ARK research: Bitcoin leads Ethereum and Solana in decentralization

Joint research report of ARK Invest and Glassnode: Comparison of the degree of decentralization of Bitcoin, Ethereum and Solana

ARK Invest and Glassnode released a joint research report on September 1, pointing out that three entities may cross the block production threshold of Bitcoin and Ethereum, while Solana requires 19 entities.

Summary

According to a joint report, Bitcoin reached the 51% computing power threshold through three mining pools. Ethereum requires three pledge entities to have more than 33% shares, but pooled commissions complicate the direct control assumption. Solana's Nakmot coefficient is 19, but almost all of the infrastructure measured runs in commercial data centers. Bitcoin's infrastructure is relatively decentralized, with 63% of measurement nodes running anonymously through the Tor network. About 49% of Ethereum's execution-level nodes are hosted in a cloud environment, with 20% running through Amazon Web Services.

The 32-page report, titled "Decentralized Spectrum: Design Tradeoffs for Digital Assets," compares networks in terms of ownership, exit liquidity, verification costs, key resilience, reconstruction costs, and infrastructure distribution.

The research results do not mean that three companies control Bitcoin or Ethereum. This indicator treats mining pools and pledge platforms as entities, and even if the underlying hardware, pledge or node operators belong to different participants, these participants can withdraw or reallocate their resources.

Bitcoin's three mining pool thresholds do not equal ownership

The report applied a 51% computing power threshold to Bitcoin. Foundry USA accounted for 27.27% of the measurement power, followed by AntPool with 17.06%, and F2Pool with 16.96%. These three mining pools add up to more than 61%. This yields a Nakmot coefficient of 3, defined as the minimum number of measuring entities needed to cross the network's critical production threshold. ViaBTC controls another 9.50%, and SpiderPool accounts for 5.82%.

Mining pools coordinate block construction and allocation of rewards, but they do not necessarily have machines that generate computing power. Independent miners connect to mining pools to generate more stable income and can transfer computing power elsewhere. This liquidity limits the equivalence between pool concentration and permanent control. The report estimates that it takes approximately 29 seconds for Bitcoin miners to switch between 1% of computing power positions. A coordinated attack or censorship attempt may prompt participants to leave the relevant mining pool.

Mining pools still affect transaction inclusion and sorting because they often provide block templates used by miners. Therefore, pool concentration represents an operational risk, even if it exaggerates the concentration of underlying mining ownership. This question is not new. Previous cryptocurrency news reports found that at the end of 2022, two mining pools produced most of the Bitcoin sample blocks. Since then, the pool share has changed, but production remains concentrated in the hands of several large coordinators.

Ethereum crosses lower thresholds through pooled pledges

ARK and Glassnode set the pledge threshold for Ethereum at 33%, because controlling one-third of participants who pledge ETH can undermine finality. This is different from Bitcoin's 51% majority threshold, so the two coefficients describe different powers.

In reporting July data, Lido accounted for 23.04% of pledged ETH, Binance controlled 8.88%, and Kraken held 6.91%. These three entities together account for approximately 38.8%, putting Ethereum above the selected threshold. Lido is not a single verifier. It allocates pledges to multiple node operators, but these operators participate through a common agreement and governance framework. As a result, the report views Lido as a shared infrastructure that aggregates economic weights rather than one machine or company directly controlling each verifier.

Ethereum's exit mechanism also limits the mobility of verifiers. The report estimates that under current conditions, it will take approximately 14.6 days to withdraw a 1% pledged position, and may take as long as 55.6 days in severe congestion conditions. This is much slower than redirecting Bitcoin computing power.

Client diversity provides another layer of resilience. The study set Geth's execution client share at 34.88%, Nethermind at 26.96%, and Reth at 18.98%. Lighthouse accounted for 54.16% of consensus clients. Different clients independently implement Ethereum's rules, reducing the exposure of the network to a single software flaw. The relationship between Ethereum nodes and their software clients means that verifier concentration alone cannot describe the risk of complete failure of the network.

