A venue clearing more than $200 billion a month, holding roughly 70% of on-chain perpetuals volume, is secured by 27 validators. Its foundation ran every one of them at launch. Both the critics and the defenders are working from stale numbers, so here is the audit: what the set looks like now, which powers actually exist, and where the honest gap remains.

The most valuable thing about a decentralization argument is usually the data it forces into the open, and the Hyperliquid version has been running on stale data for eighteen months. In January 2025 a node operator published a letter noting that five foundation validators controlled more than 81% of staked $HYPE across a set of sixteen, and that number entered the discourse and never left it. In June 2026, a prominent investor declared the network not permissionless at all, citing validators concentrated in a single building, node software that remains closed, and a foundation that can jail operators and force upgrades on them. Both interventions were treated as verdicts. Neither reflected the current state of the network, which had by then expanded to 27 validators with foundation-run nodes holding slightly less than half the stake, and neither engaged with what the protocol’s own documentation says about the powers in dispute. Meanwhile the thing being argued over kept growing: a venue processing more than $200 billion a month, holding roughly 70% of decentralized perpetuals volume, generating on the order of a billion dollars a year in fees, with an order book, a matching engine, and a liquidation system all running on those 27 machines. This piece is the audit both sides have been arguing without: the set as it stands, the powers as documented, the precedent where those powers actually fired, and the gap that survives every correction.

The set, counted

Start with the trajectory, because the direction is the part the standing critique omits.

Hyperliquid launched with a handful of validators, all run by the foundation, in what amounted to a permissioned network wearing a public ticker. The set expanded to 16 in January 2025, the moment that produced the original decentralization letter and the 81% concentration figure. In April 2025 the foundation restructured registration itself: the set moved to 21 nodes, with registration open to anyone and the 21 largest by stake forming the active set, which converted validator status from an appointment into an auction. Growth continued through 24 to 27 as of June 2026, with a stake threshold to enter that has run above a million $HYPE, a number that itself functions as the network’s real admission price.

The concentration figure moved with it. Following a round of redelegations from foundation validators in June, foundation-run nodes hold approximately 49.3% of staked $HYPE, with about 50.7% distributed across 22 independent operators. The foundation runs five validators of the 27. That is a materially different network from the one described by the 81% figure still circulating in criticism, and any honest audit has to lead with the improvement before cataloguing what remains.

JUST IN: Bitget CEO Gracy Chen declares Hyperliquid is not a true DEX but a centralized exchange, citing the JELLY incident pic.twitter.com/rVehUbfWyI

— crypto.news (@cryptodotnews) April 7, 2026

The mechanics underneath are worth stating precisely, because they define who can participate. Consensus is delegated proof of stake: validators require a minimum self-delegation of 10,000 $HYPE locked for a year, delegators face a one-day lock and a seven-day unstaking queue, and rewards accrue continuously with automatic recompounding. There is no automatic slashing anywhere in the system, which is unusual and cuts both ways: no operator loses stake for a mistake, and no operator loses stake for misbehavior either, leaving the unstaking queue and social consequences as the enforcement layer. Governance runs on delegated stake weight, with validators declaring positions and outcomes determined by the tokens behind them, not by validator headcount, which means the concentration number is the governance number, not a trivium.

The three powers, examined

Now the specific allegations, taken one at a time against the documentation, because two of the three survive and one does not.

Jailing. The claim that has traveled furthest is that the foundation can jail a validator for any reason and remove it from the active set. The protocol documentation describes something different: validators can be jailed through peer voting for latency and reliability failures, and a jailed validator stops producing rewards for its delegators until unjailed, with no slashing attached. Peer-triggered removal for performance is standard practice across proof-of-stake networks and is not foundation discretion. The residual concern is real but narrower than the accusation: when foundation-affiliated nodes hold close to half the stake, peer voting weighted by that stake is not fully independent of the foundation, so the mechanism is only as neutral as the distribution underneath it. That is an argument about concentration, which is the argument this piece keeps returning to, and not an argument about arbitrary power.

