For two weeks, the allocators who intend to buy into Anthropic’s listing have been editing the same cell in the same model. The figure at the top of the page used to read $1.5 trillion. The working range is now $1.2 trillion to $1.3 trillion, according to people familiar with the matter, and some of those buyers still describe the earlier number as a stretch.
The calendar behind that revision is now specific. Anthropic will start its formal roadshow next week and aims to trade in the week of Nov. 16, completing the listing before the Thanksgiving holiday. Investors get a compressed round of meetings. The company gets a price, and a date it can hold its own employees to.
The Thanksgiving window is doing real work. A deal that prices before the holiday reaches the market while trading desks are fully staffed, leaves time for the underwriters to support the stock through its first weeks, and lands before portfolio managers close their year. Slip past it and the next credible slot is January, when a new fiscal year resets every comparison and buyers ask why the transaction did not get done.
Two weeks of violent swings in technology shares supplied the rest. Institutions that had been briefed on a $1.5 trillion mark have been rebuilding their models around a figure 13% to 20% lower, which is what a range of $1.2 trillion to $1.3 trillion represents against the old number.
What buyers are arguing about is the arithmetic of the range. At $1.25 trillion, the price is a claim on revenue years into the future, made about a company with a short financial history and no traded peer. Its closest comparable, OpenAI, is private as well. The large software companies and the large chip companies, the two reference points a banker would reach for first, are priced on very different things, one on renewal and retention, the other on a cycle.
That absence of a benchmark is why the roadshow matters more than the paperwork. A book that fills at the top of the range tells the company demand is deeper than the public conversation suggested, and it gives the next AI listing a floor. A book that fills at the bottom, or that fills mostly with a handful of crossover funds, tells the opposite story and sets a ceiling the next issuer has to argue against.
There is a mechanical cost to listing in this window that does not appear in the valuation. A public company has to report every quarter, disclose customer concentration and defend its accounting in front of people paid to find problems. For a business whose largest costs are multi-year commitments for compute, and whose revenue arrives on contracts that can be canceled, that disclosure is not neutral. It changes what salespeople can promise and what finance can recognize.
The other cost falls on employees. Restricted stock that converts into tradeable shares is easier to value and easier to leave. A fixed listing date helps retain senior staff who have offers from competitors with liquid stock, and it starts the clock on a lockup that will be watched for signs of insider selling.
The competitive frame is unavoidable. OpenAI has been moving toward its own listing, and the order in which the two arrive will shape how each is priced. The first comparable to trade gives the second a reference, and references cut both ways. A strong debut lifts the follower. A weak one hands every short seller an argument built from the leader’s own numbers.
Some of the caution inside the process is about size rather than quality. A company that lists above $1 trillion needs buyers able to write checks large enough to move the book, and the number of institutions with that capacity is small. Those buyers know it and negotiate accordingly, which is one explanation for a range that opens below the last private mark.
They also have a memory of the past few years. Several large technology listings priced at the top of their ranges and spent the following quarters below the offer price, and the funds that absorbed those deals have been slower to commit the next time. That memory is part of why the range moved, and part of why the company is willing to accept it.
What to watch in the next ten days is coverage rather than price. Bankers call a book covered when the orders on hand would fill the deal. A book covered several times over at the low end usually prices near the top of the range. A book covered once, by accounts that intend to sell into the first rally, does not, and the difference shows up in the first month of trading rather than on the cover of the prospectus.
None of this settles what Anthropic is worth. It settles what a group of institutional buyers will pay for it this month, in a market that has spent two weeks deciding the answer is lower than it was in October. The listing will produce the first public number either of the two leading AI labs has had to defend.
