Solana's Nakamoto coefficient is 19. Bitcoin's is 3. Those two numbers, printed side by side in the decentralization dashboard that ARK Invest and Glassnode released from their September 2025 data snapshot, made the expected round-trip through crypto Twitter: Solana has suddenly decentralized past its critics; Bitcoin is structurally fragile; Ethereum is barely better. I checked the arithmetic, the weighting scheme, and the entity graph behind each number. The arithmetic is fine. The conclusion is not.
I do not read the whitepaper; I read the bytecode. For a decentralization metric, I read the entity relationships behind the dashboard. What I found is a classical error dressed in quant clothing. The Nakamoto coefficient measures the minimum number of independent units that must collude to break the network. But independence is the very assumption the model cannot verify. Solana's 19 coordination units are not 19 independent failure domains. Behind them sits a infrastructure layer that two data-center operators can threaten. The headline metric is real. Its implied safety margin is not.
The Finding in Context
The ARK Invest and Glassnode dashboard is not a whitepaper. It is a live analytical tool, drawing on node data, mining-pool templates, validator registries, staking distributions, and infrastructure disclosures. The September 2025 snapshot was intended to answer a question that has haunted the industry since the collapse of FTX: when a regulator or an institution asks whether a layer-1 network is decentralized, what number do you show them? The dashboard's answer is built on the Nakamoto coefficient, a concept popularized over a decade ago by Balaji Srinivasan and refined by Coin Metrics. It defines the smallest group of entities that, if coordinated, could compromise the liveness or the safety of the protocol under a specific attack threshold. The accounting rules differ by consensus mechanism. For Bitcoin, the unit is the mining pool, weighted by observed block templates. For Ethereum and Solana, the unit is the validator operator, weighted by delegated stake.
The contrast in the headline output is stark. Bitcoin's top three pools — Foundry USA, AntPool, and F2Pool — together touch roughly 59 percent of discovered block templates, which collapses Bitcoin's coordination threshold into the single digits. Ethereum's staking market produces a similar single-digit outcome when the same arithmetic is applied. Solana, at least in the report's framework, produces 19 coordination units, a figure that Solana Compass data cross-verifies. The conclusion that most readers extracted: Solana has become the most decentralized major proof-of-stake chain by consensus coordination threshold. That conclusion is true only inside the dashboard's own narrow definition of consensus coordination. It is dangerously incomplete everywhere else.
What the Coefficient Actually Counts
The first problem is semantic confusion about what a mining pool is versus what a validator operator is. Bitcoin's coefficient is calculated on three pools that together control the majority of block templates. Those pools do not own the hashrate. They are transaction-scheduling and template-building services. The physical miners behind them can redirect their hashrate through a different pool within minutes, and they have done so repeatedly across Bitcoin's history. When Foundry or AntPool changes policy, the market discipline is fast. Miner exit velocity is high. The entity counted as a coordination unit in September 2025 may control an entirely different share of network hash by the time one reads this article. That is not a get-out-of-jail card for Bitcoin — pool concentration remains a censorship exposure, and a pool operator can indeed refuse to include transactions. But it is a qualitatively different failure class from a validator operator that controls a large block of delegated stake.
Solana's 19 units are stake-weighted validator operators. Delegated stake does not exit quickly. It is locked, warmed up, cooled down, and often managed by custodial products that make silent re-delegation a slow process. The committee of validators that the Nakamoto number counts is therefore a committee, not a temporary market snapshot. If a threshold attack requires coordinating 19 large validators, the coordination assumption is more durable than it is in the Bitcoin mining pool case. But durability also means a hidden correlation problem: the 19 entities are stable, which means they are predictable, and predictability is an attack surface only if the model tracks physical relationships. The ARK Glassnode framework partially acknowledges this inside its infrastructure module. The viral headline number simply ignores it.
The deeper methodological issue is threshold mixing. Ethereum's protocol documentation is unusually explicit about this. It separates the one-third threshold that can stall finality from the two-thirds threshold that can compromise safety. These are two different failure modes requiring two different coordination groups. Bitcoin's proof-of-work has its own distinctions between a 51 percent double-spend attack, a 33 percent censorship delay, and a 0 percent eclipse attack on a specific participant, yet the dashboard collapses these into one scalar. Solana's 19 likewise depends on which threshold was chosen to stop the summation of the largest entities. Change the threshold from one-third to two-thirds, and the coefficient for every chain moves in a different direction. The report is not unaware of this — it labels its panels carefully. But the industry does not consume dashboards panel by panel. It consumes a single number, and that number obscures more than it reveals.
