top of page
Exponential Logo.png

In $ANT IPO Analysis-Exponential shows institutions sold $12.1 billion of Anthropic on IPO day while retail bought

13 minutes ago
8 min read

Global Flow Strategy · By Exponential Technology · November 21, 2026


Institutions sold $12.1 billion of Anthropic on its first day and retail bought. The confidence was in SpaceX, where holders barely sold at all.


On the day Anthropic went public, institutions sold $12.10 billion more of the stock than they bought. Retail did the opposite: it bought $109 million net. Big accounts were leaving, small accounts were arriving, on the same day, in the same stock. That was more institutional selling than Nvidia, AMD and SpaceX combined that session.


One session is not a trend. But it is a clean look at something most market commentary gets wrong. The market is not one actor. Four large AI names traded that same day, and the people selling were not the people buying.


Compare it to SpaceX, and the distinction gets sharper. SpaceX went public on June 16 and traded $66.09 billion on its first day. Its institutions moved $61.5 billion of that — 93% of everything that changed hands — but they finished nearly balanced, selling $31.94 billion against buying $29.53 billion. Anthropic's institutions moved $83.4 billion, or 96.5% of its volume, and finished badly lopsided: $47.73 billion sold against $35.63 billion bought.


That is the real contrast, and it is not the one the net figures alone suggest. Both first days were overwhelmingly institutional. The difference is who institutions were trading with. In SpaceX they were mostly trading with each other. In Anthropic they were net handing stock to somebody else, and the buyer on the other side was overwhelmingly retail.


What this note covers, section by section:


1. The Four, Side by Side — one session, one method, four very different pictures.

2. What the Numbers Say About Each Name — and why Anthropic's selling was a different kind of selling.

3. Why Every One of Them Sold — the common thread, and what it tells you about the quarter.

5. What One Session Cannot Tell You — the limits, stated plainly.

6. Bottom Line — what to watch next, and when.



Net institutional flow across Anthropic, Nvidia, AMD and SpaceX


1. The Four, Side by Side


Same session, same method, four different pictures of who was on which side.


The table below is what our Equity Flow product produces every day. Every trade on the US consolidated tape gets classified by the type of account behind it, institutional or retail, and by direction. Add them up and you get a net figure per name per session, which is a fact rather than an inference.


Anthropic traded $86.41 billion across 5.33 million trades on its debut. Institutions bought $35.63 billion and sold $47.73 billion, leaving them $12.10 billion more sold than bought. Retail bought $1.57 billion against selling $1.47 billion, for a net $109 million in.


Nvidia traded $44.66 billion. Institutions sold $2.84 billion net and retail sold $84 million net. AMD traded $34.91 billion, with institutions $370 million net sold and retail roughly flat at plus $9 million. SpaceX traded $16.09 billion, with institutions $690 million net sold and retail plus $48 million.


Stack those up and the pattern is hard to miss. Every one of the four was a net institutional seller. Only two had retail buying at all, and in Anthropic's case retail's $109 million is small against $12.10 billion of institutional supply.


What would make this section wrong is a classification error, which is why the method matters more than the number. The product does not guess who the buyer was from the price. It reads the trade and the counterparty structure, and the classifications are reproducible from the same tape.


2. What the Numbers Say About Each Name


Anthropic's selling was a different kind of selling from the other three, and the difference is what makes the day interesting.


Start with concentration, which is the figure Equity Flow makes easy to see and hard to argue with. Anthropic's institutional selling was 14% of its total volume. Nvidia's was 6.4%, SpaceX 4.3%, AMD 1.1%. The debut did not just see more dollars sold than the incumbents. It saw a larger share of its own trading driven by institutions stepping back.


Then consider what the selling means in each case. When institutions sell Nvidia, they are reducing a position they have held. It is a decision about an existing allocation. When institutions sell Anthropic, most of them are selling something they could not sell until that morning. There is a lock-up on pre-IPO shares, and a lock-up is a contractual period after a listing during which early holders are not allowed to sell. Once it lifts, or once the offering allows a partial release, that supply becomes real and it arrives at once.


That is why a first day can look strong on price and weak on flow at the same time. The stock opened at $111.50 against an $83 offer and closed at $97.17, a good day by any measure. Underneath it, institutions were handing $47.73 billion of stock to the market and taking back $35.63 billion. The price held because there was enough demand to absorb it, and the flow says that demand was mostly smaller accounts.


Anthropic intraday price and volume, regular session, 19 November 2026

The chart shows the shape of that day. The stock printed near $106, faded through $95 inside the first half hour, and closed at $97.05 — still a 17% gain on the $83 offer. The heaviest volume came in the opening half hour, 208 million shares, which is what distribution looks like: institutions handing stock to the market at the moment of maximum attention.



AMD is the quiet one, and the quiet is informative. Equity Flow shows almost no institutional activity in the name relative to its volume. Institutions sold $370 million net on $34.91 billion of volume, which is 1.1%. Almost nobody moved. When a name trades that heavily and the classified flow nets out near zero, buyers and sellers were roughly balanced among institutions, and the stock is being repriced by conviction rather than by flows.


SpaceX sits between the two. A small net sale, 4.3% of volume, with retail a modest net buyer. It is the profile of a name being held rather than traded.


