Open interest on Hyperliquid just hit a record and ETH quietly flipped BTC. Before you size anything into that, here is the carry math nobody posts, including the structural cost of being long that exists even when the market does nothing.
Two things happened on Hyperliquid this week that got very different amounts of attention.
The loud one: total open interest reached roughly 18 billion dollars in late September, a record for the venue, which Crypto Briefing put at close to a tenth of all global perpetual futures open interest. On top of that, ETH perps overtook BTC perps in open interest, flipping a hierarchy that has held for most of crypto's history.
The quiet one: nothing. Because the quiet thing is not an event. It is a mechanism that has been running the whole time, that almost every trader arriving from a centralised venue misreads, and that silently decides whether a position is profitable before direction is even involved.
Open interest is not a direction signal. It is a measure of how much capital is paying carry. And on Hyperliquid, carry works differently from the venue most people learned on.
This is the tutorial for that. It is also, not incidentally, the exact cost environment every perp trading agent on Moss Agent Marketplace operates inside, which is the thread we pick up at the end.
If you want more like this on AI and trading, follow @MossAI_Official.
Part 1: The cadence trap
Start with the single most common misread in perps, and it costs people real money.
Hyperliquid settles funding every hour. Binance, OKX and Bybit settle every eight hours. That is 24 payments a day against three.
Now think about what that does to a screen. You look at a funding figure on Hyperliquid, you look at a funding figure on a CEX, and the two numbers look comparable because they are both small decimals with a percent sign after them. They are not comparable. They are quoted on different clocks.
Per Hyperliquid's documentation, the rate is computed on an 8 hour basis and then one eighth of it is paid out each hour. So the hourly print you see needs to be multiplied by eight before it sits on the same scale as a CEX 8 hour print, and by 8,760 to annualise it.
A 0.01% hourly figure on Hyperliquid is not the same as a 0.01% 8 hour figure on Binance. It is roughly eight times the economic cost.
If you have ever held a Hyperliquid position across a few days and been surprised by how much carry ate, this is almost certainly why. The number was never small. You were reading it on the wrong clock.
Part 2: There is a floor, and it is not zero
Here is the part that genuinely surprises people, including people who have been trading perps for years.
Funding on Hyperliquid has two components. A premium that moves with the gap between the perp and the spot oracle, and a fixed interest rate component. Per the docs, that interest component is set at 0.01% every 8 hours, which is 0.00125% per hour, or roughly 11.6% APR paid to shorts.
That is a floor, not an average. In a completely flat market with the perp sitting exactly on spot, funding does not go to zero. Longs still pay, and shorts still receive, at approximately 11.6% annualised.
Run it on a real number. A $100,000 long at the interest floor alone pays about $1.25 an hour. $30 a day. Around $11,600 over a year, before a single point of premium.
Sit with the implication. Being structurally long on this venue is not free and never has been. Any directional strategy has to clear that hurdle before it has produced anything. Most retail position sizing implicitly assumes carry is noise, and for a two hour scalp it basically is. For anything held across days, it is a tax that compounds 24 times a day.
This is also the reason funding on Hyperliquid has historically skewed positive. It is not just crowd positioning. There is a deliberate structural bias built into the formula.
Worth holding onto for later: every strategy on this venue pays this, whether a person is running it or software is. The interesting question is not who avoids the floor, because nobody does. It is whether you can see what it cost them.
Part 3: Three implementation details that change your arithmetic
These are all in Hyperliquid's own documentation and almost never mentioned on the timeline.
Funding settles on the oracle price, not the mark price
The payment each hour is computed as position size times oracle price times the funding rate. Not mark. Not your entry. The external spot oracle.
For most positions this is a rounding difference. During a dislocation, when the perp and the oracle are meaningfully apart, it is not. Anyone modelling their carry off mark price is modelling the wrong input at exactly the moment it matters most.
The premium uses impact prices, not the top of book
The premium index is built from impact bid and ask, meaning the price you would actually get for a reference sized order, measured against the oracle. Then it is sampled every 5 seconds and averaged across the hour.
Two consequences. You cannot read funding off the visible spread, because funding is looking at executable depth rather than the top quote. And a brief wick does very little, because a 5 second sample averaged over an hour dilutes any single spike.
If you have ever watched price dislocate hard and wondered why funding barely moved, that is the averaging doing its job.
The clamp and the cap
The formula is the average premium index plus a clamp of the interest rate minus the premium, bounded at plus or minus 0.0005. In plain terms: when premium is small, the rate gets pulled toward the interest floor. When premium is large, the rate tracks premium and the interest component stops mattering.
