Is Blockchain Really Decentralized — From Mining Pools to MEV, Where Power Reassembles

The "Decentralisation Illusion" According to the Bank for International Settlements

In 2021, the Bank for International Settlements (BIS) diagnosed that blockchain-based finance rests on a "decentralisation illusion." The reasons were that centralized governance to set and change rules is unavoidably necessary, and that blockchain consensus mechanisms tend to concentrate power. Consensus, as defined in "What Is a Blockchain", is the process where multiple parties who do not particularly trust each other agree on a single answer as to what is written in a ledger. A legal paper (preprint) published on SSRN gathers evidence supporting this diagnosis. And it finds at least part of the reason why people continue to be drawn to the narrative that "blockchain will scatter economic power" in everyone's own normative preconceptions about how much concentration of economic power they are willing to accept (DOI: 10.2139/ssrn.7042298).

"The Right to Write a Line in the Ledger" defined decentralisation as a state where the power to write to the ledger is not concentrated in one or two places but scattered among multiple independent participants. This article expands that question to the places where stakes are delegated, transaction order is decided, servers are run, and rules are changed. However, academia has not yet reached a consensus on what "decentralisation" means and how to measure it. Depending on what is measured and how, the same network looks both decentralised and centralised.

Many of the cited papers are preprints, meaning early drafts released before peer review, and for those with a heavy load, we marked them within the sentences. Because concentration changes significantly over time, we also recorded the time points of the data. This article does not make judgments on where and what percentage is gathered right now.

The Promise of a System No One Controls

The promise is like this. Blockchains settle transactions without intermediaries. Bitcoin made it unnecessary to rely on a third-party financial institution through cryptographic proofs, and Ethereum layered smart contracts on top of that—programs that execute according to conditions written in a ledger—to run even complex computations (arXiv:2404.04841).

Web3 is an internet built on blockchain that champions shifting power from governments and corporations to token holders, and its raison d'être is decentralization. Yet, among industry analysts, few can explain what "decentralization" means, how it differs from "distribution," or with what observational data it can be measured. One organizational theory study therefore calls Web3 "decentralization theater," where actors play roles to fit a fictitious narrative (DOI: 10.1108/s0733-558x20240000089010).

Expectations went further. The hope is that when code is law, it will be fairer than the governance of biased humans, bringing about a society where token holders jointly decide the rules to be etched into smart contracts (DOI: 10.3389/fbloc.2023.1109544).

This promise could become an unexamined premise. A study systematically reviewing 89 papers from 2009 to 2019 notes that because Bitcoin's original contribution of decentralization is assumed to naturally exist, it could be left out of examination. The same review summarizes that studies from around that time conversely observed a growing trend of centralization in Bitcoin and Ethereum, and argues that centralization must be measured because it affects the security of blockchains (arXiv:2009.12542).

Measured Again, Almost Every Layer Is Narrow

In this article, layers are the multiple folds that make up a blockchain. The consensus layer that decides the turn to write in the ledger, the market where stakes are deposited, the order of transactions, the servers and network beneath them, and the governance above them are each one fold.

The Mining Layer — Concentration Under the Name of Pools

In a proof-of-work ledger, the right to write the next page is bought by paying with computation ("The Right to Write a Line in the Ledger"). Hashrate is the amount of computational operations per year spent on that calculation. When mining alone, turns come rarely and erratically, so the vast majority of miners pool their shares together to aim for turns collectively, mining in mining pools that distribute the rewards based on what was contributed. One study notes that this increased pool participation reduces the decentralization level of most cryptocurrencies (arXiv:1905.04624).

Looking inside the pools reveals an even narrower picture. An empirical study from 2019 (preprint) deeply analyzed the reward distribution of three out of the four major Bitcoin mining pools. It showed that in every pool analyzed, a small number of fewer than 20 actors took home over 50% of the BTC payouts issued by that pool. The same miners were sometimes active across three pools simultaneously. However, the authors add the caveat that the extent to which pool operators actually control the pool's resources is still an open debate (arXiv:1905.05999). The fact that there were multiple pools did not mean there were multiple participants behind them.

The passage of time points in the same direction. A modeling study from 2024 (preprint) observed that over the preceding 10 years, the number of mining pools had dropped by roughly one-third. A truncated power-law appeared in the hashrate distribution—where a small number holds a large share, but the largest side is capped at a certain line—and the authors interpret this as a "rich-get-richer" effect occurring within the limits of global energy supply (arXiv:2411.10249).

Bitcoin was narrow from its starting point. A study (preprint) that reconstructed the early blockchain using a technique grouping pseudonymous addresses by common owners reports that from its launch (January 3, 2009) until the day its price reached $1 (February 9, 2011), the vast majority of Bitcoin was mined by just 64 actors. Around that time, nearly all addresses were linked to them within six transactions (arXiv:2206.02871).

