Anthropic Pre-IPO Contract Arbitrage Retrospective: Key Event Timeline and Position Micro-Management

The Anthropic pre-IPO contract arbitrage is a rare market opportunity with both substantial capacity and a substantial profit margin. If you did not get on board this time, or did not size up enough, then I think it is well worth reviewing the entire arbitrage process with me, to understand why such an enormous opportunity still did not attract heavy positioning after we made it public.

For the full logic of this arbitrage, see my previous post. Today I will review only two things:

• Timeline: over the two weeks, which events drove the spread tighter, which widened it again, which affected the win rate, and which affected the odds;
• Micro-management: with the big thesis unchanged, how I managed the position in real time based on those external events.

First, the conclusion: the spread narrowed from about 30% to about 2% in two weeks, and I have closed everything and booked the gains. Before we published our analysis, we detected essentially no peers; after publication, there were actually two very good entry points—the few hours right after our tweet went out, and 11-07 when the roadshow was confirmed to proceed as scheduled: the rules were all confirmed, yet about 15% of profit margin remained, and the entry window was long enough.

I. What This Trade Was Earning

Anthropic has not yet listed, but several exchanges have launched "pre-IPO contracts" that will convert to the stock price and continue trading after listing. The key is how many shares each contract converts into at listing:

• Binance quotes at "valuation ÷ 1 billion"; with the market rumoring $1.3 trillion, Binance quoted about $1,300. But at listing it converts only by post-issuance total shares outstanding, so one contract converts into only about 13.1 shares.
• The valuations cited by the market are all on a fully diluted basis (total shares plus options, RSUs, convertible bonds, and other items that may become shares in the future—about 17.5 billion shares). $1.3 trillion works out to about $74 per share, so 13.1 shares are worth only about $970, yet Binance sells for $1,300. The share count is about a quarter lower, which makes the price about a third more expensive.
• Hyperliquid has two contracts based on shares: xyz:ANT (trade.xyz) quotes directly per share; io:ANTH (Entropy) converts by fully diluted share count, with each contract converting into about 17.5 shares after listing. Neither has this mispricing.

So the trade was to go long HL and short Binance, sized so that the share counts each converts into at listing offset each other. Regardless of whether Anthropic rises or falls after listing, the share-count gap is locked in at the moment of conversion. A few terms below:

• Route: long xyz and short Binance is the "xyz route"; long io and short Binance is the "io route."
• Lockable profit margin: the return that can be locked in if a position is established now and held until listing conversion (calculated on the amount used to buy HL). At launch, the xyz route was about 30%.

II. Before Publication: No Competitors

To get into this trade at this stage, you had to thoroughly research the prospectus and the rules of four exchanges, and confirm that Binance would convert by total shares outstanding; the bar was relatively high. Based on our monitoring at the time, the only people doing this arbitrage during this period were essentially me and my collaborators.

We built our position during this period, starting with the xyz route. After that, io's conversion rules changed twice:

• ① 10-29: io first published its conversion rule: after listing, shares would be converted by "total shares + options + RSUs," which worked out to about 15.3 billion shares. The io route was about 20%, below the xyz route's about 24%, so we stayed with xyz.
• ② 10-31: io added convertible bonds: Amazon's convertible bonds were also counted in full, for about 16.9 billion shares. This was the complete fully diluted basis. The number of shares each io contract could convert into rose from about 15.3 to about 16.9, and the io route jumped to about 31%, more than 10 percentage points above the xyz route.

Position micro-management #1: switch routes. The io route's profit margin changed with io's rules and the prices on both sides, so we switched to whichever route was higher. On the day ② came out, we switched the long leg from xyz to io. By the end, that switched portion earned about 11% more than if we had stayed in xyz.

III. Publication Day: Half the Spread Gone in Hours

③ 11-04 midday, my collaborators Longwang, Ray, and others each published articles analyzing the logic of this arbitrage from their own perspectives.

