The number arrived without a chain.
Fourteen hundred million dollars in protocol revenue. One point two six billion of it routed into buying back the native token. Fifty-six percent of all on-chain perpetual open interest. Three figures, delivered in a single ecosystem bulletin from a source called HLHUB, dated October 10, year unstated.
I read them three times, and each pass made me more uneasy — not because the numbers are implausible, but because they are unverifiable. There is no transaction hash. No auditor's attestation. No third-party dashboard to cross-reference. Just a revenue claim, a buyback claim, and a market-share claim, all stamped with the logo of an outlet whose name suggests it lives inside the very ecosystem it describes.
In seven years of dissecting protocols, I have learned that the most dangerous sentence in this industry is not "this is a scam." It is "the numbers speak for themselves." Numbers never speak. Someone speaks for them. So let's open the hood, and let's be honest about what we can and cannot see.
Context: why the architecture is the real story
Before we touch a single dollar figure, we need to understand what Hyperliquid actually built, because the architecture is the only part of this story that isn't a claim.
Hyperliquid is not a rollup. It did not queue up behind Ethereum's data availability layer and call itself a Layer 2. It built its own Layer 1 — a consensus called HyperBFT, an order book engine called HyperCore, and an execution environment called HyperEVM. The perpetual futures matching engine runs entirely on-chain as a central limit order book, or CLOB.
That choice is worth pausing on. For two years, the entire industry has been promised "decentralized sequencing" as the fix for rollup centralization. It has remained, in almost every case, a PowerPoint slide with a testnet attached. Hyperliquid simply sidestepped the debate: if the sequencer is the problem, stop having a sequencer. Build the chain yourself and put the matching engine inside the consensus.
It is a bold answer. It is also a different trust assumption, not an absence of one. A self-built L1 means you trust a validator set you cannot fully see. The bulletin tells us nothing about how many validators secure HyperBFT, how they are selected, or how concentrated the stake is. High throughput, low latency, low fees — the three properties that make an on-chain CLOB viable — are usually purchased with exactly this currency: a smaller, faster, more permissioned set of block producers.
It helps to place this against the field. dYdX also chose to build its own chain, but its market share has been under pressure as it migrated away from the liquidity it once borrowed from Ethereum. GMX took the opposite path entirely, using an automated market maker model where traders trade against a pooled counterparty rather than each other, which produces a very different user experience and a very different risk profile. Jupiter Perps aggregated Solana liquidity into a formidable venue but remained, at its core, a router. Hyperliquid's bet is that the matching engine itself belongs in consensus — that the trust-minimizing version of a derivatives exchange is one where the order book is the chain.
That bet has a cost. An AMM can scale by adding liquidity; an on-chain CLOB scales by adding throughput, and throughput on a bespoke L1 is a function of how much decentralization you are willing to trade away. This is the same tradeoff every Layer 2 sequencer makes and then quietly refuses to discuss. Hyperliquid at least put it on the table by owning the whole stack.
The token side is simpler. HYPE has a hard cap of one billion, launched in November 2024, and functions as utility, governance, and staking collateral. The mechanism under scrutiny is the loop connecting the two halves: fees flow in, and a large share of them flow back out to buy HYPE.
So the architecture earns a genuine nod. Now the accounting has to earn its own.
Core: the ninety percent question
Here is the claim that has traveled furthest: of roughly $1.4 billion in cumulative protocol revenue, $1.26 billion — about ninety percent — was used to repurchase HYPE.
If you have watched DeFi long enough, your first instinct is to ask where the money came from. And this is where Hyperliquid genuinely separates itself from the crowd. The buyback is funded by trading fees — real, non-inflationary economic activity — not by minting new tokens and paying them out as yield. That distinction is the whole ballgame. A protocol that prints tokens to reward depositors is borrowing from the future to pay the present. A protocol that collects fees from people who actually want to trade, and uses them to buy its own token, is doing something structurally different.
Based on my audit experience, that is a real signal. But it is a signal about structure, not about magnitude, and the magnitude is where the bulletin goes quiet.
What, precisely, is "revenue"? Is it gross fees before the various allocations? Is it net of the HLP market-making vault and other reward pools? Does the ninety percent figure count the buyback gross, or does it net against tokens distributed elsewhere? The bulletin never defines its terms. This is not pedantry. The 2020 Uniswap V2 audit I ran during DeFi Summer taught me exactly this lesson. The core constant-product formula was fine. It was a subtle rounding error in the price oracle for low-liquidity pairs that quietly disadvantaged retail traders — small enough to escape attention, large enough to matter at scale. A definitional ambiguity in a revenue metric is the same species of bug. It hides in the gap between what a word means to a marketer and what it means to an accountant.
The bulletin also never mentions the HLP vault, Hyperliquid's market-making pool that historically absorbed a meaningful share of protocol economics. If a portion of "revenue" is really vault profit redistributed to depositors, then the ninety percent buyback and the vault allocation compete for the same pool of fees — and the headline ratio could be flattering the buyback by excluding the rest.
The second unanswered question is sharper still: what happens to the HYPE after it is repurchased?