Proof-of-Stake Layers — Exchanges and Liquid Staking

In 2022, Ethereum shifted to proof-of-stake, a method of reaching consensus by proving one's own share in the ledger instead of performing computations ("The Right to Write a Line in the Ledger"). A validator is a participant who deposits a share in the proof-of-stake ledger and, based on that share, proposes and confirms the next page. Energy consumption dropped significantly, but concentration did not.

A study (preprint) that followed the beacon chain—which handles Ethereum's consensus—from the beginning after the transition observed two things. First, periods of dropped network participation all stemmed from upgrades, issues with major consensus clients (programs run by validators), or problems at service providers operating many validators at once. Second, the level of decentralization of staking power was similar to the proof-of-work era, and (liquid) staking services had a massive influence on this (arXiv:2306.10777). This means validators were clustered under operators to the extent that a problem at an operator with many validators manifested as a drop in network participation, and the authors conclude that this high centralization raises security concerns.

Staking is the act of depositing one's share into the ledger and earning the right to write based on that share, and while deposited, that share cannot be withdrawn and used. Liquid staking is a method where, instead of the deposited share, you receive a token that serves as a certificate of deposit, allowing you to use and sell that token even while your stake remains deposited. Ethereum's Lido issues such tokens (stETH) (arXiv:2410.23422). By the time that study (preprint) was released in 2024, liquid staking had become the largest category in decentralized finance (DeFi)—financial services running on smart contracts—based on total value locked (TVL). Some view this as a threat to the security of proof-of-stake blockchains (arXiv:2401.16353).

Measurements point in the same direction. In a study comparing the openness of eleven proof-of-stake networks, Ethereum scored high in terms of the number of participants and operating costs, but relatively low in capital concentration and staking ratio. The authors trace the reasons to a low staking ratio relative to the circulating supply and the large stake held by staking pools like Lido (arXiv:2305.13259). A study (preprint) measuring ten major blockchains using the Nakamoto coefficient, the Gini coefficient, and the Herfindahl-Hirschman index also reports that stakes are heavily concentrated among a small number of validators (arXiv:2504.14351).

The Nakamoto coefficient is a value counting the minimum number of entities that must collude to halt or break the system, and the smaller it is, the more power is concentrated. The Gini coefficient is an inequality indicator that grows larger as shares become more uneven, and the Herfindahl-Hirschman index is a concentration indicator that increases as shares are concentrated among a few, reaching its maximum when a single entity holds everything.

There is also an opposing view. Restaking is a method that allows validators to secure multiple services using the same deposited stake. One study (preprint) argues that restaking and liquid staking lower the barrier to becoming a validator and free up locked capital, thereby actually aiding decentralization (arXiv:2410.23422). This debate is far from over.

The Power to Determine Order — MEV

The newest center is the position that determines transaction order.

The transaction I sent does not go straight into the ledger, but waits along with other transactions trying to get into the next page. A mempool is a waiting room where transactions that have not yet entered the ledger wait, and in a public mempool anyone can see the waiting transactions. The party that builds the next page—miners in proof-of-work, or the designated validator (proposer) in proof-of-stake—decides which ones to include and exclude, and in what order to place them.

Even the same transaction yields different results depending on the order. Suppose someone uploaded a transaction to the waiting room to buy a large amount of a certain token. When that transaction goes in, the price rises. The party determining the order, or a party buying front and back turns by offering higher transaction fees, can insert their own purchase immediately before and their own sale immediately after it to make a profit. This technique of sandwiching someone else's transaction back and forth is a sandwich attack.

MEV (Maximal Extractable Value) is the profit that can be extracted from the position of including, excluding, and reordering transactions that will go into the next page. The automated trading programs on Ethereum seen in "The Right to Write a Line in the Ledger"—programs that anticipate ordinary users' transactions and buy and sell ahead of them—are an example targeting this profit. A literature review in the field of information systems views MEV in the same lineage as traditional finance's front-running, where a party ahead in information or power gains unfair profits at the expense of others. The same review notes that even though the blockchain ledger is decentralized and transparent, MEV is particularly common and difficult to prevent (DOI: 10.1007/s12525-024-00727-x).

An SoK (Systematization of Knowledge) preprint divides the history of MEV into three eras. The first (2014–2020) is the proof-of-work era of priority gas auctions where parties trying to get the front spot in the public mempool competitively bid up fees; the second (2020–2024) is the era of designs that separate out the task of building blocks and running them via MEV-Boost, along with Ethereum's transition to proof-of-stake; and the third (2024–) is the cross-chain era where value is extracted between devices connecting multiple ledgers, such as rollups, bridges, and sequencers (arXiv:2603.07716). Those devices are covered in "Is an L2 Really as Safe as Ethereum?" and "A Bridge Does Not Inherit the Chain's Guarantees".