• Within 40 minutes, the xyz route fell from 15.3% to 10.5%, and the io route from 22.7% to 19.6%. On an hourly basis, the xyz route bottomed at about 6% and the io route at about 12%.
• The compression was so fast partly because newly arrived arbitrage flow placed orders on both sides at once, and partly because people who had been on the opposite trade (long Binance, short HL) were closing out and admitting they were wrong.
• Acting the moment you saw the tweet could earn substantial profit. Missing the fastest wave was not fatal: later, some traders took profits early and closed positions, widening the spread back somewhat; during the next day, the xyz route still offered 7–8% and the io route 13–15%.

IV. Three Exchanges Speak: Rules Move from Inference to Confirmation

Within a day after publication, all three exchanges had made their positions official:

④ Binance: a Binance staff member clarified the total-shares rule under Longwang's tweet. The gist was: the contract tracks the price of one ordinary share of Anthropic. Share capital counts only issued and outstanding ordinary shares. Unexercised options, unsettled RSUs, unconverted convertible bonds, the underwriters' over-allotment, and shares reserved under incentive plans are not ordinary shares now and are not counted;

• This was not an ad hoc interpretation: when the contract launched on June 2, Article 8 of the Contract Specifications already stated it;
• ⑤ OKX: OKX's Anthropic contract is the same type as Binance's (quoted by valuation). It directly changed its rules page to state in black and white, "the total number of ordinary shares outstanding immediately following the IPO," explicitly not using fully diluted shares.
• io (Entropy): its rules page states that it copies the "fully diluted total" from the prospectus, and io's CEO also publicly said it would convert by fully diluted share count.

At this point, "which share count each exchange uses" was no longer an inference but official confirmation. The only major remaining risk was whether the IPO would be delayed.

V. Crash Night: The Only Remaining Risk Was "Delay"

⑥ 11-05 evening, US stocks crashed: the Nasdaq 100 fell about 6.4%, the chip index fell about 12%, and the VIX rose about 70% in a single day.

• The market began speculating about whether Anthropic would delay or even cancel its listing. Polymarket's "list this year" odds fell from about 91% to a low of about 66% (Fig. 3, lower panel).
• Some people panicked and closed positions, widening the spread: the io route went from about 15% back to around 18%, approaching 20% intraday.
• Our judgment: it would most likely still list as scheduled. Our friends at investment banks had heard nothing about a canceled roadshow; moreover, a mere delay would not actually affect the strategy's fundamentals, and at most peers closing early could widen it by a few points.

Position micro-management #2: on the news, trim a small amount as insurance and keep the main position. The risk had shifted from "rules" to "timing." We sold a small portion on the night of the crash, but most of the position stayed put.

VI. Roadshow on Schedule: Maximum Certainty with 15% Margin—the Best Post-Public Entry

The night before the roadshow news (11-06 21:12), Bloomberg reported that some potential investors were only willing to value it at $1.2–1.3 trillion (previously the talk was $1.4–1.5 trillion). That night Binance fell about 11%, while io fell only about 7%, and the io route briefly compressed to about 11%.

⑦ 11-07 early morning, FT and CNBC reported one after another that Anthropic was proceeding as scheduled and would begin its roadshow the following week.

• Polymarket's "list this year" odds went from about 69% back to 90% within an hour.
• The spread widened again instead: after the news, io was sold lower while Binance barely moved, and the io route went from about 11% back to about 15%, then stayed around 15% for more than a day.

Position micro-management #3: add as soon as the news came out. All three exchanges had confirmed the rules, and the delay worry was gone. Certainty was the highest of the two weeks, yet about 15% of profit margin remained. This was the most worthwhile entry point of the entire process; when we saw the reports, we added the position back and then used unrealized gains to increase it further.

The window lasted another two or three days, with the spread slowly converging and the profit margin drifting lower. By the morning of 11-10, the io route was about 9% and the xyz route about 6%.

