Here is the question I keep coming back to whenever someone says a chain is “truly decentralized”: how many actors does it take before the network starts to wobble? Not in a slogan. In production. In the messy middle where pools, staking desks, cloud regions, and client software actually sit. A new comparative study put that question on the table for Bitcoin, Ethereum, and Solana, and the answer is less tidy than the marketing decks suggest.
I’ve found that decentralization debates usually collapse into a single number. That number then gets treated like a championship trophy. The report I read this week does something more useful. It spreads the comparison across ownership, how quickly capital or hash power can leave, how expensive it is to verify the chain yourself, how brittle the critical threshold looks, what it costs to reconstruct history, and where the machines physically live. Bitcoin still comes out ahead overall. Ethereum sits in the middle. Solana looks broader at the validator layer and tighter underneath the floorboards.
What The Decentralization Spectrum Actually Measures
The researchers did not pretend one score settles the argument. They built a spectrum. That matters, because a network can look open at the validator set and still lean on two hosting companies. It can look concentrated at the pool layer and still let individual miners walk away in half a minute. If you only quote the headline coefficient, you miss the tradeoff.
The six lenses in the paper are easy to remember once you stop treating them as academic labels:
- Who appears to own or coordinate production
- How fluid an exit really is when something goes wrong
- What it costs an ordinary operator to verify the chain
- How few entities it takes to cross a critical threshold
- How painful it is to rebuild history from scratch
- Where infrastructure actually lives, geographically and commercially
In my experience, that last item is the one retail conversations skip. People argue about token distribution for hours and never ask whether half the nodes sit in the same cloud region. Geography is not glamorous. It is also where correlated failure likes to hide.
Why A Coefficient Is Not A Crown
The study uses a Nakamoto coefficient style count: the smallest number of measured entities needed to cross a production threshold. For Bitcoin that threshold is majority hash rate. For Ethereum and Solana it is the one-third stake band associated with disrupting finality. Those are not identical powers. Mixing them in a bar chart without that caveat is how arguments go sideways on social feeds.
A coefficient tells you how concentrated production looks on a given day. It does not tell you who owns the machines, who can leave, or what breaks if one data hall loses routing.
That is the sentence I wish more explainers led with. Three mining pools crossing 51 percent is not the same claim as three companies owning Bitcoin. Three staking platforms crossing 33 percent is not the same claim as three desks rewriting Ethereum history. Nineteen Solana validators crossing a third of delegated stake is not a free pass on data-center risk. Each number is a map of one pressure point.
Bitcoin’s Three-Pool Headline Needs A Closer Look
On the hash-rate side, the measured picture is familiar if you have watched mining for a few years. One large United States pool sat near 27 percent. Two other major coordinators followed around 17 percent each. Add those three and you clear 61 percent. Another pair of pools filled out a long tail that still matters, but the first three already cross the majority line used in the paper.
Does that mean three firms run Bitcoin? No. Pools assemble work and hand out rewards. They usually supply the block template miners follow. That is real operational influence over inclusion and ordering. It is not the same as owning every ASIC in the set. Independent operators plug in for smoother payouts. They can unplug. The report’s own mobility estimate is almost startling: shifting a 1 percent hash-rate position took roughly 29 seconds in the modeled conditions.
I keep circling that number because it changes the emotional temperature of the statistic. If a pool starts filtering transactions in a way the market hates, hash can leave faster than a governance forum can finish its first thread. Concentration still creates a tempting coordination surface. Fluidity is the counterweight. Both can be true at once, which is why a single ranking slide never feels honest enough.
Pool power is also not a brand-new story. Earlier cycle snapshots already showed two coordinators printing a majority of sampled blocks. Shares rotate. Names rise and fade. The structural habit remains: a handful of templates sit in front of a much larger field of hardware owners. If you care about censorship resistance, watch the templates. If you care about asset control, watch who can reassign the machines.
| Network | Threshold Used | Entities To Cross It | What That Power Implies |
| Bitcoin | 51% hash rate | 3 mining pools | Majority block production coordination, not automatic ownership of hardware |
| Ethereum | 33% staked ether | 3 staking entities | Ability to disrupt finality, not the two-thirds bar for stronger consensus actions |
| Solana | 33% delegated stake | 19 validators | Broader producer set, still exposed if hosts and regions cluster |
Ethereum Crosses A Lower Bar Through Pooled Stake
Ethereum’s coefficient looks identical to Bitcoin’s at first glance: three. The similarity is cosmetic. The threshold is one-third of stake, because that is the band associated with stalling finality. A liquid staking protocol held a little over 23 percent in the July snapshot used by the authors. A large exchange desk followed near 9 percent. Another exchange sat near 7 percent. Together they cleared about 39 percent.
