Honestly, I sold most of my spot crypto and offboarded today. It's just not worth the headache.
I don't want to check every single headline about AI math and cryptography research wondering if I'm about to get rugged. I don't want to wonder if every ancient BTC address making a transfer is ECDSA being cracked.
I had already refocused most of my time and energy on equities at this point, and Justin Drake's post just killed the rest of my desire to hold crypto. This isn't impulsive, I read hours of counterarguments.
1. People say this is incredibly unlikely. I just don't trust any human to predict the capability ceiling of superhuman mathematical AI. Have seen people be wrong about AI capability ceiling over and over.
2. What if I miss the next great bull cycle? Honestly, that's fine, there's plenty of other assets to trade with less stress of existential left tail risk.
3. People say that we would have bigger problems. Yes, I know that banks etc would also be at risk and society would be a clusterfuck, but the government should step in to save my money (and society) at that point. I don't know that they will for crypto. Plus, why would a malicious practitioner rather burn the whole world down over just stealing infinite crypto?
4. People say this is just like Quantum FUD. Nah, this is much more concrete and near term. This doesn't require a new scientific breakthrough, only the continued march of AI progress that we've been scaling on for years.
I got into crypto as a techno optimist but there's other investment expressions of that now. The development of AI (and recently the intersection of bio x AI) is more intellectually interesting to me. Being in crypto feels like being short AI progress, and I do not want to be short AI progress.
This is not investment advice, everyone should decide for themselves how they feel about the topic. I am just one random human being on the internet, expressing my thoughts.
btw i think you may be arguing against a chatgpt X bot but for the sake of continuing the conversation...
> Wintermute is and has always been a net positive for crypto
i broadly agree with this. wintermute is synonymous with functioning crypto markets. 'dont hate the player, hate the game'. if anyone blames them for their bags going down, maybe they should ask if their bags ever had intrinsic value in the first place.
> First of all it’s a read priority, which means that they pay for the access of incoming tx. They don’t get the ability to execute earlier than others
ok. so, the crux of the issue is not about picking others off on HL. the problem is this: they get to see when their orders will be filled, first - always, 50ms (60ms now) ahead of anyone else on hyperliquid, and then HEDGE before anyone else on other exchanges.
in doing so, they can size up to the point where they squeeze every last drop of juice out of smaller markets on HL, by consuming ALL of the profitable hedging liquidity on binance (et al).
the only reason they WOULDN'T do this is if they were being charitable and letting others take some pnl (or maxed out on position size). and wintermute is not a charity.
if there's only $100,000 of profitable hedging liquidity for UNI perps out to 50bps on binance+bybit+okx, then they can just quote that full amount / depth knowing they can consume ALL of the hedge liquidity, 100% of the time (simplifying a bit)
this means alts converge to having 1 market maker. which is structurally - and this is not debatable - bad. they could even quote wider than other MMs and still hedge ahead of them because they get to see their fills that much earlier. 60ms is an eternity in HFT. round trips between tokyo boxes is under 1ms, so that's 60+ round trips worth of network time. while everyone is listening to a dial tone.
so the actual question is, are we ok with having only a single market maker quoting markets on hyperliquid?
i would strongly argue no, we are not. it's not democratic, it's not sound microstructure. it's bad for hyperliquid, and it's bad for crypto.
> The 227M PnL is irrelevant because it was generated by a ton of alpha instead of just them having most of the priority slots
agree. also, for the sake of argument, they might have lost 228M on their hedges against these positions. they didn't, and the original number is probably not that far off, but you can't make any profitability claims based on their account history on a single venue. also wintermute is (mostly) not wearing naked HL positions w/o hedges.
> Why didn’t you mention on your other tweet that all crypto got nuked? You blame WM and HL for the 82% liq collapse but you don’t see you mentioning that nance is also down like 60%+ on that same period just like every other exchange lol
that's still a big difference, but causality here is difficult to prove. factually, i can tell you we can see binance's hedges hitting the binance tapes 60ms+ ahead of ours, and size-match it to their HL wallets fills. this is what is driving smaller MMs out.
we have recently cut our quoted symbols in half on HL from 120 to 60.
> It literally converts MMs trading edge in fees
no, that's not how it works. as mentioned, they could even quote *wider* than they otherwise would, because they have guaranteed hedge liquidity.
also, it's similar to how 0 fees doesn't mean 0 spreads. market makers do not just eat fees, they widen spreads according to costs.
there are shades of gray here, it's not binary, and i am open to a multiple equal-speed slot architecture with open, easy peering access for any slot winning wallet. that was my first recommended redesign in fact. but that may not be possible given the network/architecture.