VII. The Leak, the Prospectus, and the Exit

⑧ 11-10 morning, the price range leaked: Reuters reported $70–80 per share, followed by reports that the terms had been withdrawn. We judged the range was very likely real: $75 times the fully diluted share count came out right in the previously rumored $1.2–1.3 trillion range. Since the terms had leaked, a formal priced prospectus was very likely to be released that evening. At the moment the news came out, xyz fell from about $104 to about $95, even more than Binance, and the xyz route briefly widened to about 11%. The reason was that xyz is quoted per share, so people compared it directly with $70–80.

• From that point, we began adding to a one-sided short Binance position; the logic is in Section VIII.
• ⑨ 11-10 evening, the priced prospectus: $70–78 per share, with post-issuance total shares of about 13.09 billion. The spread compressed immediately: the xyz route went from about 10% to just over 2%. Binance fell about 4% within an hour.
• ⑩ 11-11 early morning, io fixed at about 17.5 billion shares: Entropy stated that the estimated listing conversion was about 17.5 billion shares, slightly less than the ~18 billion estimated under the previous basis. The io route fell from about 5% to about 2%.
• On the morning of 11-11, we closed everything. The xyz route is now about 1% and the io route about 2%; this arbitrage is over.

VIII. The One-Sided Binance Short: Why We Kept It and Why We Added on the Leak

In addition to the arbitrage position, we always held an extra short in Binance (one-sided, with no corresponding long). The reason starts with "who was moving."

Binance was the one with the wrong share count, yet it was HL that got pulled up. The share-count mispricing belonged only to Binance, so convergence should logically have come mainly from Binance falling. What actually happened was:

• Before publication, it was indeed mainly Binance that fell: Binance −12.6%, while xyz fell only 1.9%.
• After publication, arbitrage flow placed orders on both sides. But Binance had the best liquidity, so the same money moved it the least; instead, xyz and io on HL were bid up more: from after the tweet to before the crash, Binance −3.6% and xyz +3.4%; after the roadshow was confirmed, Binance +3.6% and xyz +6.8%.
• Only on the evening of the prospectus did Binance catch down by 5.0%, while xyz barely moved.

Binance's premium had two layers. Using the rumored valuation: $1.2–1.3 trillion (fully diluted) worked out to about $71 per share; on 11-07, Binance's total-share basis worked out to about $105 per share, about 50% higher. Of that 50%:

• About 10 points were the share-count mispricing, unique to Binance—the arbitrage spread itself—which arbitrage flow would gradually compress;
• The remaining ~40 points reflected market-wide optimism on valuation; xyz was the same (about $97 per share).

The one-sided Binance short bet on both layers at once: the first was highly likely to compress; the second required waiting for the market to admit it was wrong.

Why we added on the leak. As soon as the price range leaked, we added to the one-sided short for two reasons:

• The range was very likely real ($75 times the fully diluted share count lined up exactly with the $1.2–1.3 trillion rumor);
• With the range, the second-layer premium had a direct yardstick. xyz is quoted per share, the same unit as the offer price; put $100 per share next to the $70–80 range, and anyone could see it was expensive. As xyz converged toward the range, and with Binance and xyz tied together by arbitrage flow, Binance would be pulled down as well.

IX. The Final Mistake: Staying Short After the Prospectus

After the prospectus came out, we kept the Binance short. Our thinking at the time was:

• Binance's total-share basis was still around $98 per share, about 30% above the $70–78 prospectus range (Fig. 4, lower panel: the distance between the yellow band and the Binance line);
• Once the media wrote, based on the prospectus, that the market cap on a total-share basis was about $1 trillion, people would realize the price was off, and the premium should start to converge.

But when we woke up the next morning:

• The premium had not converged; Binance had actually risen a bit, and xyz rose too;
• Just then, news (FT) said underwriters were preparing to raise the valuation above $1.5 trillion.

At that point, we knew the call was wrong: the longs had not capitulated just because the media wrote "market cap about $1 trillion." We stopped out immediately and closed everything. That short had once been up substantially on paper on the night of the prospectus, but it ended in a loss.

Join the official Coincamps community:

X: https://x.com/coincamps

Telegram: https://t.me/coin_camps