The 45.7 Percent Problem
The number that should have dominated the discussion was not 19. It was 45.7. According to the infrastructure data that Solana Compass tracks and that the ARK Glassnode report reflects, two data-center operators — TeraSwitch and Latitude — are associated with approximately 45.7 percent of Solana's active stake. That is not a statement about malicious intent. It is a statement about physical correlation. Solana's high throughput architecture favors validators with colocated hardware, fast networks, and predictable low-latency access to the cluster's turbine protocol. The market response was rational: sophisticated validator operators moved into professional data centers. The consequence is that Solana's validator set is professionally diverse but geographically and physically concentrated. The coefficient counts 19 independent governance entities. The physical layer counts two facilities and perhaps a handful of buildings.
Any attacker with a lower attack cost will choose the lower layer. Why coordinate 19 validators, each with independent keys, independent legal entities, and independent security teams, when the same objective can be achieved by compromising a single network operations center or a single cloud billing account? The Nakamoto coefficient measures the minimum colluding set under the assumption that the only coordination necessary is between consensus participants. It does not measure the cost of coercing those participants, nor the number of shared dependencies that convert 19 entities into 2. This is not an academic abstraction. In my own stress tests of the Compound governance system during DeFi Summer 2020, I simulated a 51 percent governance attack by aggregating COMP wallets and found that the mathematically precise number of tokens was less important than the actual relationships between the top holders. Vote-coordinating wallets shared exchange withdrawal patterns, shared funding sources, and in some cases shared operational infrastructure. The token-weighted Nakamoto coefficient of that system looked healthy on paper. The address-correlation graph looked like a spiderweb. The lesson applies directly to validator sets on every high-performance chain.
A Nakamoto coefficient measured on validator identities is an upper bound on distributed security, not a lower bound. It tells you how many distinct organizations must agree to attack. It does not tell you how many of those organizations share a transformer, a Tier-4 cooling system, or a common software binary.
The Software Exposure
The dashboard's own risk mapping identified software client concentration as the second-largest exposure for the networks under review. This is the part of the report that deserves the attention it is not receiving. Solana's consensus history contains incidents where the network restarted after a cluster halt. Ethereum's consensus layer has survived client bugs because of its client diversity campaign. Bitcoin's nodes run multiple compatible implementations, though the actual node count distribution is uneven. The reason client diversity matters is not political. It is mathematical. A Nakamoto coefficient of 19 assumes independent failures. If all 19 validators in the minimum colluding set run the same client implementation, a single software bug can reproduce the exact failure mode that the coefficient says would require 19 entities to execute deliberately. The network becomes one bug away from behaving as if its coefficient were 1. Client concentration is a coordination amplifier.
This is an especially acute tension for Solana. The chain's performance narrative depends on tight software optimization, tight hardware requirements, and tight operational discipline. Each of those optimizations reduces the diversity of the stack. Every validator that must run the same optimized client, on the same GPU generation, inside the same data-center ecosystem, contributes to a system whose separate entities are functionally interchangeable parts. The report goes some way toward addressing this by listing client software across its decentralized layers, and honest readers will notice that no major chain scores well on all axes simultaneously. But the single scalar — the 19 that gets retweeted — carries none of that nuance. The market hears one number, not five correlated variables.
The Conflation of Bitcoin and Ethereum Numbers
Critics will say that comparing Bitcoin's pool-based number to Solana's stake-based number is like comparing the melting point of steel to the bandwidth of a fiber cable. There is truth in that objection, but it is only half the story. Bitcoin and Ethereum genuinely have low Nakamoto coefficients, and those low numbers have different real-world meanings. Ethereum staking is dominated by a relatively small group of node operators, partly through liquid staking derivatives. That concentration is a validator-layer fact, and it exposes Ethereum to a different coordination risk than Bitcoin's pools. Bitcoin's pools are highly visible, highly regulated in some jurisdictions, and capable of being pressured by legal action. If a government demanded that three pool operators censor transactions, Bitcoin's liveness could survive, but its permissionless property would suffer concrete damage. The three is not a benign number for Bitcoin either. The question is what the dashboard does with those similar outputs — it pronounces both networks approximately equal in coordination terms, then places Solana far above both, creating an implicit ranking of decentralization that the evidence does not support.
The correct reading of the data is that Bitcoin, Ethereum, and Solana have different threat models, not different grades of decentralization. Bitcoin's threat is pool-level censorship and legal coercion. Ethereum's threats are staking-derivative concentration and client monoculture. Solana's threats are physical infrastructure concentration and performance-driven stack homogeneity. A useful metric would place these exposures on separate axes, not merge them into a single rankable number.
The report's own nuance is its best defense. ARK Invest and Glassnode did not publish a press release claiming Solana was more decentralized than Bitcoin. They published a multi-panel dashboard explicitly designed to show that no single metric explains systemic risk. They even warned, in the report's framing, that the Nakamoto coefficient covers only one portion of the attack surface. The problem is the incentive structure around such research. Institutional investors, project marketing teams, and exchange listing committees all want a single answer to the decentralization question. The dashboard provides them a convenient illusion of one.