There are also studies that measured the scale. A 2021 study (preprint) estimated the value extracted by sandwich attacks, liquidations (the process of paying off a loan instead where the collateral value has dropped below the threshold and taking over the collateral cheaply), and decentralized exchange arbitrage from 32 months of on-chain data to be 540.54M dollars across 11,289 addresses. This value also reaches consensus. If the value staked on one side is large, an incentive can arise for the party writing it to rewind the already written page and pocket that value for themselves. As an earlier result taken as a starting point by this paper, a rational miner with 10% of the hashrate has an incentive to fork the chain if MEV opportunities exceed 4 times the block reward. A fork is when a chain splits into two branches ("What Is a Blockchain"), and here it is a branch to recreate already attached blocks and hijack that profit. The authors conclude that a system where miners receive MEV transactions from a centralized relay server instead of the public waiting room actually worsens this consensus-layer attack (arXiv:2101.05511).

Devices created to reduce the side effects of MEV have also failed to dissolve centralization. Flashbots is a private channel (private pool) that sends transactions directly to miners instead of the public waiting room, sharing only the name with mining pools. In a 2022 measurement (preprint), miners using Flashbots accounted for over 99.9% of Ethereum's hash power, and over 80% of MEV extraction went through Flashbots. Yet over 90% of Flashbots blocks came from just two miners. The authors note that mining is as concentrated as it was before Flashbots, and while powerful miners' earnings have more than doubled, the share of non-miner parties has decreased by that much (arXiv:2206.04185).

The position that determines transaction order is power that becomes money in itself, and power concentrates in that position as well.

Other Layers — Infrastructure, Geography, Governance

The decentralization story usually stops at the consensus layer. There are many narrow places both above and below it.

First is infrastructure. A study analyzing public data acknowledges that blockchains have the potential to significantly lower infrastructure costs for financial services, yet reports that the Web3 landscape it observed relies heavily on centralized cloud providers for server operation, scaling, and performance. The authors call this a paradoxical concentration of value and power within the decentralized ecosystem (arXiv:2311.08234). The network layer is no different. A study quantifying the connections of computers holding ledger copies, namely nodes ("What Is a Blockchain"), using graph models and simulations, found that hub-type nodes where many nodes cluster together, like relay gateways, reduce the degree of decentralization. The same study also revealed that the average shortest path affecting network speed and the degree of decentralization are traded off against each other (DOI: 10.3389/fbloc.2022.812957). This means a fast network and a widely scattered network do not go in the same direction.

Geography can also be measured. One study observed that in actual networks, most validators are clustered geographically close together. A minority of distant validators easily failed to meet performance requirements and were mistaken for being broken, with the result that they tended to be kicked out of the validator set (jailing) or have their staked funds slashed as a penalty (slashing). The authors report that they obtained the same results in controlled simulation experiments as well, and together showed the effectiveness of a fix that can be easily plugged into consensus protocols (arXiv:2305.17771).

In governance, a corporate governance study argues, based on the cases of Bitcoin and Ethereum, that blockchains have evolved into a technocracy—a system ruled by technological elites—where developers, foundations, and corporations hold disproportionate control. The author concludes that blockchain governance is no exception to the coordination problems, collective action failures, and centralizing tendencies that have plagued traditional governance (DOI: 10.1111/corg.70008).

DAOs (Decentralized Autonomous Organizations) are organizations where token holders vote on proposals to set rules (arXiv:2605.18990). An empirical study of on-chain voting (governance on the ledger) across 21 DAOs discovered high concentration of voting power, significant hidden costs, and a surprisingly large amount of pointless governance activity. (arXiv:2302.12125). Where decisions are actually made is also a problem. A multi-case study summarizes that a substantial portion of governance decision-making factors still reside off the ledger (off-chain) and only some are truly on the ledger, noting that highly centralized elements often exist even within decentralized systems (DOI: 10.3389/fbloc.2023.1256651).

Why Concentration Happens

When observations point this way, it is time to ask for the cause. Various studies arrive at similar explanations at different levels. Mainly, model studies explain concentration as the result of each participant acting rationally and arising from the structure of incentives.

A Rational Choice to Reduce Income Volatility

Miners joining a pool is a choice to reduce risk and smooth out their income. The previous study on pool participation starts precisely from this premise, presenting a model that reflects the miners' degree of risk aversion to optimally split and wager their hashing power across multiple pools. The conclusion points toward a prescription: they showed with past Bitcoin data that splitting across multiple pools increases a miner's return relative to risk, and the authors argue that such splitting also increases the decentralization of Proof of Work (arXiv:1905.04624).

This overlaps with the logic where scale itself becomes an advantage. A study applying an oceanic games framework—which analyzes situations consisting of a few large participants and a multitude of individually negligible participants—to mining shows that individual miners have an incentive to band together to increase the value of their resources. This is another perspective on observing the observed concentration of mining power (arXiv:1904.02368).

