This time, I’m the one getting interviewed. I joined Ethan Kho on Odds on Open to talk about what I actually did trading semiconductor options at IMC, why my personal portfolio looks very different from an institutional trading book, and why I left to build Freeport Markets.
Ethan hosts conversations with traders, portfolio managers, and founders about where their edge comes from. He was particularly good at pressing on the distance between how I think about risk professionally and how I take it personally. At one point, I describe being down an entire Ferrari in a day in the PA.
The 74-minute conversation covers how structured products create predictable flows, why a hedged options book can still blow up, where the AI supply chain captures value, how information becomes a market narrative, and what young people should be doing with their careers when the tools are changing this quickly.
What Options Traders Bet On
Ethan’s first question was how a bet gets made at a firm like IMC. The word “bet” itself needs some unpacking. When you buy a prediction-market contract, you usually have a view about something that has not happened yet. At our desk, a more useful starting point was the current state of the market: what is expensive, what is cheap, and why someone is willing to trade with us at those prices.
For an options trader, cheap and expensive often refer to volatility. Once you hedge an option’s directional exposure by trading the underlying stock, a large part of what remains is exposure to how much the stock moves relative to what the option price implied. Buying a call does not mean expressing the view that stock is going up.
Take Nvidia and AMD. Suppose Nvidia’s implied volatility is 50% above its historical realized volatility, while AMD’s is 70% above. Those are hypothetical relative premiums, not volatility levels. The difference alone does not establish a trade, but alongside other signals it might suggest that AMD vol is expensive. You could sell it and buy Nvidia vol to offset some shared semiconductor risk. Or, if Nvidia looks fairly priced, you might simply sell the expensive AMD vol within risk limits.
Usually the work was relative: one stock against another, one expiry against another, one part of the volatility surface against another. We rarely wanted an enormous outright long- or short-volatility position.
Who is Forced to Trade
One of the most useful inputs was flow. A large options market maker sees a meaningful share of the market because it is on the other side of so many trades. In the conversation I estimated that, in certain markets, the share could be something like 20–40%. That visibility is an unusual asset.
Counterparty information needs interpreting. You may know the executing broker without knowing the ultimate client or the reason for the order. Goldman buying an option does not tell you whether the bank is hedging a structured product or executing for a hedge fund with a strong view.
Asian banks’ structured products are a good example. A bank sells a retail client a complicated payoff and then has to hedge the exposure it has created. Issuance data, the terms of the product, and the current stock price can help you estimate what the bank will need to buy or sell.
Imagine a long-dated Nvidia product whose hedge requires the bank to sell two-year volatility as the stock falls. The selling depresses option prices. If the stock then crosses a contractual barrier, the product’s exposure can change abruptly and the bank may need to buy that volatility back. The precise directions depend on the product, but the important feature is that the flow is driven by a contractual obligation.
That gives a trader something to work with. If issuance data points to that exposure, the relevant selling appears, and the price looks attractive, you might buy. If the barrier is crossed and the hedge reverses, you might sell into the buying.
I jokingly described the job as trading against the stupid people and following the smart ones. In practice, a very smart counterparty may simply have to execute a trade for reasons unrelated to whether its price is attractive to you. Recognizing that constraint is often the edge. It also explains why discretion still matters: the data gives you a plausible explanation for the flow, not certainty about the person behind it.
You Want to Keep the Risk
The standard picture of a market maker is someone buying, selling, and collecting a spread while staying close to flat. Across an options business, making that happen requires a large risk-management system.
Each trade brings delta, gamma, vega, theta, and other exposures into the book. The firm aggregates them across instruments, nets what it can, and adjusts prices to encourage trades that reduce the risks left over. Eventually, unwanted risk is externalized: someone else is paid to take it.
Our desk could make a different judgment about particular trades. If the price was good and the flow looked benign, we might want to warehouse that exposure rather than let it automatically push the rest of the firm’s quotes toward getting rid of it. We could keep the options, hedge their delta over time, and potentially hold them to expiry. If more of the same attractive flow arrived, we could try to get more of it.
The distinction is economically important. The system is trying to manage the firm’s aggregate inventory because accrued inventory contains signal about correct valuation. A discretionary trader can sometimes identify particular risks that the firm should be happy to keep.
That was much closer to our mandate than waking up with an independent forecast for where the entire stock market should go. Other firms or desks have broader mandates, stronger earnings analysts, or more direct OTC relationships with institutions. Those differences change which opportunities they can see and how confidently they can take them.
Correlation Is Where the Hedge Breaks
It is easier to estimate what an option does if its underlying stock moves 20% than to know what every other stock will do alongside it.
Nvidia and Broadcom once traded closely as expressions of the same AI theme. As the market began distinguishing Google’s TPU ecosystem from the Nvidia, Microsoft, and OpenAI ecosystem, the relationship became less stable. Broadcom’s custom ASIC business was suddenly part of a different argument about who would win.