Forced upgrades. The claim that validators must adopt protocol upgrades is essentially accurate and largely unremarkable. Every chain running a single client implementation faces the same reality: nodes that decline an upgrade fall out of consensus, which is a coordination fact, not a governance power. What makes it sharper here is the single-binary architecture. Hyperliquid runs one implementation, which the foundation has defended by pointing out that Solana operated the same way for years. The defense is honest and incomplete: single-client networks concentrate the risk that a bug or a decision in one codebase becomes the whole network’s bug or decision, which is precisely why Ethereum’s client diversity is treated as a security property instead of an inefficiency.

Closed source. This one stands, and it is the most consequential of the three. The node software has remained closed, with the foundation’s position since early 2025 being that the code will open when it is stable, citing development speed and security. Eighteen months and considerable growth later, the promise is still outstanding, and it is the crux of the June criticism: a validator running a binary it cannot read is trusting the author in a way that no amount of stake distribution fixes. Users can verify state on-chain, but nobody outside the team can independently verify what the software does before it produces that state. For a venue clearing $200 billion a month, that is the single widest gap between what the network claims and what an outsider can check.

The precedent: when the powers fired

Governance arguments stay abstract until an incident makes them concrete, and Hyperliquid’s arrived in March 2025 with a memecoin called JELLY.

A trader opened a large position and manipulated the thin spot market underneath it, engineering losses that landed on the protocol’s liquidity vault, the pool that absorbs liquidated positions on behalf of depositors. With the vault facing an eight-figure hit, validators voted to delist the market and settle it at a price favorable to the protocol, and the loss was contained. The intervention worked, users were protected, and the affair was over within hours.

JUST IN: CZ calls Hyperliquid’s invention awesome for filling a Binance gap. He highlights their no-KYC model while questioning decentralization claims pic.twitter.com/WYQdYOM2H7

— crypto.news (@cryptodotnews) June 18, 2026

It also answered the governance question empirically. A market that traded on a network can be closed by a stake-weighted vote when the network’s own capital is at risk, and the vote at that time ran through a validator set in which the foundation held a decisive share, which is why the episode was described in the trade press as a validator put: an implicit guarantee that the house will intervene when the house is losing. Two readings follow, and both are defensible. The generous one is that any exchange, decentralized or otherwise, must be able to halt manipulation, and a venue that let a vault be drained by an obvious attack would deserve the criticism it received instead. The unforgiving one is that decentralization is only tested at the moment intervention becomes attractive, and Hyperliquid intervened. What the incident settles is not whether the network is good or bad but what it is: a venue with a functioning emergency brake and a small number of hands on it. Traders should price that accordingly, in both directions, since the same brake that protected vault depositors in March 2025 is the brake that could close a market a trader is winning in.

The comparison that survives every correction

Strip out the stale numbers and the overstated claims, and one gap remains that no redelegation fixes: the set is very small relative to what it secures.

Twenty-seven validators sits against roughly 1,800 on Solana, several hundred on Cosmos Hub, and hundreds of thousands on Ethereum. The technical counterargument is legitimate and worth stating properly: Byzantine fault tolerant consensus does not require thousands of participants for safety, it requires an honest supermajority within whatever set exists, and a small high-performance set is exactly how the network achieves the sub-second finality that makes an on-chain order book viable at all. Hyperliquid’s entire product advantage, matching and finality fast enough to compete with centralized venues, is purchased with validator-set size. That is a deliberate trade, not an oversight.

The question is whether the price is right at this scale, and the arithmetic is uncomfortable. A set of 27 secures a venue processing over $200 billion monthly, with open interest, vault deposits, and now equity-linked and other builder-deployed markets on top. The attack surface that matters is not cryptographic but social and regulatory: 27 operators are 27 phone calls, 27 jurisdictions to subpoena, 27 relationships to pressure, and the foundation’s near-half stake means a much smaller number of conversations would decide most outcomes. The delegation program that expands the set applies identity checks to participants, which improves accountability and simultaneously means the expansion is curated, not open, in practice. Each of those facts is defensible on its own terms. Together they describe a network whose decentralization is best characterized as a managed trajectory: real, measurable, improving, and still a long way from the property its marketing language implies.