Again, labels hide mechanics. The leading liquid protocol is not one validator in a closet. It routes stake across many node operators under a shared framework. Treating it as a single entity is a modeling choice about aggregated economic weight. It is a fair choice if you care about coordinated policy, upgrades, and social pressure. It overstates the idea of one machine flipping a switch.
Exit speed is the quieter problem. The same report estimated that leaving a 1 percent position could take around 14.6 days in ordinary conditions and stretch past 55 days if the exit queue clogs. Compare that with Bitcoin’s half-minute hash pivot and you see why I refuse to rank these coefficients as twins. Stake is sticky. Hash is restless. Sticky capital can be a feature for security. It is a bug if you needed to flee a compromised coordinator yesterday.
Client diversity offers a second kind of insurance. Execution-layer shares in the study put one long-running client near 35 percent, another near 27 percent, and a newer implementation near 19 percent. On the consensus side, one client still held a majority near 54 percent. Different teams implementing the same rules reduce the blast radius of a single bug. They do not erase staking concentration. They sit beside it.
Perhaps the most interesting aspect is how often public arguments flatten those layers into one vibe. “Ethereum is centralized because of liquid staking.” “Ethereum is fine because of client diversity.” Both statements can be half right and still leave you unprepared for a messy week. Finality risk, software risk, and custody risk are different animals. Feed them separately.
Solana’s Nineteen Looks Strong Until You Ask Where The Boxes Live
Solana posted the highest coefficient on the selected production threshold. Nineteen validators were needed to gather more than a third of delegated stake. The largest names in the measured set sat in the mid-to-low single digits. No single operator looked like a shadow monarch on that chart. If your only question is “how many distinct producers before we cross 33 percent,” Solana wins the heat.
Then the infrastructure page arrives and the mood changes. Nearly all of the measured footprint sat in commercial data centers. Europe held about 68 percent. North America held about 21 percent. One hosting firm carried a little over 30 percent of measured stake. The top two hosts together reached around 36 percent. Many operators, shared buildings. That is a different kind of concentration.
You did not need a theoretical paper to see the pattern. In August, a routing incident at a major host knocked 102 of 699 validators offline. The chain kept processing. That part is important and often gets dropped in panic posts. The other part is equally important: independent brands can fail together when they rent the same pipes. Decentralized letterhead, correlated electricity.
One awkward detail for purists: some Solana geographic figures in the report were older than the Bitcoin and Ethereum hosting snapshots. Timing gaps make neat league tables a little less neat. Distribution may have shifted since the older sample. Even with that caveat, the qualitative contrast remains. High-throughput design leans on serious hardware. Serious hardware leans on serious rooms. Those rooms cluster.
The Quiet Contest: Who Can Actually Check The Books
I care about this section more than the coefficient horse race, and I will not pretend otherwise. A chain that only specialists can fully verify becomes a chain the public must take on trust. Trust is fine until it is not. Verification cost is where philosophy meets a shopping cart.
The study’s hardware estimates were blunt. A Bitcoin full node setup landed near $289. An Ethereum full-archive style setup landed near $730. A Solana RPC or validator-class configuration landed near $21,478. Storage followed the same slope: about 753 gigabytes for Bitcoin’s measured full chain, roughly two terabytes for an Ethereum archive posture, and an estimated 480 terabytes if you try to reconstruct Solana history the hard way because so much of that history lives with external providers.
Verification reality check Bitcoin node hardware: about $289 Ethereum archive-class path: about $730 Solana validator or heavy RPC path: about $21,478 Full history burden rises from hundreds of GB to hundreds of TB
Those figures will age. Disk gets cheaper. Clients get leaner. Pruning strategies improve. The ranking of difficulty still tells a story about design intent. Bitcoin optimized for a world where a stubborn person with mid-range consumer gear can keep a copy. Ethereum asks more, especially if you want the deep archive rather than a working execution view. Solana asks for a small business budget and a serious pipe if you want to sit close to the firehose.
