Network, measured

Solana validators: who holds the stake, and what they run

Validator counts are easy to find. What is harder to find is how concentrated the stake is at the thresholds Alpenglow actually uses, and what the client mix looks like now that most validators run one codebase with a different block builder on top.

Read from the RPC; hosting per stakewiz Data as of Updated November 16, 2026 8 sources
Short answer

662 validators were voting on November 16, 2026. 8 of them hold more than 20% of all stake, 18 more than a third, and 25 more than 40%. Under Alpenglow, Solana's stated safety margin is 20% malicious stake plus 20% offline (SIMD-0326), so 8 is the number that maps to the protocol's own bound; most published "Nakamoto coefficients" use one third. Agave-based clients run 81.6% of voting stake and Firedancer-based clients 18.1%.

How many validators

  • Voting
    662
  • Delinquent
    38
  • Delinquent stake
    0.08%
  • Active stake
    429.3M SOL
  • Nodes in gossip
    3,838

Read at 2026-11-16 00:42 UTC, epoch 1087. A validator (an operator's computer that votes on blocks and produces them in turn) is counted once per vote account. "Delinquent" means its vote account has stopped voting. Nodes in gossip include RPC servers, which do not vote. Alpenglow caps the voting set at the 2,000 highest-staked validators (SIMD-0326).

How many validators it takes to pass each threshold

Sort validators by stake, largest first, and count how many it takes for their combined stake to pass a share of all stake. 20%, a third and 40% are lines in Solana's consensus rules, old and new; 50% and two thirds are shown for scale.

  • More than 20% 8 validators
  • More than a third (33.3%) 18 validators
  • More than 40% 25 validators
  • More than 50% 40 validators
  • More than two thirds (66.7%) 77 validators
  • 20%: Alpenglow's Byzantine bound. SIMD-0326 describes "a distinctive 20+20 security model": the protocol stays safe with up to 20% of stake acting maliciously, plus up to another 20% offline. 8 validators together hold more than 20%.
  • 33.3%: the classic rule. Under Tower BFT, the consensus Solana ran until October 21, 2026, a third of stake could halt the chain. Most published "Nakamoto coefficients" still use it: 18 validators here.
  • 40%: Alpenglow's crash bound. The 20% + 20% case. SIMD-0326 calls it "40% crash failure resilience". 25 validators.

The largest single validator holds 3.94% of all stake, and the ten largest hold 24.8%.

Which threshold each published figure uses

Different sites publish different "Nakamoto coefficients" for Solana because they measure different things. Here is what each one counts.

SourceFigureThresholdCountsRead
SolPilled (this page)820% (Alpenglow, Byzantine)validatorsNovember 16, 2026
SolPilled (this page)18one third (classic)validatorsNovember 16, 2026
SolPilled (this page)2540% (Alpenglow, crash)validatorsNovember 16, 2026
Nakaflow (Chainflow)6one third, by entity (validators grouped by operator)operators (grouped)November 10, 2026
Solana Compass18one third, by validator (matches our per-validator count at 33.3%)validatorsNovember 10, 2026

Grouping matters: one company can run many validators, so a per-validator count overstates how many independent parties are needed. Nakaflow groups by operator using its own entity map, which we have not checked; that is why its figure is far lower. Neither approach uses the 20% line that Alpenglow's safety argument rests on.

What software the stake runs, on two axes

Each node reports a client name in gossip. Those names now mix two separate questions: whose validator code it is built from, and what builds or orders the blocks it produces. Asking them separately gives a clearer picture than "Agave vs Firedancer vs Jito".

By codebase

  • Agave and Agave-based81.6%
  • Firedancer and Firedancer-based18.1%
  • Unknown0.4%

By block-building layer

  • BAM (Jito's Block Assembly Marketplace)40.5%
  • Harmonic (block-builder marketplace)32.4%
  • Jito's original client19.4%
  • None (stock client)6.1%
  • Rakurai scheduler1.2%
  • Unknown0.4%
  • Raiku0.1%
Every client name reported, by share of voting stake
Client name (clientId)CodebaseBlock-building layerValidatorsShare of stakeMain versions
AgaveBamAgaveBAM36038.49%4.4.1, 4.4.0
HarmonicAgaveAgaveHarmonic5021.44%4.4.1, 4.4.0
JitoLabsAgaveJito12219.35%4.4.1, 4.4.0
HarmonicFiredancerFiredancerHarmonic4410.95%26.10.8, 26.10.6
FiredancerFiredancernone345.11%26.10.8, 0.1.1
FireBAMFiredancerBAM211.99%26.11.0, 26.10.8
RakuraiAgaveRakurai81.22%4.4.1, 4.4.0
AgaveAgavenone180.99%4.4.1, 5.0.0-alpha.0
unknown idotherother30.32%0.0.0, 4.4.1
RaikuAgaveRaiku10.09%4.4.1
not in gossipotherother10.04%

What the names mean (BAM, Harmonic, Rakurai and Raiku as each describes itself): Agave is Anza's validator client, the descendant of the original Solana Labs code; Jito's client is an Agave fork with Jito's block engine. BAM (bam.dev) is Jito's network of scheduler nodes that order transactions privately and attest to the ordering. Harmonic (harmonic.gg) is "configurable block building for Solana": each validator picks a block builder. Rakurai is a high-throughput scheduler; Raiku runs trading infrastructure next to its own validator. Firedancer is Jump's independent client written from scratch; Frankendancer was its earlier hybrid.

One reason the Firedancer share is not higher: the switch to Alpenglow on October 21, 2026 required Agave, Jito-Solana or Harmonic builds, and solana.com states that "Firedancer and Frankendancer were not supported for the migration." Firedancer-based clients were back to 18.1% of stake when this was read.