A Fork Makes Larger Miners Wealthier

In proof-of-work, unintentional forks always happen. When two miners find the next block at almost the same time, the chain briefly splits and one block is discarded. A 2025 theoretical study (preprint) analyzed this discarding by assuming that the time it takes for a found block to spread to other miners—that is, the block propagation delay—is fixed. As a result, mining yield depended linearly on the hash rate ratio. This means the share earned per unit of computation is larger for larger miners, and the authors call this phenomenon Bitcoin's "The Rich Get Richer" (arXiv:2506.13360).

There is one more inconvenient result in the model study that observed the aforementioned drop in solvability. The more concentrated the hash rate is and the faster block propagation is, the lower the fork rate becomes. The authors showed this through both observation and modeling (arXiv:2411.10249). Because forks can increase resource waste and network instability, if we look at it solely through the yardstick of waste, concentration makes the system "efficient." The fact that efficiency and decentralization do not point in the same direction is what makes this problem difficult.

Reward Design Invites Centralization

Proof of stake is no different. A stake pool is a pool that gathers stakes together to aim for a validator spot. Research that designed reward splitting so that stake pools form fairly proposes a method to settle the target number of pools into a balance—in other words, a state where no one gains more by unilaterally changing their strategy ("The Right to Write a Line in the Ledger")—while simultaneously warning that a seemingly "fairer" simple reward split instead converges to a centralized balance (arXiv:1807.11218).

Research analyzing the long-term dynamics of stake shares shows an even more extreme case. Depending on the reward method, the share of large investors remains stable while the share of medium and small investors can fluctuate or converge to zero, and under geometric rewards, a "chaotic centralization" occurs where all stake eventually clusters randomly in a single investor. The authors also characterized the threshold where stability and concentration diverge (a phase transition critical point) (arXiv:2206.02227).

The Financialization of Staking

The deposit structure itself can give birth to concentration. Tezos is a proof-of-stake ledger where holders can temporarily deposit (delegate) their share to a node, and while this method is called liquid proof-of-stake, it is a different mechanism from the liquid staking mentioned earlier. A study analyzing Tezos points out that delegation creates super-additivity. Super-additivity is a property where the probability of the combined party being drawn as the ruler to forge the next block can be greater than the simple sum of the probabilities when they are separate (DOI: 10.3390/info13120556). This means a node that pools shares can be drawn more often than the proportion of what it pooled.

Numbers within the rules also change the game. A game-theoretic model of the Ethereum staking market suggests that solo stakers running validators directly on their own equipment may be more sensitive to reward changes than holders going through centralized exchanges or liquid staking providers. The authors find the reason in the cost structure of solo staking and its competition with other ways of accessing MEV and DeFi revenue sources. Therefore, this model predicts that reducing the amount of newly issued coins is likely to push out solo stakers (arXiv:2503.14385).

The most radical prediction comes from a 2026 macro model (preprint). The wealth of investors accessing external yields in traditional finance (TradFi) grows exponentially, and that capital floods in to crush staking yields inside the protocol to effectively zero. Then, a cap is placed on the wealth of consumers who must hold money to use for transactions, and consumers eventually stop staking. The authors' conclusion is that the larger the scale, the more completely the proof-of-stake consensus layer is captured by the investor side (arXiv:2604.26076). This is the conclusion of a model built on specific assumptions.

The Position That Decides the Order Becomes an Axis of Power

MEV also creates feedback loops. Proposer-Builder Separation (PBS) is a design that strips the task of organizing which transactions go into a block and in what order away from the proposer, handing it over to a separate entity called a builder. Builders receive transaction bundles from those seeking MEV opportunities, construct blocks, and bid on them to proposers, with proposers submitting the most profitable block. A study (preprint) analyzing this auction as a game reports that leaving MEV alone can centralize the distribution of stake and undermine the security of consensus, noting that Ethereum introduced this design to protect decentralization on the proposer side. The authors show that there are alternative auction mechanisms that create equilibria more beneficial to participants than the method at that time, and back this up with actual data records (arXiv:2405.09465). The very fact that a design emerged to separate the authority to create blocks from the authority to organize the order shows that the position of deciding the order has already become an axis of power.

What, at Which Layer, and Why We Measure It

Up to this point, these are the arguments from those who say it is "concentrated." There are at least three holes in this picture.

How to Measure Decentralization

The most fundamental gap is measurement. A SoK paper (preprint) that systematizes decentralization measurement methods divides previous studies into three axes: which resources are measured, how their distribution is extracted, and what function turns them into a single number, and reports two uncomfortable facts by examining actual data.

