation, is not a securities transaction, and a March 2026 release reaffirmed that view. Mining-related derivatives fall to the CFTC, and sanctions compliance is a Treasury and OFAC matter, not the SEC.

Yuki Tanaka covers Bitcoin mining, protocol politics and market structure for HOGE Wire.