Jack Selby has watched a few of these cycles from the inside. He was an early employee at PayPal, spent years running Peter Thiel’s family office, and now leads Copper Sky, a venture firm. This week he attached his name to an uncomfortable prediction: that Anthropic’s initial public offering, the most anticipated listing in the artificial-intelligence business, could turn out to be the high-water mark for the whole trade.
His argument has two parts, and the second is the one that should worry anyone holding the stocks. The first is that a listing of this size, priced at a valuation that assumes years of uninterrupted growth, tends to arrive when the buying is already done: the institutions that wanted exposure have found it privately, the retail public gets its first chance at the top of a run, and the price has to be justified by results rather than by narrative from the day it starts trading. The second is mechanical. When the lockup period on insider shares expires, a wave of stock becomes sellable at once, and the resulting supply can pressure a price that was set by scarcity.
Lockups typically run six months for a listing of this kind, though underwriters can release shares earlier and often negotiate staggered schedules for large holders. The effect is well documented across market history: the weeks around a first expiry are frequently the weakest part of a newly public company’s first year, not because the business has changed but because the float has. In a company valued in the hundreds of billions, the arrival of even a modest percentage of insider shares is a sum large enough to absorb weeks of demand.
The chain-reaction part of Selby’s warning is where it gets less comfortable. Artificial-intelligence valuations have been set by comparison for two years. A private company’s round is priced against the last round of a similar company, a public company’s multiple is defended by pointing at a peer, and entire portfolios have been marked on the assumption that the category trades as a block. When the largest name in the category stumbles, the marking mechanism runs in reverse, and the companies with the least revenue and the longest path to profit are repriced first and hardest.
None of that is a prediction about Anthropic’s business, and it is worth being precise about what Selby did and did not say. He did not argue that the company is a poor business, or that the models are worse than advertised, or that the revenue is fictitious. He argued about the price and the timing, which are different things, and about the composition of the buyer base at the moment of listing, which is a subject on which a venture investor has unusually good information.
That information cuts both ways. The people who warn loudest about a top are rarely indifferent to it. A venture firm with positions in artificial-intelligence companies has an interest in a healthy public market and an interest in being quoted when the mood turns.
Selby’s view is a view, offered by someone with a stake in the conversation, and it should be weighed as such. It is also, notably, the kind of thing that is easier to say before a listing than after one, when saying it looks like sour grapes.
The broader market has heard similar warnings at every stage of this run, and has so far been rewarded for ignoring them. That is how late-stage rallies work: the skeptics are right early, wrong for a long stretch, and then right again in a way that costs them their credibility at exactly the wrong moment. What makes this particular warning worth recording is not its novelty. It is that it names a specific mechanism, a specific date range and a specific company, which means it can be checked.
The mechanism deserves particular attention because it is the one part of the cycle that does not depend on anyone’s opinion about technology. Lockup expiries are calendared. Insider selling is disclosed.
The float of a stock is a number, not a sentiment, and the relationship between supply and price is the oldest thing in finance. A company can be worth every penny of its valuation and still fall when a hundred million shares become eligible on the same Tuesday.
What happens after that depends on buyers who are not currently in the stock. Index funds will take a position on the day it qualifies, regardless of price. Active managers will decide whether the growth rate justifies the multiple once they can see two quarters of public reporting. Employees will decide how much of their net worth should stay in the shares of the company that pays them, a decision that is made under different constraints than the one made by an investor with a diversified portfolio.
Selby’s version of the outcome is a decline that spreads from the listing to the peers, from the peers to the private marks, and from the private marks to the willingness of funds to commit capital at the next round. That is a plausible sequence. It is also the sequence that has been described, incorrectly, at every previous peak in this cycle. The question for the next six months is whether the calendar finally does what the arguments could not.