A position can look hedged under the old relationship and carry much more risk under the new one. The contracts have not changed; the market’s understanding of the businesses has.
The first DeepSeek selloff supplied another example. Nvidia fell roughly 17% in a day. Being right about that move would not have told you exactly how far AMD or Broadcom should fall. Once margin calls and forced selling begin, the relative moves can diverge sharply from the relationships used to size the hedge.
July brought a related lesson about flow. Former colleagues described a market in which persistent selling seemed disproportionate to the news. I recalled Intel rallying on a strong earnings release and then selling off again: to traders watching the tape, that looked like someone using good news to find buyers. The observation was a clue about positioning, not proof of the seller’s identity.
Fifty Stocks, One Thesis
My personal book is much more directional. At the time of the interview, I described roughly fifty AI and semiconductor names across the supply chain, with about 2x exposure.
The breadth comes from uncertainty about value capture. Suppose AI creates $10 trillion of value in some future year. How much goes to consumers through cheaper services? How much becomes corporate earnings? Within those earnings, what accrues to the labs, GPU designers, data centers, fabs, equipment makers, memory suppliers, or the companies connecting it all?
You can be very bullish on the technology without knowing the answer.
So the portfolio reaches from Nvidia and the labs’ infrastructure suppliers through CoreWeave, TSMC and Intel, equipment companies such as Lam Research, Applied Materials, and ASML, and optics and connectivity companies such as Lumentum and Coherent. Materials and substrates sit underneath it all. The idea is to own exposure across the stack rather than require one precise forecast about where the profits land. Fifty names still leave me concentrated in a single broad theme.
Many of my trader friends are also heavily exposed to AI. I joked that the average beta in our group chats was around three and the median July drawdown was at least 30%. This is a very selected group of people, and my own concentration and appetite for risk are not representative of everyone at a trading firm.
There is a reason that group is receptive to the thesis. Trading firms have seen AI change their own work: internal tools, research, and deep-learning systems running on substantial compute. In my view, firms that invested early are seeing the benefits, while some that dismissed the technology lost time they could not easily recover.
But experience with a useful technology is still different from knowing what its stocks should cost. Institutional trading offers many repetitions and relatively fast feedback. Long-term investing is a much noisier judgment about whether businesses will turn capital into more capital over years. A convincing thesis can spend a long time looking wrong.
Before a Narrative Reaches Bloomberg
A lot of my research starts with conversations with semiconductor traders and analysts I know, then continues through Twitter and Substack. I think narratives often pass through three stages: conversations among people close to an industry, specialist discussion on Twitter or Substack, and broad financial-media coverage.
At IMC, the information I worked with generally arrived at the Twitter or paid-newswire stage. Some better-connected funds can hear the industry conversation days or weeks earlier. By the time an idea is a familiar Bloomberg or Wall Street Journal story, much of its novelty may already be gone.
Finding the right accounts requires judgment. Does this researcher understand the technology? Is the claim credible? And, separately, is this something other investors are likely to notice and trade?
Those are different forecasts. A claim can be correct and fail to capture attention. It can also capture attention long before anyone has established that it is correct. The Keynesian beauty contest is very much alive in the AI trade. You can build algorithms to predict which small narratives will become large ones, but I do not think that is a solved problem.
Services such as Citrini and SemiAnalysis can be useful partly for their research and partly for their proximity to the conversations that generate it. Understanding how an executive evaluates a data-center contract, for example, may help you assess a Bitcoin miner’s prospects of becoming an AI infrastructure provider. That is a plausible informational advantage without assuming access to a confidential contract.
Down an Entire Ferrari
Ethan also asked how I size these views. I told him I was not sure my personal account qualified me to give a risk-management lecture. By my recollection, the portfolio had risen roughly 470% from a low to its peak, then suffered a drawdown of about 70%. At the time we spoke, it was still around 60% below that peak. During one of the sharp daily declines, I texted a friend that I was down an entire Ferrari.
I do not think “AI is a bubble” is a sufficiently precise objection. Some companies can be overpriced while the technology produces enormous value. A business can be worth owning over a long horizon and still experience a 50–70% drawdown along the way. My bullishness is a view about long-run compounding, not a forecast for a smooth path.
The theoretical argument for my sizing is closer to Kelly: choose exposure with long-run compound growth in mind. Under sufficiently favorable assumptions about returns, volatility, and financing, exposure above 1x can make sense. Change those assumptions and the answer changes.
My own estimate was that roughly 2x could be defensible given my view of AI. Maintaining that exposure can mean selling on the way down and buying on the way up, so the rebalancing path matters too. I also acknowledged that my sizing might be too high. Believing a position has positive expected value does not settle whether you can carry it through the bad outcomes.