First, seemingly trivial data processing choices, such as thresholds placed on the length or distribution of the estimation interval, significantly alter the calculated level of decentralization. Second, in an exploratory factor analysis checking whether various metrics measure the same thing, the consensus layer participation of proof-of-work blockchains captured an isolated signal unrelated to decentralization (in proof-of-stake, multiple metrics were bundled into a single factor). The authors write that this result challenges the community's long-standing assumption that "more participation means higher decentralization" (arXiv:2501.18279).

Changing the direction of measurement also changes the conclusion. Almost all existing metrics measure who builds blocks, that is, the starting point. One preprint measured the opposite direction: where the money leaving validator wallets ultimately flows. Treating wallet-to-wallet transfers as a Markov chain (a probability model where the next step depends only on the current position) and finding the final destination after a long flow, the fund flows of 76,855 Ethereum wallets over four years converged into just four destination accounts, none of which were the three dominant miners of this network (DOI: 10.2139/ssrn.7081020). This means who creates blocks and where money goes can show different concentrations.

The verdict can also split depending on what and how far one looks. In a study measuring mining power distribution with 2019 data using three metrics—the Gini coefficient, Shannon entropy (a value that grows larger as shares are scattered more evenly), and the Nakamoto coefficient—across three time units of days, weeks, and months, the same trend emerged regardless of the metric and time unit changes: Bitcoin was more decentralized and Ethereum was more stable (arXiv:2101.10699). In a study comparing Bitcoin and Steem, a Delegated Proof-of-Stake chain where holders cast votes proportional to their stakes to elect representatives who build blocks, the verdict split depending on the scope observed. Bitcoin was more dispersed among top miners but less dispersed overall, and neither was clearly better than the other (arXiv:2002.02082).

The sentence "Blockchains are centralized" is not yet in a verifiable form. Unless it specifies which resource, at which layer, and measured in what time unit, it is neither true nor false.

Is Concentration Failure Right Away?

The second hole is the distance between concentration and actual harm. Operational concentration does not automatically lead to censorship or theft.

Early Bitcoin is a case in point. Back when those same 64 people mined most of it, they could at any time carry out a 51% attack—rewriting an already recorded ledger by using more than half of the right to write to write the same coin over and over again ("The Right to Write a Line in the Ledger"). Then the community collapses. Researchers modeled this dilemma as an N-person Centipede game, where anyone can siphon off a common good whose value is rising, but doing so ruins that good. In practice, potential operators always chose cooperation, and combining theory and economic experiments showed that anonymous groups playing the game cooperated more frequently even when individual payoffs were the same. The authors' conclusion is that Bitcoin's early success relied on cooperation among a small group of founders instead of decentralization (arXiv:2206.02871). It is a record of concentration not leading to attacks, yet also a record of that safety relying on people's choices instead of mechanisms.

There are also cases where regulatory shocks changed the landscape of concentration. According to a study analyzing all 97,005 Bitcoin blocks mined from May 2020 to February 2022 along with the corporate locations of pool operators, before and after China's 2021 mining ban, the block share of China-affiliated pools dropped from 78.47% to 60.01% and the non-China-affiliated share rose from 3.50% to 27.02%. Measured within the non-China side, the normalized Herfindahl-Hirschman index fell from 10,000, the maximum of the scale, to 4,191, meaning dispersion among non-China operators grew. However, the same paper also notes that a single non-China operator absorbed about 67% (15.7 percentage points) of the increased share. The landscape changed, but it did not scatter evenly. The authors state that the pool-level regression coefficient (-5.45 percentage points) did not reach the 5% significance level, and note that the main driver was a shift in aggregate composition (DOI: 10.3390/fintech5030063).

There is also evidence that country-level shares of Bitcoin mining actually spread out more evenly over a certain period. A preprint analyzing a panel of hash rate shares across 105 countries (September 2019–January 2022) and the Cambridge Electricity Consumption Index reports that the Theil index, which measures inequality among countries, decreased over time and that $\sigma$-convergence occurred, with the spread of country shares clearly shrinking. The same study also reveals that the asymmetry between the core and the periphery persists in long-term shares (DOI: 10.20944/preprints202603.0926.v1).

On the proof-of-stake side, there are theoretical results pointing in the opposite direction. A study treating the interaction between centralized exchanges and the proof-of-stake ecosystem as a mean-field model—a mathematical model that follows the average movement instead of countless individual participants—obtained the result that, under certain assumptions, active trading on centralized exchanges can actually increase staking participation and scatter the staking distribution more in numerical experiments (arXiv:2606.09003). Whether this result will be borne out by real data is not yet known.