What would make this wrong is a second session that reverses the first. If institutions come back as buyers in Anthropic next week, Thursday was a clearing event rather than a verdict, and that is a real possibility worth naming.


3. Why Every One of Them Sold


Four names, four stories, one shared direction. That is not a coincidence, and it is the most useful thing in the data.


This is the kind of question Equity Flow was built to answer, because it lets you compare across names rather than guess at one. When every large name in a sector shows net institutional selling on the same session, the cause is usually portfolio-level rather than name-level. A fund that needs to raise cash, reduce risk, or meet a redemption does not sell its favorite idea. It sells what it can, and it tends to sell across the book. The technical word for the resulting pressure is de-risking, and the word undersells how indiscriminate it can be.


The timing fits. That Thursday was the Anthropic debut, which meant a very large new issue was absorbing capital on a single day. New supply of that size does not appear from nowhere. Money that would have gone into existing names goes into the new one instead, and some of it comes from selling the existing names to fund the purchase. That is a rotation, and rotations show up in flow before they show up in price.


There is a second explanation that runs alongside it. This was the same week the energy complex was unwinding, with crude down about 13% after the midterms, and the AI complex was carrying the year's gains. A week like that invites profit-taking in the winners regardless of what else is happening. Both readings point the same direction, which makes the observation more robust rather than less.


What would make this wrong is a session where these four diverge sharply while the rest of the tape is calm. That would suggest the selling was name-specific after all, and the rotation story would need replacing.


4. Compare It To SpaceX


SpaceX went public on June 16 and its first session looked nothing like Anthropic's. The difference is the most useful comparison in this piece.


SpaceX traded $66.09 billion on its first day. Its institutions moved $61.5 billion of that, and finished nearly balanced: $31.94 billion sold against $29.53 billion bought, a net $2.41 billion out. Retail bought $430 million net, roughly four times what Anthropic retail bought on a day with fewer dollars changing hands.


Net institutional and retail flow, Anthropic first session versus SpaceX first session

Put the two side by side and the share-of-volume figures make it plain. Anthropic's institutional selling was 14.0% of everything traded; SpaceX's was 3.6%. In dollars, five times as much left Anthropic on its first day.


There is a reasonable reading here, and it is the one worth taking seriously. The listing where existing holders were least eager to sell was the one where the people closest to the business had the most confidence in it. SpaceX holders looked at a $66 billion first day and mostly held. Anthropic holders looked at an $86 billion first day and sold $12 billion into it. That does not make Anthropic a worse company. It does suggest the people who knew it best were taking money off the table at the first legal opportunity, and that is information about how they saw the price rather than about how they saw the business.


The alternative explanation deserves equal weight, because it is just as plausible. Anthropic's shareholder base may simply have been more concentrated among early venture funds sitting on enormous paper gains, and a fund with a ten-times return has a fiduciary reason to realize some of it. SpaceX's base includes a large strategic holder whose position is not managed for return on the same schedule. Different owners produce different first-day behavior, and neither pattern is a verdict on the company.


What would make this comparison wrong is a different pair of listings that reverses it, or a second session in either name that flips the direction. One comparison of two events is an observation, not a law.


5. What One Session Cannot Tell You


The honest limits of a single day of data, stated up front.


One session cannot establish a trend. It can establish what happened in one session, and that is all we are claiming here. The Friday data had not been delivered when this was written, so the read covers Thursday only.


Classification is a judgment, not a measurement, and we say that about our own product. Equity Flow decides whether a trade came from an institution or a retail account by reading the structure of the trade. Deciding whether a given trade came from an institution or a retail account is an inference from the structure of the trade. The method is consistent and reproducible, but it is not the same thing as a census. Anyone who tells you the split is exact is overselling.


And a net number hides the gross. Anthropic institutions bought $35.63 billion and sold $47.73 billion. Someone bought $35.63 billion of stock that day, and calling the day institutional selling does not mean no institution was buying. It means more was sold than bought, which is a different and weaker statement. We report both figures for that reason.


The last limit is the one that matters most for anyone making a decision. Flow tells you who moved. It does not tell you why, and it does not tell you what happens next. It narrows the range of plausible stories, which is valuable, but it is not a forecast.


6. Bottom Line


We think the useful takeaway is not that Anthropic sold off. It is that the whole AI complex was being sourced for cash on the day its newest member listed.


Near term, watch the next few sessions for whether institutions return as buyers in Anthropic. If they do, Thursday was a clearing event and the supply is behind us. If they keep selling, the debut marked the start of a distribution rather than the end of one.


Here is what would change our mind, with dates. The Friday session, once delivered, either confirms or breaks the pattern. The following week shows whether the rotation persisted. And the next large AI listing gives us a second natural experiment to compare against this one.


Near term we are neutral on the group and watching the flow rather than the price. Longer term the question is whether this complex can absorb new supply without the incumbents funding it.


We will update this when the Friday data lands.


Same tape, four names, one direction.


This material is provided to institutional investors for informational purposes only. It is not investment advice, an offer, or a solicitation. Flow classifications are model-based inferences from trade structure, not a census of participants. Data are as of the session shown and are subject to revision. © Exponential Technology Inc.



Comments


Read the research,
then test the data.

Request a white paper from our research library, or book a call with the research team behind it.

bottom of page