> Abolishing the auction doesn’t change anything in fact it might actually make it worse. Do you know how much funds spend on infra every year? Take it away and now have the same exact thing happening but behind closed doors and with fewer participants
no. our all-in HL-related infra bill was around $15,000 last month, and i know for a fact we are within 1ms of wintermute's HL latency, absent read-priority.
wintermute spent $828,676.86 on read priority in the past 30d. this fact is available from hl core chain data.
$828,676.86 > $15,000. could argue the point on engineering, but there's not that much engineering required once everything is in place.
furthermore i doubt they reduced their infra costs with read priority, but you'd have to ask them
> The auction is better for the smaller MMs because they can’t afford the infra cost to race for ms, but they can afford to buy temporary priority
no. this was the stated intention, but i struggle to see how they didn't see the obvious outcome of the deepest pocketed player dominating the auction. which makes me think this was intended as a v0 that just hasn't been updated yet. so now it's time to update or simplify (abolish).
> I never personally worked in HFT/MM but Id argue that since it massively helps against adverse selection it also allows MMs to quote with much deeper liq at top of book and sometimes maybe even tighter spreads
the incentive could also be the opposite, as stated above. my observed behavior, as an HFT MM competing directly with them, staring at this stuff 90 hours per week, is that alt markets are and will continue to converge to having a single market maker because nobody else can quote profitably. if wintermute blows up, or decides to stop quoting, or their systems take a dump for 12 hours on a wednesday, those markets go quiet. bad.
> 1. WM is buying 76% of slots because they’re better than everyone else
> 2. They’re buying it so others don’t have access to it
> Sorry to say but in both cases it’s based af lmao
i more or less agree. they're just playing the game, and winning. i would again like to state unequivocally, wintermute is NOT the problem here! they're just playing the game and winning.
but if we're going to house all of finance the game cannot be designed in this way. it has serious risks, and it is not good enough. i've spoken to many smart people about this and they all agree. respectfully, the only one who disagrees is the one who's winning the game.
hyperliquid should stick to its day-1 values and be a democratic, open ecosystem that is NOT designed for the few. it should be designed to support the deepest liquidity, with the most participants, to serve a global market.
and under the current design, it's simply not so.
New essay out on the @multicoin blog, which is an evolution of our thinking on DeFi value capture
TLDR is in a world where software costs trend to zero, finance is one of the most obvious industries where moats will exist and DeFi is perfectly positioned https://t.co/iXBtFnZndw
best x accounts on hft/mm/arbitrage
yesterday i asked you guys to share, now it’s my turn
and yeah, no ranking here, just random order
might sound weird but first i wanna shout out jeff @chameleon_jeff. before the $HYPE tge he was writing insanely good threads on hft, and reading that from the guy who built a multi-billion dollar business is never a waste of time. for example, here’s his thread on why you should use rust for hft - https://t.co/2dS3WbNvAu
@LumaoDoggie - you can tell right away this guy knows his stuff. seen a few posts from him with optimization tricks, like how to shave tens of ms on hl with non-obvious stuff, example - https://t.co/KeYRFnD491
@0xLUKE777crypt - literally a whole series on getting into market making, part one here - https://t.co/DGFrj6wu0k
@Tristan0x - guy from the bullet perp dex team, and he pulled the most gigachad move possible for growing a project: mm’d 10k up to 16k and started a series on how to mm on bulletxyz, part one - https://t.co/qjlsbRLFsS
@PlanemoTrading - been following this guy for a while. he’s basically got a treadfi-style saas but for stat arb, and sometimes drops useful arb posts. recently showed ~1k profit in a week off 10k - https://t.co/zZUmbcBeOd
@0xpsyche1 - great guy, found him through his post comparing latency across different hosting providers - https://t.co/ALW9oDe3oX
@annanay - ex-hft at major firms, now founder at qfex, shares different ideas from time to time. if you’re new to mm, highly recommend his almost hour-long podcast intro to the topic - https://t.co/PoHeXDH8h8
@0xevolve - tons of posts on rust/low-level hft, here’s one on config - https://t.co/LAku15UmGP
@supermomonga - cool japanese guy, recently wrote about self-improving software via ml - https://t.co/lXT616mQGh
@yourQuantGuy - insanely talented chinese guy, lots of arb-specific info. funding, prices, and once wrote about an interesting options arb situation - https://t.co/DC0QVtkWNH
@UST_mean - solid chinese guy, from what i get he’s fairly new to hft but posts a lot of useful stuff from personal experience, e.g. rust vs python for latency - https://t.co/pObdqibFYF
@Boywus - very strong chinese guy, tons of info on hft, market making, adverse selection, latency, lead-lag arb. here’s a detailed breakdown of anthropic pre-ipo and the spreads between markets - https://t.co/he6Y2Q6z07
@ApriloneHS - writes about his arb experience with actual numbers, e.g. results from farming arcus <> lighter rh arb - https://t.co/vsQfXgOlm6
@StepOneAi - straight up a goldmine of hft info, wrote a massive series, absolute gem. part one of his “high-frequency infrastructure” series - https://t.co/mhCSURvBS9
@0xDC_MA - super useful chinese account, not really about auto arb, more manual stuff, but he posts a ton of his experience and pretty much every arb trade he takes. here’s one on a ~5% price spread - https://t.co/oh3Y4ewblO
@JXiaoLoong - kinda similar to the one above, also chinese and also mostly manual arb. picked up a few tradfi arb ideas from him, some food for thought -
mic drop
@HyperliquidX's IOC speed-bump mechanism has a flaw. Normally designed to protect makers in the interests of a more liquid market, it breaks down in periods of high activity. When blocks become full, pending modifies and cancels queue up for later blocks, so their latency increases. However, IOCs behave as though they have a separate blockspace dedicated for them, and their latency barely budges. You can see this in the lower left chart, showing latency of different order types during 19 activity bursts.