Can It Be Fixed, and at Which Layer

The third hole is the prescription. Applying a weighting model that compresses large stakes by taking the square root or logarithm of a validator's influence to the stake distribution of ten blockchains, it reports that square-root weighting improved decentralization metrics by an average of 51% and logarithmic weighting by 132% (arXiv:2504.14351). A prior study reported that square-root weighting improves the Gini coefficient by an average of 37.16% and the Nakamoto coefficients for liveness (ensuring the system does not halt) and safety (ensuring the ledger does not become incorrect) by an average of 101.04% and 80.09% (arXiv:2312.13938). Because both are calculations applied to the current stake distribution, they do not state whether the improvements are maintained if a single holder splits their stake into multiple shares.

When that splitting is examined through voting, an impossibility result emerges. Quadratic voting is a rule that dampens the influence of large holders by ensuring that votes increase less as holdings grow. A 2026 preprint writes that it proved no rule that draws voting power solely from wallet balances can succeed in permissionless blockchains (arXiv:2605.18990). Because anyone can create a wallet without permission, a sybil attack—an attack that inflates headcount with fake identities ("What Is a Blockchain")—in which tokens are split across multiple wallets is possible, and as long as even a single wallet receives votes, the total voting power of an attacker who splits their holdings grows at least linearly with respect to their holdings under any of the concave rules proposed to suppress large holders. When the authors recalculated the ten most recent finalized agenda items at the time of the paper across five actual DAOs (ENS, Compound, Uniswap, Arbitrum, and ZKsync) using these rules, the multiplier by which votes swell from splitting wallets (the sybil amplification multiplier) ranged from 1,172 to 4,039 times in quadratic voting and exceeded 229,000 times in power-law rules that suppress large holders more heavily.

A similar limitation exists on the MEV side. A study formalizing arbitrage-style MEV that can be extracted from liquidity pools—pools that hold two types of tokens and exchange them for each other—(a preprint) proves that if block time is fixed and the pool satisfies the natural properties of actual pools, this MEV remains unchanged even if the ordering mechanism or block time distribution is altered (arXiv:2304.11010). This means the size of this type of MEV cannot be altered simply by tweaking ordering rules. There are also studies approaching from the regulatory side. One legal study examines the extent to which sandwich attacks, decentralized exchange arbitrage, liquidations, and transaction-ordering manipulation are permitted under the market abuse provisions of the EU's Markets in Crypto-Assets Regulation (MiCAR), and whether the actors involved can be punished for market manipulation or insider trading (DOI: 10.5771/9783748970873).

The biggest question remains. Which layer of decentralization was promised in the first place? The earlier systematic literature review divides centralization concerns into six architectural layers and 13 dimensions—governance, network, consensus, incentive, operational, and application—and emphasizes that the implications of centralization vary depending on which dimension is observed (arXiv:2009.12542). Mining can be concentrated while nodes remain dispersed, and validators can be dispersed while governance remains narrow. The question "Is a blockchain decentralized?" cannot be answered without specifying the layer.

Bitcoin is not free from this ambiguity either. A study analyzing Bitcoin through the lens of a DAO evaluates that Bitcoin implements the DAO principles of open participation and decentralized decision-making through Bitcoin Improvement Proposals (BIPs), miner signaling, and user-driven rule changes activated by users (User Activated Soft Forks) (arXiv:2504.20838). At the same time, it points out the risk that economic power is concentrated among large stakeholders, obscuring "who actually votes."

Trust Does Not Disappear, It Relocates

So is the conclusion that "blockchains are not decentralized"? This article does not make that definitive claim.

What the evidence supports is simply this: Decentralization is a set of distinct scales that must be measured separately at each layer. In mining, a small number of pools; within pools, a small number of actors; in proof-of-stake, staking services; in sequencing, harsh private channels for the majority of MEV extraction; in infrastructure, the cloud; in governance, evidence that developers, foundations, and large holders concentrate power—these repeat at every layer. At the same time, there are cases where that concentration did not lead to actual attacks, cases where regulatory shocks reshaped the landscape, and instances where Bitcoin mining dispersed across countries for a time. Sentences that make a definitive claim in either direction usually run ahead of the data.

The conclusion of one regulation and governance study is that using "trustless" technology does not make the problem of trust disappear, but rather requires rebuilding new trust relationships across multiple layers. Abstract properties attributed to technology do not realize themselves automatically in concrete applications (DOI: 10.1111/rego.12553). Research theoretically analyzing power relations arrives at the same place. It argues that Web3's society of "code is law" fails to untangle unequal power relations, and merely shifts the arena of power struggles between existing and new actors, much like other technological revolutions have done (DOI: 10.3389/fbloc.2023.1109544).

Value judgments are also at stake in this debate. In a study surveying 3,710 participants in the blockchain ecosystem, respondents were sharply polarized regarding how they perceived economic power to be distributed and how they believed power should be distributed in governance (DOI: 10.3389/fbloc.2023.1125088). The answer to "how concentrated must it be to be a problem" does not emerge from data alone.