There are two possibilities, and they are not remotely equivalent. If the tokens are sent to a verifiable burn address and destroyed, the supply genuinely contracts, and the buyback is real deflation. If they are moved to a treasury wallet and held, the float only appears to shrink — the tokens still exist, still overhang the market, and can return at any moment. The bulletin does not say which. Until someone points to the address, the buyback is a balance-sheet gesture, not a monetary fact.
Now the ratio itself. Ninety percent is extraordinary, and extraordinary numbers deserve skeptical arithmetic. If a protocol commits ninety percent of revenue to buybacks, it is retaining roughly ten percent for everything else — engineering, security, operations, legal, and the risk buffer every derivatives venue eventually needs. That is a thin margin for a platform running an on-chain order book against a live adversary set. It is a wonderful story for token holders today and a potentially fragile one for the protocol tomorrow.
One more piece of the puzzle is worth naming. The ninety percent figure may not be generosity so much as arithmetic. A protocol with no venture capital overhang and no team unlock cliff has no captive sellers to satisfy, which frees it to route nearly all income back to the market. If that reading is right, the buyback is less a marketing decision than a structural consequence of a fair launch — and structural consequences are more durable than campaigns.
The market-share claim needs the same treatment. Fifty-six percent of on-chain perpetual open interest is a staggering figure, and it is plausible given the liquidity flywheel that derivatives naturally produce: deeper books attract traders, traders deepen the books, and the leader pulls away. But read the qualifier carefully. Fifty-six percent of on-chain open interest is not fifty-six percent of the perpetual market. The largest perpetual venue on earth is still a centralized exchange, and against that denominator the number shrinks considerably. And the figure is self-reported. Third-party dashboards count differently — whether they include centralized venues' on-chain legs, whether they span all chains — and a self-reported number from an ecosystem outlet is a number with an agenda.
Then there is the quiet part that is arguably the most interesting: the wallets. Phantom earned over $25 million and MetaMask over $10.5 million through something called Builder Codes. Strip the branding away and this is revenue-sharing for distribution — wallets route orders to Hyperliquid and take a cut of the fees. It is a genuinely clever flywheel, because it turns every major wallet into a salesperson with skin in the game. But it also means the protocol is giving away a slice of the very revenue that funds the buyback. Distribution bought with margin is still margin spent.
If I were handed the mandate to verify this bulletin, I would not start with the press release. I would start with the fee contract: pull every fee-collection event, sum the inflow over the period, and reconcile it against the claimed $1.4 billion. Then I would trace the outflow to the repurchase wallet and follow it to its final resting address. Two queries. That is all it takes to turn a claim into a fact — and the fact that the bulletin does not invite them is itself a data point.
And the team behind all of this remains, as it has been, largely anonymous — the founder known only by a first name. The bulletin discloses nothing about vesting, about who controls the buyback treasury, or about how the repurchase decision is actually governed. In a system whose entire pitch is trust minimization, the human layer is the least visible and the most load-bearing.
Contrarian: the buyback is a story, not a floor
Here is the blind spot that the ninety-percent headline is designed to cover: a buyback is not the same thing as net buying.
Every dollar of repurchase meets a seller. If early airdrop recipients, market makers, and opportunistic farmers are distributing into that bid, the net change in circulating supply could be flat, or negative, no matter how impressive the gross number looks. The bulletin reports the size of the bid. It says nothing about the size of the offer. In my 2021 work on the Axie Infinity contracts, the lesson was the same in a different costume: a mechanism can function exactly as documented and still be unsafe, because the documentation describes the intent and the market decides the outcome. Audit the intent, not just the syntax — and then audit who is selling into it.
There is a regulatory layer here too, and it cuts against the decentralization story rather than for it. A programmatic, protocol-directed buyback strengthens the "expectation of profits from the efforts of others" prong of the Howey test. And the more centrally that buyback is decided and executed, the weaker the claim that the token is sufficiently decentralized to escape securities characterization. The mechanism that makes HYPE attractive to holders is the same mechanism that makes it legible to regulators. That tension does not resolve; it compounds.
Finally, there is the source itself. HLHUB reads as an ecosystem outlet, and ecosystem outlets have a structural bias: they report the buyback and not the net float, the market share and not the validator count. That is not proof of dishonesty. It is a reason to treat every figure in the bulletin as a claim awaiting corroboration rather than a fact awaiting citation.
Takeaway: watch the burn address, not the press release
So here is where I land. Hyperliquid is very likely winning, and the architecture is genuinely interesting — an on-chain order book on a purpose-built L1 is a real answer to a real problem. But the scoreboard in this particular bulletin is being kept by the home team.
The next time someone quotes the $1.26 billion, ask one question: show me the burn address. If repurchased HYPE is verifiably destroyed, the flywheel is real and the number means what it implies. If it sits in a treasury multisig, the "deflation" is a photograph of deflation. Watch net float, not gross buyback. Watch the validator set, not the market share. Code is law, but trust is the currency — and right now, we are being asked to spend a great deal of it on a number with no chain attached.