The result is that under relatively modest load (25-30k tps), IOC + 8 bps (priority fee) orders can pick off resting liquidity before it has a chance to get out the way. Meanwhile, under higher load (35-40k tps), even IOCs without any priority fee can pick off resting liquidity.
This, in my opinion, is antithetical to the ethos and design of the ALO and cancel prioritization mechanism and needs investigation by the team. It exists to reduce toxicity from HFT takers, and is breaking down precisely when most needed.
It may also be exploitable. If IOCs really are insulated from the congestion affecting cancels and modifies, a malicious user could spam orders, delaying makers’ cancels while their own IOCs go through, and taking the liquidity left stale.
One single example that occurred on AVAX today caused 1.3M$ of volume to be adversely selected for a total loss of ~13k$ over the subsequent 10 seconds. This affected us materially. Faced with such episodes of adverse selection, spreads and liquidity are undoubtedly worse than they could be otherwise.
Would appreciate if anyone with a contact at HL could share this with the team.
Inspired by recent discussion about whether it's possible to learn trading outside an elite firm.
If you're the kind of person that reads this project and thinks it sounds fun, imo you'll be able to learn employable trading skills just fine on your own. https://t.co/czZMfucmpT
I built a Barra factor model to see what drives crypto perp returns. Out of sample since 2022, includes: what traits tended to make money, how long the effects last, and how much of a coin's move is coin-specific versus systemic https://t.co/IyHQuLSOvZ
Thoughts from my Shanghai field trip
Sentiment among Greater China investors was far worse than I expected. Unlike the US market, A shares had not seen a single rebound since July. Still holding Chinese and Hong Kong AI stocks that had crashed and Korean memory stocks that had collapsed, investors kept pressing me on when memory would bounce back.
Many investors said the AI trade was over, and a few said they doubted whether even 2027 capex, not just 2028, would go ahead as planned.
Overall, the bearish sentiment was so strong that even I, a bull before coming to China, started to waver.
It got to the point where, faced with the bear case that memory prices could start falling in the third quarter of next year, I took it seriously and couldn't refute it.
As the conversations went on, I gradually began to understand where their pessimism was coming from. In the end, they were all asking the same question.
So what comes after AI coding? What will be the next engine of ARR for Anthropic and OpenAI?
Several candidates came up. AI4S, AI chip design, and so on... But none of them clearly convinced investors.
I couldn't offer a clear answer either. Sure, there are rumors that Anthropic has made major breakthroughs in both AI4S and AI chip design, but what we actually want to know is when any of that turns into money.
With AI4S, there could be beneficiaries right away, such as companies that sell lab monkeys or diagnostic kits. In fact, there already are. But while we can trade those names in the short term, what we ultimately want to know is whether Anthropic can make money by developing drugs. That is still several years away.
As for AI chip design, the TAM itself is hard to calculate, and even if you do calculate it, it is hard to monetize. That makes it difficult for the market to believe in.
Honestly, this feels like the hardest market in history. If a company isn't making money, selling makes sense. But right now the market sells even when companies are making money, even when they beat, and even when they say demand will stay solid all the way through 2028 and 2029.
So my recommendation to investors was to buy memory, and within memory, DRAM. Not because I have conviction in the DRAM cycle, but because I see it as a relatively defensive choice within the AI trade.
Memory stocks are now trading close to 1x P/B on 2027 BPS. Historically, memory stocks have staged technical rebounds when they fell below 1x P/B.
And even if the broader market falls further from here, the three DRAM makers will keep buying back their own shares. I thought that would guarantee a certain level of buying and concluded the stocks would have downside support.
Of course, I don't know whether memory stocks can break above their previous highs. That depends on whether sentiment across the broader market recovers. But I think there is a fairly good chance that memory outperforms the rest of the AI trade.