Therefore, when confronted with sentences like "blockchains are decentralized" or "they are actually centralized," there is only one question to ask in return: Which layer, at what time, measured by what metrics are we talking about? Claims lacking answers to that question, whether in support or refutation, are narratives that cannot yet be verified.

---

This article is a summary of research on the technical and governance structures of blockchain systems and is not a recommendation or advice to invest in any specific asset.

Source Papers

• "Blockchain-Based 'Decentralization': A Rorschach Test for Normative Priors" (2026) — DOI: 10.2139/ssrn.7042298 (preprint) — BIS's diagnosis of the "illusion of decentralization" and supporting evidence.
• "Web3 as Decentralization Theater?" (2024) — DOI: 10.1108/s0733-558x20240000089010 — Absence of an operational definition for decentralization, and the alternative framework of authority dispersion.
• "Taxonomy of Centralization in Public Blockchain Systems" (2020) — arXiv:2009.12542 (preprint) — Review of 89 papers, classifying centralization across 6 architectural layers and 13 dimensions.
• "Unveiling Decentralization: Bitcoin, Ethereum, and Solana" (2024) — arXiv:2404.04841 (preprint) — The starting point of exchange without intermediaries.
• "A Deep Dive into Bitcoin Mining Pools" (2019) — arXiv:1905.05999 (preprint) — In 3 out of 4 major pools, fewer than 20 people received over 50% of BTC payouts.
• "Diversification Across Mining Pools" (2019) — arXiv:1905.04624 (preprint) — Pool participation lowers income volatility but reduces the level of decentralization.
• "How the interplay between power concentration, competition, and propagation affects the resource efficiency of distributed ledgers" (2024) — arXiv:2411.10249 (preprint) — Pool count decreased by about 1/3 over a decade; concentration actually lowers the fork rate.
• "The Rich Get Richer in Bitcoin Mining Induced by Blockchain Forks" (2025) — arXiv:2506.13360 (preprint) — Under a fixed propagation delay assumption, the return rate depends linearly on the hashrate share.
• "Oceanic Games: Centralization Risks and Incentives in Blockchain Mining" (2019) — arXiv:1904.02368 (preprint) — Incentives for miners to merge and increase the value of their own resources.
• "Cooperation among an anonymous group protected Bitcoin during failures of decentralization" (2022) — arXiv:2206.02871 (preprint) — 64 people mined most of Bitcoin from 2009 to 2011; attacks were possible, but they chose cooperation.
• "Ethereum Proof-of-Stake Consensus Layer: Participation and Decentralization" (2023) — arXiv:2306.10777 (preprint) — Measuring the beacon chain after the merge, the causes of falling participation rates, and the impact of staking services.
• "Decentralization in PoS Blockchain Consensus" (2025) — arXiv:2504.14351 (preprint) — Stake concentration across 10 chains; square root and logarithmic weighting improve metrics by 51% and 132%.
• "How Does Stake Distribution Influence Consensus?" (2023) — arXiv:2312.13938 (preprint) — Square root weighting improves the Gini coefficient by 37.16% and the Nakamoto coefficient by 101.04%/80.09%.
• "Empirical and Theoretical Analysis of Liquid Staking Protocols" (2024) — arXiv:2401.16353 (preprint) — Liquid staking is the largest DeFi category by TVL, sparking debate over PoS security threats.
• "Mitigating Challenges in Ethereum's PoS: EigenLayer and Lido" (2024) — arXiv:2410.23422 (preprint) — The opposing view that restaking and liquid staking lower barriers to entry.
• "Network Participation and Accessibility of PoS Blockchains" (2023) — arXiv:2305.13259 (preprint) — Comparing 11 networks; Ethereum has a high number of participants and a low capital concentration score.
• "Towards a Formal Framework of the Ethereum Staking Market" (2025) — arXiv:2503.14385 (preprint) — The possibility that issuance cuts will squeeze out solo stakers.
• "The Financialization of Proof-of-Stake" (2026) — arXiv:2604.26076 (preprint) — A macroeconomic model showing how external financial returns lead to the institutional capture of the consensus layer.
• "Proof of Stake economy under centralized exchanges — a mean field model" (2026) — arXiv:2606.09003 (preprint) — Counter-results showing that centralized exchanges can actually promote stake dispersion.
• "Stability of shares in the Proof of Stake Protocol" (2022) — arXiv:2206.02227 (preprint) — Chaotic centralization where stakes concentrate in a single investor under geometric rewards.
• "Reward Sharing Schemes for Stake Pools" (2018) — arXiv:1807.11218 (preprint) — Seemingly "fairer" reward distribution schemes instead converge to a centralized equilibrium.