Cross-check: stakewiz's own client field, read the same run, puts Agave-based clients at 81.6% of its live stake, against our 81.6% from gossip. clientId is self-reported, and a node missing from gossip at the moment of the call is left out.

Commission: what validators keep

A validator's commission is the share of the staking rewards it keeps before passing the rest to delegators. Since July 2026 it can be set in basis points (SIMD-0291). The most common setting is 5% (265 validators); 220 charge 0%.

CommissionValidatorsShare of stake
0%22030.60%
0.01% to 5%31225.16%
5.01% to 10%6521.01%
10.01% to 99.99%40.27%
100% (no rewards passed on)6122.96%

The 100% row matters for anyone reading an "average commission": 23.0% of all stake sits with validators that keep every reward, mostly operators staking their own SOL. That drags a stake-weighted average far above what a delegator choosing a validator actually pays. What staking pays at typical commissions is on what staking really pays; what to look at when choosing is in how to choose a validator. We do not rank validators.

Where the stake is hosted, per stakewiz

Hosting and country are not on-chain. These shares come from stakewiz, which maps each validator's IP address to a network operator and a country (geo-IP), read November 16, 2026 across 661 live validators. Geo-IP places the server, not the operator, and some hosts are mislabelled.

By network operator

  • TeraSwitch Networks Inc.25.0%
  • OVH SAS8.2%
  • Allnodes Inc7.9%
  • UAB Cherry Servers7.7%
  • Latitude.sh7.5%
  • Private Customer3.5%
  • UAB Nacionalinis Telekomunikaciju Tinklas3.1%
  • Waverly Community Schools3.1%
  • 105 others34.1%

By country

  • Germany35.2%
  • Netherlands18.9%
  • United States13.7%
  • United Kingdom9.6%
  • Japan5.1%
  • Lithuania4.7%
  • Singapore4.1%
  • Sweden2.2%
  • 20 others6.5%

Hosting concentration is not theoretical. Decrypt reported that on August 12, 2026 a routing fault at a single hosting provider took almost 29% of staked SOL offline, against the 33% threshold in force at the time; the chain did not halt. See the incident record.

Bars are drawn at twice the share so small rows stay visible; the number is the share.

How this is measured

Total stake = activated stake of every vote account, voting or delinquent. Validators are sorted by stake; the count needed to exceed X% is the smallest number of the largest validators whose stake sums to more than X% of the total. Client mix: each voting validator's node is looked up in getClusterNodes and its self-reported clientId is joined to its stake; shares are of voting stake. Codebase groups the clients by whose code they are built from (Agave or Firedancer); block-building layer groups them by what orders their blocks (BAM, Harmonic, Jito's original client, none). Commission is the inflation-reward commission in basis points (SIMD-0291), for validators with stake.

  • Counts are per validator, not per operator. One company can run several validators, so the true number of independent parties needed is lower. Nakaflow, which groups validators by entity, published 6 against a one-third threshold on 2026-11-10.
  • clientId is self-reported in gossip and can be wrong; a validator missing from gossip at the moment of the call is counted as "not in gossip".
  • Hosting and country come from geo-IP via stakewiz: they place the server, not the operator, and some hosts are mislabelled.
  • Which threshold matters is a judgement. Under Alpenglow the protocol's stated safety bound is 20% malicious stake (SIMD-0326); 33.3% is the bound most published figures use; 40% is the stated crash tolerance.
  • Shares move every epoch as stake is delegated and withdrawn.

Questions people ask

How many validators does Solana have?

662 vote accounts were voting and 38 were delinquent (not voting) on November 16, 2026, per getVoteAccounts. Gossip showed 3,838 nodes in total, most of them RPC nodes that do not vote.

What is Solana's Nakamoto coefficient?

It depends on the threshold and on whether validators run by one company are grouped. Counting validators one by one: 18 to pass one third of stake, 8 to pass 20%. Nakaflow, which groups validators by operator, published 6 (one third) on 2026-11-10.

What share of Solana runs Firedancer?

18.1% of voting stake ran a Firedancer-based client (Firedancer, HarmonicFiredancer or FireBAM) on November 16, 2026, by the clientId each node reports in gossip. Plain Firedancer alone was 5.1%.

Which client do most Solana validators run?

By stake, AgaveBam (38.5%), an Agave build that orders blocks through Jito's BAM. The old split "Agave vs Firedancer vs Jito" no longer describes the network: most stake runs an Agave codebase with a separate block-building layer on top.

What does a 100% commission validator mean?

It keeps all the inflation rewards of the stake delegated to it, so delegators earn nothing from it. 61 validators charged 100% on November 16, 2026, holding 23.0% of stake: mostly operators staking their own SOL.

Sources

  1. Solana RPC: getVoteAccounts
  2. Solana RPC: getClusterNodes
  3. SIMD-0326: Alpenglow (the "20+20" security model)
  4. solana.com: Alpenglow (migration clients, Rotor status)
  5. stakewiz validator API (hosting and country by geo-IP)
  6. Nakaflow: Nakamoto coefficient tracker and methodology
  7. Solana Compass: decentralization statistics
  8. SIMD-0291: Commission rate in basis points

The data behind this page

Free to reuse under CC BY 4.0. Collected every two hours from Solana's public RPC. How every number is made

Related

SolPilled is independent and unofficial: not the Solana Foundation, Anza, or any validator. Figures are read from public endpoints at the time shown and can be wrong if an endpoint is; for the network's state right now, check status.solana.com and a block explorer. Nothing here is investment advice.