The more extreme example was a user I recalled taking an account from roughly $70,000 to $1 million and then back to zero in July. He told me he was taking a break from perps. The screenshot of the peak survived; the capital did not. Other users take longer-term positions, including pre-IPO exposure, but a successful trade does not establish that a particular amount of borrowing was prudent.
A Moat Around Your Career
The same technology raises a personal question: what should you own, and what should you learn to do? Ethan described the two extremes of the debate: AI is a bubble, or people without AI exposure in their careers or portfolios risk becoming a permanent underclass. I do assign a nonzero probability to the second scenario.
One response is to own some of the capital benefiting from AI. If more value accrues to capital, participating in that ownership matters. For a young person, there is also a distinction between current savings and lifetime earning capacity. A graduate with a small portfolio and substantial expected future income has a different economic balance sheet from someone living off accumulated assets.
That was the logic behind my comments about young people taking more investment risk. I used the example of someone entering trading with $50,000–$100,000 saved and a very high expected salary, and argued that even 3–4x exposure could be reasonable relative to that broader balance sheet. It is an aggressive view, and it depends heavily on the reliability of those future earnings and the consequences of losing the savings they already have. A trading offer is not a bond, and an anticipated salary cannot meet a margin call before it has been earned.
The other response is to learn the tools. I do not start from the assumption that greater AI adoption must mean fewer jobs overall; new tasks and businesses can grow alongside the automation. Use AI in the job you have. Automate work, build things, and see what becomes possible. If your environment prevents that, the cost of trying an idea independently has fallen dramatically. The fact that software is easier to build also makes weak ideas harder to excuse: removing the implementation bottleneck does not automatically create demand.
Ethan framed his own podcast as a way to build three assets together: a recognizable brand, relationships, and knowledge from the guests. He wanted to know what someone without such a moat should do.
My answer was that growth sometimes matters more than defensibility. A restaurant competes for a relatively fixed number of meals; differentiation is essential. In a rapidly expanding market, demand can grow fast enough to reward a business before anyone has a satisfying account of its permanent moat. Cursor was the example we discussed. It is possible for AI to improve someone’s productivity, income, and quality of life without giving that person a skill nobody else possesses.
Relationships and reputation still help. I think people often underestimate the return on simply putting their work and thinking on LinkedIn or Twitter. Becoming recognizable does not necessarily require an enormous effort, and it can make subsequent conversations much easier.
In college, I helped lead our quant trading club and organized recruiting events. By application season, I already knew many of the recruiters. I could apply and follow up with someone I had worked with, rather than send another résumé into a queue. I recalled getting first-round interviews at nearly all the firms I approached that way.
The useful part was planning ahead and doing something valuable in the relationship. Sending an application is easy to automate. Having already helped someone run a good event is a different kind of signal.
Building the Analyst Workflow
I enjoyed the work and the people at IMC. What I missed was the scope of responsibility I had felt leading the trading club and serving as president of AEPi for two years in college. There was an itch to build something that a good job did not quite scratch.
Interest in perps grew alongside that. My friends and I were interested in market access, trading outside conventional hours, and the friction around personal trading while working in finance. That demand helped shape the initial idea.
My cofounder and I talked for months, applied to YC, and got in. Then we had about 24 hours to make a decision. That was the moment a general desire to start something became an actual departure.
Ethan’s next question was the obvious one: there are already plenty of perp exchanges. What makes Freeport different?
Our answer is a particular way of trading: events and narratives. Earnings, macro releases, industry developments, tweets, and long-form research are already part of how people make decisions. Freeport connects those sources to the markets someone follows and the positions they can take.
At a trading firm, an analyst might send a morning briefing, then update the desk throughout the day as relevant information arrives. The work includes choosing sources, filtering noise, connecting developments, and deciding what a person running a particular book needs to know. Simply giving the trader more headlines does not perform that job.
That is the workflow we are trying to make more accessible. A user selects markets, gets a feed from sources we have curated, and can add sources of their own. The system then distinguishes time-sensitive events from material worth processing more deeply. A CPI release or a sudden military development belongs on the fast path. A long Citrini piece about the photonics supply chain can go through a larger reasoning model and be broken into several relevant ideas.
Ethan asked whether frontier models are already good enough to trade. My view was that they are plausibly capable of outperforming an average retail trader, and perhaps even the S&P 500 on a risk-adjusted basis, with a sufficiently well-designed system around them.
A Few Exchanges & Many Ways to Trade
On the exchange side, I expect liquidity to concentrate. Traders want to be where the other traders are, and two contracts offering similar Nvidia exposure can compete closely if their terms and risks are comparable.
Traditional markets fragment partly because of jurisdiction and market structure. U.S. equity-market rules made displayed quotes on multiple venues economically relevant. Rule 611’s trade-through protections are one example, though the actual rule is more specific than “every market maker must connect to every exchange”: access can be indirect, and there are exceptions. The SEC’s Rule 611 guidance explains that distinction.