Solana's 19 validator results are accompanied by infrastructure costs

Solana recorded the highest Nakmot coefficient at the selected block production threshold. The report found that 19 verifiers were needed to control more than 33% of entrusted pledges. Figment was the largest single verifier with 3.78%, followed by Helius with 3.69%, Jupiter with 2.91%, Binance Pledge with 2.81%, and Ledger by Figment with 2.16%. The remaining 84.65% are among other validators.

A text in the report mentions that Solana requires 20 entities, but its charts, comparison tables, and published Glassnode summaries all show a factor of 19. The table also states that the number has increased from 18 in March 2026. Solana's validator distribution performs well on this particular metric, but its physical infrastructure is more concentrated. Almost 100% of the infrastructure the researchers measured runs in commercial data centers. About 68% are located in Europe and 21% are located in North America. TeraSwitch hosts 30.23% of measured pledges, and the first two hosting companies serve approximately 35.7%. Even if the verifier set contains many independent operators, a common infrastructure can cause related failures.

This risk became apparent in August, when 699 of the 102 Solana validators stopped voting during TeraSwitch routing issues. Solana continues to process transactions, but this incident shows how one infrastructure failure affects multiple other independent validators. The report uses Solana geographical data from November 2024, while most Bitcoin and Ethereum infrastructure data comes from July 2026. This temporal difference limits direct comparisons and leaves room for Solana's distribution that may have changed.

Bitcoin leads in infrastructure resilience and auditability

Bitcoin has the cheapest verification requirements in research. Researchers estimate that the hardware cost for a full node is $289, compared with $730 for Ethereum and $21,478 for a Solana RPC node or verifier level configuration. It measures a full-chain storage requirement of 753GB. Ethereum's full archive setup requires approximately 2TB, while reconstructing Solana's history is estimated to require 480TB, as historical data is often offloaded to external providers.

Bitcoin also has the most decentralized custody configuration. Only 16% of measurement infrastructure runs in data centers, while 63% of nodes use Tor. The other 15% are residential or self-managed. About 49% of Ethereum's execution layer nodes are located in the cloud environment, and 45% are self-hosted. AWS alone hosts approximately 20%, with the top two providers accounting for approximately 27%. Solana's high hardware and bandwidth requirements reflect its focus on throughput. The cost is that fewer ordinary users can independently rebuild or verify a complete network history using consumer-grade devices.

A single score is inconclusive on blockchain decentralization

The report ultimately rated Bitcoin as the most decentralized overall of the three networks, followed by Ethereum and Solana. Bitcoin leads in ownership distribution, auditability and geographical resilience. Ethereum is usually in the middle of the six dimensions. Solana scored high in key resilience thresholds and verifier engagement, but ranked low in ownership distribution, verification accessibility, and infrastructure diversity.

The method remains sensitive to how entities are grouped. Exchanges can hold tokens for many customers, pool independent miners, and pledge agreements coordinate multiple operators. Wallet size ranges can also consolidate assets held by thousands of users. Therefore, this comparison is more useful as a map of different concentrated risks than as a final ranking. A network may be widely distributed for block production while relying heavily on several hosting companies, software clients, or governance organizations.

Future versions can improve comparability by using synchronized data dates, separating mining pools from underlying resource owners, and distinguishing between review thresholds and thresholds that can rewrite final history.

FAQ

Do three entities control Bitcoin? No. The three measured pools exceeded 51% of the computing power, but independent miners provided most of the computing power and the pools could be replaced.

Can three Ethereum platforms rewrite blockchain? The three entity numbers in the report involve a 33% pledge threshold related to the finality of damage. It does not represent the stricter two-thirds threshold required for other consensus operations.

Why does Solana score 19? 19 is the minimum number of validators whose consolidated pledge exceeds the reporting 33% threshold.

Which blockchain does the report rate as the most decentralized? [TAG Bitcoin ranks highest overall because of its authentication accessibility, decentralized ownership, and relatively resilient geographic infrastructure.