What the Framework Gets Right
Before dismissing the report, I want to grant the bulls their central point. This framework is a genuine upgrade over the decentralized metrics that preceded it. The old approach counted node numbers, treating a Raspberry Pi in Seoul as equal to an AWS availability zone. The new approach, whatever its limitations, weights entities by actual control over consensus. That is the correct instinct. A Bitcoin node run on a residential connection does not influence block production. A Solana validator with 1 percent of stake does. Counting controlling stake or hashrate is analytically superior to counting participation. The dashboard also refuses to pretend that decentralization is binary. It treats the Nakamoto coefficient, infrastructure distribution, software concentration, and data-center redundancy as separate panels, acknowledging that a network can be centralized in one dimension and decentralized in another. That is the honest starting point for any serious risk assessment.
There is another point that the market's anti-Solana contingent will not enjoy. Solana's 19 coordination units is a real improvement, not an accounting illusion. The validator set has demonstrably broadened since the network's crisis years, and the entity count is not fabricated. If one accepts the report's definitional assumptions, the number is an accurate output. The dashboard also provides a structure that can be extended to other L1 and L2 networks, creating a standardized measure of protocol-relevant control thresholds across the industry. That is a valuable public good. Any standardization that helps regulators distinguish between a network controlled by three entities and a network requiring nineteen is a step toward more rational policy.
The missing context is the industry baseline. Many decentralization analysts treat a Nakamoto coefficient above 30 as the rough ideal threshold for a major network. Solana at 19 is above Bitcoin and Ethereum on this narrow axis, but it remains below the level that institutional risk committees typically require for classification as meaningfully distributed. The dashboard does not try to hide this; the data viewer can see that a score of 19 is not a stamp of decentralization utopia. It is a moderately healthy number for one layer of a deeply correlated stack. The team behind the report deserves credit for resisting the temptation to declare a winner. The community that converted the report into a Solana marketing asset did not show the same restraint.
The Next Metric Must Measure Exit Speed
If I could add one variable to the ARK Glassnode framework, it would be failover velocity. The Nakamoto coefficient measures a static state of power distribution at a moment in time. It does not measure how fast the network can repair that distribution when an entity becomes compromised or hostile. Bitcoin's pool problem is mitigated by fast miner exit. Ethereum's staking concentration is mitigated by the ability of token holders to withdraw and re-delegate, even if the process takes days. Solana's validator market also has exiting mechanics, but the physical infrastructure dependency creates a different bottleneck. If one data-center operator fails or is coerced, the remaining validators cannot instantly rebuild the network's performance profile inside another facility. The hardware is specialized. The network topology depends on low-latency links. The recovery is slow, and slow recovery is exactly when a coordinated attack succeeds.
The next version of this dashboard should publish a coordination coefficient for the physical layer, the software layer, and the geographic layer separately. It should disclose how many entities are required to cross a threshold at each layer and how quickly the network can restore its preferred distribution after a shock. Until then, quoting a Nakamoto coefficient of 19 as evidence of decentralization is like quoting a fire escape's capacity while ignoring the number of stories above it. The two numbers are related. They are not the same.
When I look at the September 2025 snapshot, I see data that is honest in its construction and naive in its interpretation. Solana has more consensus coordination units than its predecessors. That is a fact. But 45.7 percent of the stake sits in two data-center ecosystems, the software layer is concentrated, and the physical recovery options are limited by the very performance architecture that makes the network attractive. A regulator reading the headline number might classify Solana differently from Bitcoin. A systems engineer reading the infrastructure panel would hesitate before drawing any conclusion about systemic robustness.
The ledger remembers what the dashboard forgets. A scalar does not describe a network; it describes the comfort of the person who summarizes it. The next bull market will reward projects that can credibly claim decentralization, and the ARK Glassnode dashboard will be a reference point in those claims. Good. What I ask of every analyst who uses it is this: quote the 19, then quote the 45.7, then quote the client-diversity panel, and then explain why a network that requires 19 entities to coordinate an attack can still be neutralized by one shared binary and two buildings.
Decentralization is not an integer from zero to infinity. It is an attack-tree with multiple branches, and every branch has a different cost. The Nakamoto coefficient names the minimum number of conspirators required. It never asks whether those conspirators already share a server rack. That question is where the next report — and the next crisis — will be written. In the meantime, I will continue to read the bytecode, the stake graph, and the colocation contracts. The coefficient can keep its 19. I will keep the floor price: a network whose physical layer looks like two and whose software layer looks like one has not earned an all-clear from a number nor my trust from a headline.
