Twitxy Saved

Lihong | Freeport

Verified account Freeport

@Lihong062

Founder @freeportmarkets | Prev Semis Trader @ IMC & Jane Street | Field Notes on AI & Macro
New York, Dubai, Hong Kong freeportlogbook.substack.com Joined February 2022
698 Following 5.4K Followers 2.1K Posts

Latest posts 5

Pinned post

Lihong | Freeport Verified account @Lihong062
Had a great time on the pod with @ethanrkho.

We covered how vol desks actually make bets, why trading firms are the third-largest consumers of AI tokens, and what I'm building @freeportmarkets.

Apparently I also admitted that I was "down an entire Ferrari" on the PA June 22nd on record lmao.
Ethan Kho Verified account @ethanrkho Sep 17
Ex-IMC semiconductor options trader Lihong Wang on seeing the whole market's flow, betting the entire AI stack, & why foundation models are good enough to beat the S&P 500.

Lihong Wang (@Lihong062) | Discretionary semiconductor options trader @ IMC Trading | Now co-founder of Freeport, a YC-backed event-driven perps exchange

"You see what everyone's doing, you trade against the people who are stupid, and you get out of the way or you follow the people who are smart."

We cover:
- How a bet actually gets made on a vol desk: not predicting the future, just selling what's expensive and buying what's cheap
- The edge of the seat: being counterparty to 20-40% of flow in certain options markets, a view "maybe a few dozen people on the entire planet" have
- Knowing who HAS to trade: reverse-engineering banks' structured-product hedging from issuance data
- How desks blow up: overestimating correlation, and why NVDA -17% on DeepSeek didn't mean AMD -12%
- The trader group chats: ~3x beta to the S&P, median 30% drawdown in July, and everyone still bullish
- His portfolio: 50 chip/AI stocks covering the entire stack, up 470% bottom-to-peak, then a 70% drawdown ("I'm literally down an entire Ferrari today")
- Why trading firms may be the third-largest consumers of AI tokens on the planet
- Moats vs. growth: why Cursor got funded with no moat, and what that means for careers in the age of AI
- His hot take: foundation models can already beat the S&P 500 risk-adjusted. The model isn't the hard part, the harness is

Highlights:
00:00 Intro
01:09 How discretionary options bets get made — and blow up
10:23 A message from Onyx
10:53 Warehousing benign flow through July's deleveraging
16:40 Seat leverage and how trading firms allocate talent
22:03 Why quant traders are levered long the AI stack
26:27 Playing the AI hand: leverage, moats, personal brand
36:42 From IMC to Freeport: AI agents for event-driven trading
45:26 Perp DEX endgame: liquidity, fragmentation, and SEC rulemaking
51:21 Sourcing semiconductor alpha from trader dinners
56:05 Narrative edge, DeepSeek puts, and a 70% drawdown
01:04:24 Is AI a bubble? Compounding, path, and Kelly leverage
01:08:39 US, China, and comparative advantage under ASI
01:12:17 Risk-taking when you're post-economic
View quoted post on X
9 replies 5 reposts 75 likes 54K views View on X
Lihong | Freeport Verified account @Lihong062
Beijing and Washington both clearly care about AI.

But neither seems to really grasp how impactful it could be, since neither really behaves as though there is a material probability of imminent RSI leading to ASI.

The best working timeline for a serious recursive self-improvement loop is 8-16 months.

If you believe first-to-ASI-wins, it’s implications are much stronger than either government’s behavior.

Once AI can automate and accelerate AI research, the strategic value of having more human engineers (China) goes to zero.

American operational plus planned data-center capacity is more the next fourteen countries combined (see chart). Conditional on hitting RSI first, it wins the AI race.

If Beijing genuinely believed an American lab was months away from a decisive, irreversible advantage, you would expect it to pay extraordinary costs to delay that.

Massive treasury sales to pressure financing, oil stockpiling to complicate easing, export restrictions on critical materials, and conflict over Taiwan / Fabs.

However, it’s not doing that, and its overall posture looks far from treating the next year as the final contest for global power.

That could mean Beijing assigns a low probability to imminent ASI, doubts that its actions can help it get there first anyway, believes confrontation would worsen its position, or for other reasons is accepting of American hegemony.

The implication for those in Washington is almost the reverse.

If admin believes that the US has a substantial lead AND recursive self-improvement is imminent, “we must accelerate because China” becomes a terrible argument.

The relevant comparison is the marginal risk of China catching up against the marginal risk of an American lab building something it cannot control.

A sufficiently large lead justifies a coordinated, verifiable slowdown at the frontier to improve alignment.

There’s a related implication for the labs, in that if OpenAI believes recursive self-improvement is imminent and strategically decisive, its commercial roadmap should be subordinate to that objective:

concentrate resources on the research loop, alignment, and securing enough compute to finish.

That’s partly why I’m increasingly bullish on Anthropic’s chances.

Its disclosed cash and short-term investments are enormous ($132bn) and has a 20x over subscribed IPO,
while OpenAI had to downsize their most recent round.

I’m arguing against my own book here since I still own some AI infra.

My portfolio drops 10% when Dario merely says “pace the frontier,” so I would hate to see him actually do it.

In any case, I don’t expect governments to internalize these implications.

In practice, I expect the fate of humanity to depend on the judgment of our overlords at Anthropic.

However, I’m less pessimistic than you may think, since conversations with friends inside labs have made me more optimistic about progress on alignment and interpretability.

My personal p(doom) is therefor only around 15% right now, which is low by most AI-risk standards.

What does somewhat bother me is the gap between the extraordinary implications people claim to believe and the remarkably ordinary behavior that follows.
AI Notkilleveryoneism Memes ⏸️ Verified account @AISafetyMemes 2h
You have 11 months left.
View quoted post on X
Lihong | Freeport Verified account @Lihong062
You don’t get extraordinary founders by selecting for obedience and a "culture of compliance."

@paulg called one of the TOP qualities @ycombinator looks for “naughtiness,” as in founders who care about getting the big questions right but have little reverence for convention or rules.

They instinctively question constraints and find possibilities to hack and game systems (software, legal, social, and others).

That disposition is part of how you build something others think is impossible, but also doesn’t conveniently switch off when the obstacle becomes government policy.

Trade restrictions (like all restrictions on freedom) collide directly with that instinct. A founder sees customers who want a product, engineers who can improve it, and a market they spent decades building.

Of course they’re going to question a policy that tells them to abandon it, of course they’ll explore every legally available way to keep serving it, the determination and resourcefulness people admire in the company’s origin story are still there.

We celebrate founders for refusing to accept “that’s just how things work.” We shouldn’t be surprised when they bring that same attitude to Washington.

None of this settles whether a particular policy is right or wrong, but if you’re confused by SV's enthusiasm for Jensen, this is a big part of it: a deep distrust of anyone who claims to know better than the market and uses coercion to override voluntary exchange.

Builders tend to see trade as a way for both sides to gain and they’re naturally skeptical of a worldview that treats someone else’s progress as our loss.
Philippe Lemoine Verified account @phl43 3d
This is exactly how most China hawks think about export controls or indeed sanctions in general. They believe that, once sanctions have been imposed, anyone who opposes them is basically some kind of traitor.

Similarly, as this paragraph from last month's Bloomberg story on chip smuggling illustrates, they don't think it's enough for tech companies to follow the law, they think it's a problem that they don't share their views on sanctions and therefore don't do so enthusiastically.
View quoted post on X