The regulator arrives

Which is where the argument stopped being philosophical. On June 26, Singapore’s Monetary Authority added Hyperliquid to its Investor Alert List, the register of entities that consumers might wrongly believe are licensed. The listing is not a ban, not an enforcement action, and not a finding of wrongdoing, and Hyperliquid’s response was accurate on every point: it has never claimed authorization from the regulator, nothing about the network changed, users retain self-custody, and settlement remains on-chain. Bybit had joined the same list nine days earlier, KuCoin in February, Binance since 2021, which places Hyperliquid in familiar company and suggests a regulator working through a list instead of singling out a protocol.

NEW: Singapore’s MAS flags Hyperliquid $HYPE as unlicensed, adds it to investor alert list https://t.co/5o4UCLGTBX pic.twitter.com/jAPfxaJXCj

— crypto.news (@cryptodotnews) June 27, 2026

The significance is what the listing does to the vocabulary. Permissionless has been a technical description inside crypto and is becoming a legal position outside it, because a protocol claiming to be infrastructure rather than an operator is making an argument about who, if anyone, is responsible for the venue. The critique that landed the same day, that a network with closed-source software, a curated validator set, and foundation-weighted governance does not meet the description, is therefore not merely a purity argument. It is a claim that the legal position rests on facts the network has not fully proven, and regulators reading the same debate will reach their own conclusions about which entity, if any, is running the exchange. That is the real stake of the governance question in 2026, and it is why the numbers in this piece matter beyond ideology: the distance between 49.3% and something much smaller, and between closed source and open, is also the distance between a plausible infrastructure claim and a contestable one.

The listing power, and the money behind it

One dimension of the governance question sits outside the validator debate entirely, and for traders it may be the more consequential one: who decides what trades here.

The network’s newer listing machinery, the builder-deployed markets that opened perpetuals creation beyond the core team and produced the equity-linked contracts this publication audited separately, is gated by stake rather than by approval. Deploying a perpetual market requires staking a large $HYPE position for a minimum period, and builder deployments on the EVM side run through a periodic auction for slots. Read one way, that is the most genuinely permissionless part of the system: no committee decides which markets exist, only capital does, which is why the venue could list synthetic equity exposure faster than any regulated exchange could convene a meeting about it. Read another way, it replaces gatekeeping with a wealth qualification, and it means the venue’s expanding product surface, including markets that touch regulated asset classes, is determined by whoever can post the stake.

The economics tie the two halves of the governance question together. Trading fees flow into the token’s buyback machinery, which this publication has covered as crypto’s clearest example of a network routing real revenue to its asset, and staked $HYPE is simultaneously the security bond, the governance weight, and the listing key. That triple duty is elegant design and a concentration mechanism at once: the same token that secures the chain decides its rules and controls what it lists, so any accumulation of $HYPE is an accumulation of all three powers together. On a chain where roughly half the stake already sits with one affiliated group, and where an entry ticket to the validator set runs above a million tokens, the practical question is not whether the system is permissionless in principle but how much capital it takes to matter, and the answer has been rising with the token.

That is the frame worth carrying out of this audit. Hyperliquid’s governance is not a story about a foundation refusing to let go; the trajectory shows the opposite, steadily and measurably. It is a story about a design in which influence tracks capital with unusual directness, on a venue whose scale now exceeds most regulated exchanges, with the software still unreadable from outside. Whether that is acceptable is a judgment each user makes. What it is, precisely, is now on the record.

JUST IN: Hyperliquid to expand active validator set from 24 to 27 in approximately one month pic.twitter.com/1JtvwKIN1l

— crypto.news (@cryptodotnews) May 19, 2026

What to watch

The stake distribution, not the validator count. Headcount is the easy number to grow and the least informative. Whether foundation-run stake continues falling below 49.3%, and whether any single independent operator accumulates a blocking position, is the measure that determines who actually decides outcomes.