Onchain markets can reduce some sources of fragmentation. They do not make jurisdiction, collateral, settlement rules, or product design irrelevant. My expectation was a relatively small number of large liquidity pools, with much more room for different businesses serving the end user.
Robinhood, Interactive Brokers, Schwab, eTrade, WeBull, Vanguard, and Fidelity all offer access to stocks, but appeal to different customers. The same should be true of perp interfaces and aggregators. Someone investing for years, someone following politicians’ disclosed trades, and someone trading breaking news do not want the same product. Freeport is built around the last of those workflows. That fragmentation may limit the case for one enormous winner, while leaving room for several multibillion-dollar businesses serving different customers.
The regulatory outcome matters considerably. Centralized venues such as Binance still accounted for most perp volume in the market I was describing. In the conversation I framed clearer U.S. rules as a potential accelerator for decentralized venues; an unfavorable outcome could keep them from reaching mainstream U.S. adoption in the near term.
Does Everyone Else Get Left Behind by AI?
Ethan’s broader concern was what happens to people and countries outside the leading AI ecosystems. If the most important new industries concentrate in the United States and China, where does that leave everyone else?
My starting point is that AI lets us produce more with less. That expands what is possible. How the gains are distributed is a second question, and a serious one, but concentration at the frontier does not by itself prove that everybody else becomes poorer.
Comparative advantage is useful here. Being less productive in absolute terms does not rule out gains from specialization and trade; what matters is relative opportunity cost. That is why I am reluctant to infer universal human uselessness even from an extreme scenario in which AI becomes better at every intellectual task. It is an argument for potential gains from exchange, not a guarantee that wages or bargaining power will be protected.
China’s development is a reminder of the scale of the gains that economic growth can produce. The World Bank estimates that close to 800 million people escaped extreme poverty over four decades. Trade was part of that transformation, alongside domestic reform, investment, and policy.
My optimism about AI has a similar condition: people and countries need to be able to participate in the growth. An international system closing itself off to trade could produce a very different result. I do not know exactly what every country’s role will be, but I do not think a bad outcome is economically inevitable.
The Cost of Risk-Taking?
Ethan ended by asking what a young person’s mindset toward risk should be. This is the question I find hardest to answer universally.
I come from a financially comfortable background. I did not have student loans, and I have never seriously worried about feeding myself or having somewhere to live. That makes it easier to leave a well-paid job and build a company, or tolerate a personal portfolio that moves far more than most people would want.
Many people around me in trading, consulting, and investment banking have a similar safety net. I sometimes describe them as “post-economic”: their basic needs and comforts are so well covered that optimizing every decision around another increment of security can stop making sense. For that group, I often think the right answer is to take more career risk.
But someone supporting a family, carrying debt, or facing a real possibility of losing their home is making a different decision. I cannot turn my tolerance for volatility into a general instruction for them. The Ferrari joke is funny only because while the loss was painful, it never really was ever threatening to the foundations of my economic life.
About Freeport Markets: Freeport lets users trade 24/7, weekends included, with no KYC and up to 50x leverage across stocks, indices, commodities, crypto, rates, and more, with access to pre-IPO exposure in companies like OpenAI & Anthropic.
For traders who already follow sharp sources across Twitter, Substack, hedge fund filings, corporate insider activity, and political trading disclosures, Freeport uses AI to read the sources they trust, connect the dots across markets, and surface trade ideas as narratives develop.
About the Host: Ethan Kho is the host of Odds on Open, a podcast featuring conversations with traders, investors, and founders about markets, investing, and risk. The show is produced by Patrick Kho.
About the Guest: Lihong Wang is the founder and CEO of Freeport Markets. Before starting Freeport, he traded at Jane Street and IMC Trading, including discretionary semiconductor options at IMC. He graduated from Duke with a degree in mathematics and statistics, and has raised over $2.5M to build Freeport from investors including Y Combinator, Alliance DAO, and Informed Ventures.
Disclaimer: The information provided on TheLogbook (the “Substack”) is strictly for informational and educational purposes only and should not be considered as investment or financial advice. The author is not a licensed financial advisor or tax professional and is not offering any professional services through this Substack.
Investing in financial markets involves substantial risk, including possible loss of principal. Past performance is not indicative of future results. The author makes no representations or warranties about the completeness, accuracy, reliability, suitability, or availability of the information provided.
This Substack may contain links to external websites not affiliated with the author, and the accuracy of information on these sites is not guaranteed. Nothing contained in this Substack constitutes a solicitation, recommendation, endorsement, or offer to buy or sell any securities or other financial instruments. Always seek the advice of a qualified financial advisor before making any investment decisions.