Is that a moral failing? Not automatically. Throughput is a product choice. Users who want cheap blockspace and fast settlement are not villains for preferring a heavier machine profile. The honest sentence is simpler: fewer ordinary operators will independently recreate the full record. When reconstruction leans on specialized vendors, auditability becomes a professional service. That can still be robust. It is a different social contract than “download the chain on a quiet Sunday.”
Where The Machines Sleep Changes The Risk Story
Bitcoin’s measured hosting mix was the most scattered of the three. Only about 16 percent of observed infrastructure sat in data centers. Roughly 63 percent of nodes hid in Tor. Another 15 percent looked residential or self-hosted. That profile is inconvenient for neat dashboards. It is also inconvenient for a single landlord, a single cloud region, or a single compliance team hoping to tap one rack and call it a day.
Ethereum landed in a split personality. About 49 percent of execution-layer nodes sat in cloud environments. About 45 percent looked self-hosted. One major cloud provider alone held around 20 percent. The top two providers together reached about 27 percent. That is not a cartoon monopoly. It is enough overlap that a regional outage or a policy shock would be felt.
Solana, as noted, was almost entirely commercial. High performance loves dense halls, redundant power, and fat transit. Those amenities do not grow in spare bedrooms at the same rate. If you design for speed, you inherit the geography of professional hosting. I do not think that point gets enough airtime in “which chain is most decentralized” threads, because it sounds less ideological than token-holder charts. It is also closer to how real outages happen.
- Bitcoin: more anonymous routing, more home and self-hosted footprints, fewer measured boxes in classic halls
- Ethereum: a near even split between cloud and self-hosting, with a visible overweight in one large cloud brand
- Solana: commercial density, regional clustering in Europe on the measured sample, shared-host correlation risk
None of this means Bitcoin nodes are magically immortal. Tor helps with privacy and scatter. It does not abolish ISP pressure, power cuts, or amateur operational mistakes. Self-hosting is a culture, not a force field. Still, when I stack the three maps, Bitcoin’s map is the one that looks hardest to scoop into a single building.
Ownership, Custody, And The Way We Group Names
Methodology is where a ranking can quietly smuggle its conclusion. Group exchanges as single entities and custody looks terrifying. Split every customer wallet and the same chain looks like a festival of tiny holders. Group a liquid staking protocol as one actor and Ethereum’s coefficient collapses. Treat every underlying operator as sovereign and the picture loosens. The authors flagged that sensitivity themselves, which I respect. Too many scorecards pretend the grouping choice was obvious.
Wallet-size bands have the same trap. A band that looks like a whale may be a custodian holding thousands of client accounts. A mining pool that looks like a titan may be a bulletin board for rented hash. A validator brand that looks independent may share a cage, a switch, and a maintenance window with twelve neighbors. If your model cannot see those layers, your model is grading letterhead.
I’ve sat through enough conference panels to know how this gets abused. Bulls quote the dimension that flatters their bag. Bears quote the dimension that damns it. A grown-up reading is duller and better: list the concentration surfaces, estimate how fast users can leave each surface, and ask which failures are correlated. That is work. Work does not trend as well as a trophy graphic. It ages better.
The comparison is more useful as a map of separate concentration risks than as a definitive ranking.
What “Most Decentralized” Means After You Finish The Paper
The authors still offered an overall order. Bitcoin first. Ethereum second. Solana third. Bitcoin led on ownership scatter, auditability, and geographic resilience. Ethereum occupied the middle across most dimensions. Solana scored well on the critical producer threshold and participation breadth, then lagged on verification access, ownership optics, and infrastructure diversity.
That order matches my own bias only in part. I have long thought cheap verification is the unsexy foundation of the whole cypherpunk pitch. If a network needs a specialist class to know what the ledger says, politics creeps back in through the side door. Bitcoin protects that door better than the other two on current evidence. Ethereum buys expressiveness and pays in operational weight. Solana buys speed and pays in hardware and housing.
Does that make Bitcoin “done”? Hardly. Pool templates remain a live operational risk. Mining geography can tighten when energy policy shifts. Hardware manufacturing is not a cottage industry. A coefficient of three at the pool layer should keep people uncomfortable even after you explain miner mobility. Comfort is how bad coordination habits calcify.