The open-source commitment. The promise to publish node software has been outstanding since early 2025 and is the single change that would most alter the audit. Its continued absence is itself information, and the longer it runs, the weaker the stability rationale becomes.

The next intervention. JELLY showed that the network will act to protect its vault. The next comparable event, and whether the decision runs through a stake distribution that no longer has a foundation majority behind it, is the test of whether governance changed or only its arithmetic did.

Regulatory follow-through. The Singapore listing has no operational effect today. Whether other jurisdictions follow, and whether any of them treats the foundation as the operator of an unlicensed exchange, is the scenario in which every fact in this audit stops being a debating point and becomes evidence.

Disclaimer: This article is for information and educational purposes only and does not constitute financial or investment advice. Validator counts, stake distributions, and protocol parameters change continuously, and figures reflect data reported at the time of writing. Nothing here is a recommendation to buy, sell, hold, or trade any asset or on any venue. Always do your own research. Information is accurate as of July 26, 2026.

Frequently Asked Questions

How many validators does Hyperliquid have?

Twenty-seven as of June 2026, up from four or five at launch, then 16 at the start of 2025, 21 in April 2025, and 24 later that year. Registration is open to anyone, with the largest stakes forming the active set, and entry has required a stake above roughly one million $HYPE. Validators must self-delegate a minimum of 10,000 $HYPE locked for one year.

Who controls the stake?

Foundation-run validators hold approximately 49.3% of staked $HYPE following redelegations in June, with about 50.7% spread across 22 independent operators. The foundation operates five of the 27 validators. This is a substantial change from early 2025, when a widely cited analysis put foundation-controlled stake above 81% across a set of 16.

Can the foundation remove validators at will?

Not according to the documentation. Jailing is described as peer-triggered for latency and reliability failures, with a jailed validator ceasing to earn rewards until unjailed, and there is no automatic slashing in the system. The legitimate concern is indirect: because peer voting is weighted by stake and foundation-affiliated nodes hold close to half of it, the mechanism’s independence is limited by the same concentration issue that affects governance generally.

Is Hyperliquid’s code open source?

The node software has remained closed, with the foundation stating since early 2025 that it will open the code once development is stable, citing security and shipping speed. That commitment is still outstanding, and it is the most substantive of the standing criticisms: validators run a binary they cannot audit, and no distribution of stake compensates for that.

What was the JELLY incident?

In March 2025 a trader manipulated a thinly traded memecoin market to push losses onto the protocol’s liquidity vault. Validators voted to delist the market and settle it at a price that protected the vault, containing an eight-figure loss. The intervention worked and was also read as evidence of a validator put, meaning the network will act when its own capital is at risk, through a stake distribution the foundation then dominated.

How does the validator count compare to other chains?

It is far smaller: roughly 1,800 validators on Solana, several hundred on Cosmos Hub, and hundreds of thousands on Ethereum, against 27 on Hyperliquid. Byzantine fault tolerant consensus does not require large sets for safety, and the small set is what delivers the sub-second finality an on-chain order book needs, but it concentrates social, regulatory, and coordination risk for a venue processing over $200 billion a month.

What did the Singapore listing mean?

The Monetary Authority of Singapore added Hyperliquid to its Investor Alert List on June 26, a register of entities consumers may wrongly believe are licensed. It is not a ban or an enforcement action, and Bybit, KuCoin, and Binance appear on the same list. Its importance is that it moves the permissionless question from a technical debate into a legal one, since the claim to be infrastructure rather than an operator depends on the governance facts being what the protocol says they are.

What should traders take from this?

That the network has a functioning emergency brake with a small number of hands on it, and that this is a property to price, not a scandal to condemn. Decentralization here is a managed trajectory: measurably improving on stake distribution, unresolved on source code, and small relative to the value at risk. Position sizing on any venue should reflect the governance reality, not the marketing vocabulary. This is educational analysis, not investment advice.