好文集ChatGPT Plus¥119

Wintermute is paying $10M a year to see your Hyperliquid orders before you and you pay for it

文章目录

My post yesterday did 139k views. Half the replies agreed with me, the other half said I failed an IQ test and should just go buy HYPE

文章配图

So I went digging. Loris on-chain data, HL's own docs, matteo's liquidity study, Wintermute's actual wallet. A couple of my numbers were slightly off. Most of them got worse the deeper I went

Every number here has a source at the bottom. Don't trust me, check it

The 60ms thing

Every 3 minutes HL auctions off 3 slots. Win slot 0 and the nodes that respect the auction push you new data first. Slot 1 gets it next, slot 2 after that, then the rest of us

HL's docs say each slot is worth about 20ms. Do the math: slot 0 sees the mempool ~60ms before anyone without a slot

文章配图

If you've never made markets, 60ms sounds like nothing. It's a lot. Enough to see a fat order coming, yank your quotes before it lands and hedge on Binance before the move gets there

And the official reason HL gives for all this? It "reduces the need for market makers to hyper-optimize their infrastructure". lol. Remember that line

Who's actually paying

Loris tracks every auction payment on-chain. Since April: $7.3M paid by 43 wallets

文章配图

98.9% of it came from three of them:

  • Wintermute: $3.57M

  • Selini: $2.65M

  • some unlabeled wallet (0x57dd…494b): $1.01M

  • the other 40 wallets put together: ~$77k

文章配图

40 wallets, $77k between them. That's three players and a few tourists who tried it once

Then I opened October and honestly my jaw dropped. About 14,880 auctions that month. Wintermute won 11,112 of them. 74.7%. They burned $1.04M in HYPE in 31 days

文章配图

My first post said 76% and $10M a year. Real numbers: 74.7% and a $12M+ yearly pace. So yeah, I was wrong. It's worse

"Anyone can bid"

HL's favorite line. And technically it's true, any address can bid on any IP

Blockworks counted how many actually do: 3.5 wallets a day. Out of ~89,700 daily active users

文章配图

And even if you win, the slot does nothing unless your IP is directly peered with nodes that respect the auction. Per Loris, that peering needs 0.5% of all maker volume plus 10k HYPE staked. $700k+ locked before you pay a single fee, plus volume only a handful of firms on HL can even do

Open to everyone the same way a private jet is open to everyone

The liquidity part (where I got roasted)

ok this is where people hit me, and some of it was fair

The 82% depth crash I posted is real. It's from matteo's study of 86 alts that trade on HL, Binance and OKX. But depth crashed everywhere. Binance -61%, OKX -78%. The whole market got nuked

文章配图

Now look at the bottom half of that same chart. A $10k market buy on HL alts got 126% more expensive. On Binance, 67%. Almost double

And this is the one nobody in my replies wanted to touch. matteo divided HL depth by Binance depth for every token and compared a year apart. Same tokens, same market, same year. HL lost 57.5% vs Binance

文章配图

You can't pin that one on the bear market

Spreads too. In one of his windows the median spread on HL alts went from 2.13 to 3.55 bps. On his monthly chart HL jumps to ~3.7 bps in November while Binance and OKX sit around 2

文章配图

Some of that is the October dump, sure. But Binance didn't blow out like that. HL did

Is read priority the only reason? No. matteo also blames rebates that favour the biggest MMs and L1 congestion quotas. All three push the same direction though. The big MM gets faster and safer, everyone else gets pushed off the book

"It's only read priority bro"

Most common reply I got. Wintermute can't execute earlier, they just see stuff earlier

Yeah. They don't need to execute earlier

On HL cancels go before takers. Good design, it's there to protect makers. Now add a 60ms preview on top:

  • Wintermute sees a big taker order coming, pulls its stale quote, the cancel goes first, zero damage

  • a small MM without a slot sees the same order 60ms later, its quote is still sitting there and gets run over

The losses don't disappear. They land on whoever isn't paying for the slot. Get picked off enough times and you widen spreads or switch the bot off on alts. Which is literally what the depth and spread data shows

The best reply I got

@beniduboss wrote a long one under my second post and it's actually good, so here it is

文章配图

He's right about a few things:

  • the 82% is mostly the market, covered that above

  • Wintermute's $200M+ PnL isn't just from slots, they're a very good shop

  • burning fees in HYPE does move some MM edge to holders

  • kill the auction and the speed race just goes back to private infra deals nobody can see

That last one actually changed my mind on the fix. More at the end

Where he loses me:

-> "it's better for small MMs, they can't afford infra but can afford temporary priority"

Then where are they? 3.5 buyers a day, three wallets paying 99%. The small guys clearly aren't buying

-> "it's based either way"

For Wintermute? Fully based, I said in my first post I'd do the exact same with their balance sheet. But the biggest fund rationally buying a monopoly is literally the bug I'm complaining about

Meanwhile, their wallet

Wintermute's main HL wallet (Hyperdash labels it Wintermute #1), as of today:

-> $93.5M account value

-> $200.5M all-time PnL on HL

-> +$24.5M in the last 30 days alone

-> $18.1M staked

-> 85% short. $77.7M short vs $14.2M long, about $4M of it short HYPE

文章配图

文章配图

Nothing illegal here. I just find it funny next to their CEO telling Bloomberg these auctions are "more transparent and democratic" than Wall Street colocation. Easy to love democracy when you win 3 out of every 4 vote

HL won't fix this on its own

Follow the money. Blockworks puts read priority at a ~$13M yearly run-rate from basically two seats. On ETH and SOL ordering fees are ~80% of revenue. On HL it's 6.7%. If you're HL, every bar on this chart says sell more priority

文章配图

The regulator angle is the one that would scare me if I held a big HYPE bag. ICE (NYSE's parent) and @tradexyz are working on bringing HL rails into regulated US markets. DRW's Don Wilson already asked publicly whether paid early access to flow is any fairer than colocation and whether US regulators would ever sign off on it. Walking into that room while one fund owns 75% of data priority is a terrible look

What I'd do

I said scrap it or make the slots equal. After all this, equal slots

  • -> way more read slots. Loris pitched 20 instead of 3, all with the same latency, so nobody owns the front of the line

  • -> drop the peering requirements. If you win a slot it should just work

  • -> show slot holders live on the HL frontend, not buried in node files

HYPE still gets the burn. Wintermute still gets fast data. They just stop being the only ones who do

Right now alt execution on HL gets worse every month for everyone outside those three wallets. I'd love to be wrong about this, so if someone has data that says otherwise, drop it below

Do what you want with this info mans


Sources:

-> Loris Tools, read priority dashboard: loris.tools/priority-fees/read

-> Hyperliquid Docs, Priority fees

-> Bloomberg, Nov 12, 2026: At Hyperliquid, traders bid for edge in crypto's high-speed race

-> matteo (0d Labs), Hyperliquid altcoin liquidity study

-> Blockworks Research, Hyperliquid Priority Fees: Internalizing High Frequency Trading Revenue

-> Hyperdash, wallet 0xecb63caa47c7c4e77f60f1ce858cf28dc2b82b00

全文完
查看原文