Ethereum’s homework is different. Sticky exits, liquid staking gravity, and a still-heavy consensus client share are the combination to watch. Progress on client diversity is real and under-celebrated. Progress on making large staking silos less socially and economically magnetic is slower, because users like convenience. Convenience aggregates. Aggregation is the plot.
Solana’s homework is physical. More regions. More hosts. Less coincidence of fate when one vendor’s routing table has a bad afternoon. The validator set already looks competitive on the 33 percent test. The building directory does not. If the next edition of this research syncs dates and separates brands from cages more cleanly, I would not be shocked to see the coefficient stay high while the infrastructure grade becomes the main argument.
Practical Questions Investors And Operators Should Ask Next
If you hold these assets, the useful follow-up is not “who won.” It is “which failure mode am I underwriting.” A few questions have saved me from sloppy takes:
- If the top coordinators filtered a category of transactions tomorrow, how fast could production leave them?
- If one cloud region or one colo brand went dark, what share of stake or nodes would blink together?
- Can I personally run a verifying node on equipment I can buy without a purchase order?
- If I needed the full historical record, would I be reconstructing it or requesting it?
- When analysts group an “entity,” are they grouping a company, a protocol, a custodian, or a building?
Those questions cut through tribe language. A Bitcoin maxi still has to answer the pool-template question. An Ethereum holder still has to answer the exit-queue question. A Solana holder still has to answer the shared-host question. If a thesis cannot survive those prompts, it was a vibe, not a model.
There is also a portfolio angle people underplay. Decentralization is not a coupon you clip each quarter. It is an insurance property. You pay for it in throughput, fees, latency, or operational friction. You collect on it when a company, a government, or a cloud dashboard tries to become the switch. If your horizon is a weekend trade, the insurance premium can look wasteful. If your horizon is a decade of adversarial weather, the premium is the product.
How Future Scorecards Could Stop Talking Past Each Other
The paper itself sketched a better sequel. Sync the observation dates so Solana’s map is not wearing last season’s coat. Separate pools from the owners of hash, and staking brands from the operators behind them. Distinguish a threshold that censors inclusion from a threshold that rewrites finalized history. Those are different crimes. They deserve different meters.
I would add one more request. Publish the correlation matrix, not just the leaderboard. Show me which risks move together. A network can look beautifully scattered on validators and tightly coupled on clients, hosts, and governance forums. Another can look ugly on pools and surprisingly loose on geography and verification. Side-by-side columns hide covariance. Covariance is where the ugly week lives.
Until that kind of scorecard is normal, readers should treat any single ranking as a conversation starter. Bitcoin leading this comparison is a serious result. It is not a permission slip to stop measuring pools. Ethereum sitting in the middle is not a eulogy. Solana landing third on the blended score is not a claim that nineteen validators are a mirage. The spectrum is the point. The trophy is a headline.
A Few Straight Answers Before The Comment Section Lights Up
Do three entities control Bitcoin? No. Three measured pools cleared a majority of hash rate in the snapshot. Much of that hash can leave. Control, in the ownership sense, is a thicker claim than coordination of templates.
Can three Ethereum platforms rewrite the chain? The three-entity figure in this study is about the one-third band tied to disrupting finality. It is not the two-thirds band associated with heavier consensus actions. People collapse those bands because the collapse fits a tweet. The bands are not the same tool.
Why does Solana print 19? Because that is how many validators it took, in the measured set, to gather more than a third of delegated stake. The number is specific. It is not a general grade for the whole architecture. Hosts, regions, and hardware still sit underneath it.
Which network ranked most decentralized overall? Bitcoin, on the authors’ blended reading, thanks to cheaper verification, more scattered ownership optics, and a hosting mix that is harder to sweep into a handful of halls. I think that conclusion is directionally right. I also think the useful work starts after the ranking, when you pick the failure mode you are actually underwriting.
One last personal note, because these pieces get read as tribal scorekeeping no matter how carefully you hedge. I want cheap, widely verifiable settlement to remain possible for people who will never work at a market-making firm. That preference colors how I weigh node cost against raw throughput. Someone building a high-frequency consumer app will weigh the same table differently, and they will not be crazy for doing it. Design tradeoffs are allowed. What is not allowed, if we are being adults, is pretending the tradeoff is free.
Watch the pools. Watch the staking silos. Watch the halls with the loud fans. And keep a machine of your own if the asset is supposed to outlive the companies explaining it. That habit still beats a coefficient on a slide.