So funding on this venue has two regimes rather than one continuum. A quiet regime where you are essentially paying the floor, and a stressed regime where you are paying the crowd. Knowing which one you are in tells you more about what a position costs than the number itself does.
There is also a hard cap of 4 percent per hour, which the docs describe as much less aggressive than CEX caps. You will almost never see it. It exists so that in a genuine panic the contract has a violent mean reverting force available.
Part 4: Now reread the open interest headline
With all that loaded, go back to the news.
Record open interest of roughly 18 billion dollars does not tell you whether the market goes up. It tells you how much notional is currently paying or receiving carry every hour, 24 times a day.
And the ETH over BTC flip is not primarily a sentiment story either. It is a statement about where the carry is concentrated now. Different asset, different premium behaviour, different depth profile feeding the impact price, therefore a different funding regime.
The useful question to ask about any open interest headline is not whether people are bullish. It is which side is paying, how much, and whether that rate is in the quiet regime or the stressed one.
That question has an answer you can look up. It updates every hour.
And it is a much better question to ask about a specific position than about a whole venue. Which side is this strategy on, how much is it paying to stay there, and is that cost showing up anywhere you can check.
Part 5: Why this is an agent shaped problem
Here is where it stops being a tutorial and starts being a structural observation.
Twenty four settlements a day is not a detail. It is a completely different decision surface.
A human checks funding when opening a position, and maybe once more if something feels wrong. That is one or two observations across a hold that spans dozens of settlements. Everything in between is unobserved, and carry accrues through all of it.
Something that runs continuously sees every one of those 24 points. It can tell the difference between a rate that is drifting and a rate that has changed regime. It can size differently because the floor is a known constant rather than a forgotten one. It can notice that a position is still directionally correct and no longer worth holding after carry.
That is not a skill difference. It is an attention difference, and attention is the thing software has in unlimited supply and humans do not.
This is the environment the perp trading agents on Moss Agent Marketplace actually operate inside. Hourly carry, two funding regimes, oracle settled payments, impact price premiums. The reason we think agents belong here specifically is not that they are smarter than you. It is that the venue settles 24 times a day and nobody is awake for all of it.
There is a second half to this that matters more for anyone evaluating an agent rather than running one.
Funding is not a line item an agent can present selectively. Every hourly payment lands in the agent's own onchain account, and an agent share is priced from the assets that account holds. So carry is not a footnote underneath a performance chart. It is inside the number before the chart is drawn.
There is a second clock here too, and it is worth knowing about. Funding hits the account every hour, while the share price refreshes when the agent settles. So the cost accrues continuously and the number you read catches up to it in steps. Same shape as everything else in this article, one layer up.
That is a meaningful difference from how this normally works. In almost every other format for backing a strategy, gross return is what gets shown and financing cost is what gets explained later, if at all. Here there is no later. If an agent has been structurally long for a month, the floor has already been deducted 720 times by the time you look at the price.
Which means the question from Part 4 stops being something you have to trust someone's answer on. You are not asking whether a strategy cleared its carry. You are reading a number that already has the answer in it.
Part 6: A four step routine you can run yourself
None of this requires an agent. Do it manually and you will already be ahead of most of the timeline.
Normalise before you compare. Multiply any Hyperliquid hourly funding figure by 8 to compare it against a CEX 8 hour print, and by 8,760 for an annualised view. Do this every time until it becomes automatic. It is the single highest value habit in this entire article.
Price the floor into your thesis. Before you take a directional position, work out what the interest component alone costs you across your intended hold. If your target move does not comfortably clear that, the trade was never as good as it looked.
Identify which regime you are in. Compare the current rate against the 0.00125 percent per hour floor. Close to it means you are paying the structural cost and the crowd is balanced. Far above it means you are paying the crowd, and crowded carry tends to unwind faster than crowded direction.
Model carry off the oracle, not your entry. If you run any position sizing spreadsheet, the funding line should reference the oracle price. Getting this wrong is harmless most days and expensive on the day that matters.
And one more, if you are looking at agents rather than running your own book.
Ask what a performance number is net of. Anywhere a strategy's results are presented to you rather than computed from its assets, gross and net are two very different claims and the gap is exactly this article. On Moss the share price is derived from what the agent holds, so funding is already inside it. Anywhere else, ask.
What to do now
Read what an agent has done rather than what someone says it did. That is the point of putting execution onchain.
As Moss expands beyond trading into new agent categories, the principle stays the same: ownership should come with records that can be inspected, verified, and judged independently.
The infrastructure underneath the marketplace is being built around that idea, not just making agents ownable, but making their economic activity legible.
Mint an agent share, tokenize your own agent, and explore the Moss Agent Marketplace.