• "Liquid Proof-of-Stake in Tezos: An Economic Analysis" (2022) — DOI: 10.3390/info13120556 — A structure where delegation creates superadditivity in the probability of block producer selection.
• "SoK: The Evolution of Maximal Extractable Value, From Miners to Cross-Chain" (2026) — arXiv:2603.07716 (preprint) — The three eras of MEV and measurement challenges.
• "Maximal extractable value: Current understanding, categorization, and open research questions" (2024) — DOI: 10.1007/s12525-024-00727-x — Defining MEV as profit extraction based on power asymmetry.
• "Quantifying Blockchain Extractable Value: How dark is the forest?" (2021) — arXiv:2101.05511 (preprint) — 540.54M USD extracted over 32 months, and the fork incentive threshold for a miner with 10% hashrate.
• "A Flash(bot) in the Pan: Measuring MEV in Private Pools" (2022) — arXiv:2206.04185 (preprint) — Flashbots miners hold over 99.9% of hash power, with over 90% of blocks coming from two miners.
• "Flashback: Enhancing Proposer-Builder Design with Future-Block Auctions" (2024) — arXiv:2405.09465 (preprint) — The concern that ignoring MEV leads to stake centralization and compromised consensus security.
• "Invariance properties of maximal extractable value" (2023) — arXiv:2304.11010 (preprint) — Proving that MEV is invariant to changes in ordering mechanisms and block time distributions.
• "Maximal Extractable Value" (2026) — DOI: 10.5771/9783748970873 — Examining the permissible limits of four types of MEV under EU MiCAR market abuse regulations.
• "The Myths of Blockchain Governance" (2025) — DOI: 10.1111/corg.70008 — A technocracy where developers, foundations, and corporations exercise disproportionate control.
• "The Hidden Shortcomings of (D)AOs" (2023) — arXiv:2302.12125 (preprint) — Empirical study of 21 DAOs, revealing voting power concentration, hidden costs, and meaningless governance activity.
• "Concave is the New Linear: The Impossibility of Anti-Plutocratic DAO Governance" (2026) — arXiv:2605.18990 (preprint) — Proving the impossibility of wallet-balance-based voting, with 1,172 to 4,039 times sybil amplification.
• "Bitcoin, a DAO?" (2025) — arXiv:2504.20838 (preprint) — The DAO-like nature of BIPs, miner signaling, and UASFs, and the ambiguity of "who votes".
• "Analyzing decision-making in blockchain governance" (2023) — DOI: 10.3389/fbloc.2023.1256651 — A significant portion of decision-making happens off-chain.
• "Centralized Intermediation in a Decentralized Web3 Economy" (2023) — arXiv:2311.08234 (preprint) — Web3 infrastructure's reliance on centralized cloud computing.
• "Toward Quantifying Decentralization of Blockchain Networks With Relay Nodes" (2022) — DOI: 10.3389/fbloc.2022.812957 — Hub-like relay nodes lower the degree of decentralization.
• "Analyzing Geospatial Distribution in Blockchains" (2023) — arXiv:2305.17771 (preprint) — Geographic concentration of validators, and the tendency to evict and penalize distant validators.
• "SoK: Measuring Blockchain Decentralization" (2025) — arXiv:2501.18279 (preprint) — Data processing choices alter results; no correlation between participation and decentralization in PoW.
• "Where Does Ethereum Validators' Money Go? A Spectral Analysis" (2026) — DOI: 10.2139/ssrn.7081020 (preprint) — Destination-side measurement showing that the flows of 76,855 wallets converge into four destination accounts.
• "Measuring Decentralization in Bitcoin and Ethereum using Multiple Metrics and Granularities" (2021) — arXiv:2101.10699 (preprint) — 2019 data showing Bitcoin is more distributed and Ethereum is more stable.
• "Comparison of Decentralization in DPoS and PoW Blockchains" (2020) — arXiv:2002.02082 (preprint) — Comparing Bitcoin and Steem, with neither being clearly superior.
• "Pool-Operator Concentration After the 2021 China Mining Ban" (2026) — DOI: 10.3390/fintech5030063 — Full census of 97,005 blocks, China-linked share dropped from 78.47% to 60.01%, normalized HHI from 10,000 to 4,191.
• "Decentralization Under Energy Growth: Geographic Reallocation and Convergence in Bitcoin Mining" (2026) — DOI: 10.20944/preprints202603.0926.v1 — Panel of 105 countries, persistence of σ-convergence and core-periphery asymmetries.
• "Realizing a blockchain solution without blockchain?" (2023) — DOI: 10.1111/rego.12553 — "Trustless" technology does not eliminate trust problems but demands new trust relationships.
• "Algorithmic governance, code as law, and the blockchain common" (2023) — DOI: 10.3389/fbloc.2023.1109544 — Web3 does not dissolve power relations but merely shifts the arena of struggle.
• "Political, economic, and governance attitudes of blockchain users" (2023) — DOI: 10.3389/fbloc.2023.1125088 — Survey of 3,710 participants, showing polarization in perceptions of economic power.