For most of the year the argument about Anthropic was a multiple. It is now a cost of money. The company starts its roadshow next week still carrying a $1.5 trillion valuation, according to the Financial Times, and investors briefed on the process describe that figure as a stretch; several expect the final price to land between $1 trillion and $1.3 trillion. The credit market has already voted. Blackstone, the anchor investor in the $63 billion debt package arranged to finance Anthropic’s chips, has told investors in that deal to expect a high junk rating for the company after it lists, the FT reported, a reversal of earlier guidance that Anthropic would reach investment grade before 2027. High junk means the upper end of speculative grade, a band for issuers with real business risk and limited cushion, priced to compensate lenders for the possibility of default.
The rating matters because of what sits in the footnotes. Anthropic has disclosed compute purchase commitments of $895 billion. Those are contracts to buy computing capacity over many years: fixed obligations that behave like debt when demand falls and that do not appear as debt in the debt-to-earnings measure a rating agency computes. The revenue against them arrives on contracts customers can cancel or renegotiate. Rating committees have spent decades learning to distrust that asymmetry, and the memory of it is why a company with a celebrated product can still be treated as a speculative borrower.
The two markets are pricing different things, which is not a contradiction. Equity buys an option on a future that may be enormous, and the buyer can lose no more than the share price. Credit buys the floor, and the lender’s return is capped while the loss is not. A company can be worth $1.2 trillion to a shareholder and still be a BB borrower, because the questions are different: how much can this be worth if it works, against how much cash does it need if it does not.
What makes the equity case hard to argue is the absence of a benchmark. OpenAI is private. The large software companies trade on renewal rates and retention, the large chip companies on a cycle, and Anthropic is neither. At $1.25 trillion the price is a claim on revenue years out, discounted at a rate nobody has calibrated for a business whose largest cost is a multi-year commitment and whose product improves on a schedule its own researchers cannot forecast.
The buyers who can move a book that size are few. A listing above $1 trillion requires orders large enough to matter, which narrows the field to a handful of sovereign funds, the biggest index-adjacent managers and a group of crossover funds that invest in both private and public rounds. Those institutions know the deal needs them, and they negotiate accordingly. That is one explanation for a range that opens well below the last private mark.
There is also a mechanical cost that does not show up in the valuation. A public company reports every quarter, discloses customer concentration and defends its accounting to people paid to find problems. For a business that sells multi-year access to a model, that disclosure changes what salespeople can promise and what finance can recognize. It also puts a number on churn every ninety days, in a market where the competitors are private and can keep theirs quiet. The employee side carries its own arithmetic. Restricted stock that converts into tradeable shares is easier to value and easier to leave with. A date certain helps retain senior researchers who have offers from labs with liquid stock, and it starts a lockup clock that will be watched for signs of insider selling in the first two quarters. For a company whose assets leave the building at 6 p.m., that matters more than it would at a chipmaker.
The compulsion to list now is partly competitive. OpenAI has been moving toward its own listing, and whichever company trades first gives the other a reference. A strong debut lifts the follower and validates the whole category. A weak one hands every skeptic a number built from the leader’s own disclosure, and the second issuer spends its roadshow explaining the first one’s stock chart.
What to watch in the next two weeks is coverage rather than price. Bankers call a book covered when the orders on hand would fill the deal. A book covered several times at the low end usually prices near the top of the range; a book covered once, by accounts planning to sell into the first rally, does not. The difference tends to show up in the first month of trading rather than in the prospectus. The rating agencies will move within a quarter of the listing, and their first action will say more about the business than the share price does. An upgrade path requires the company to convert commitments into revenue on schedule, which means the compute it has contracted for has to be sold at a price that covers the depreciation. If it is, the junk label is temporary and the debt gets cheaper. If it is not, the label sticks, and the next round of chips gets financed at a rate that shows up in every projection the company publishes.
