Anthropic's order book is less a verdict on the stock than a tool for sizing the raise. The most revealing disclosure around the deal is not the "multiple times" oversubscription Bloomberg's sources describe. It is the line the Financial Times attributed to three of the company's backers: Anthropic is now focused on "raising as much as possible via the listing to support its heavy spending on computing power, rather than maximizing its valuation." When an issuer says that while its bankers push the price above a marketed range, demand stops being a signal of upside for buyers. It becomes the measure of how much capital the company can extract before trading opens next Thursday.
That changes how the oversubscription headline should be read. A deep book is the reason the price and size go up, which is also the reason less is left on the table for whoever buys at the offer.
Proceeds, Not Price
The amended S-1 sets out the arithmetic. Anthropic is marketing 825 million shares at $70 to $78 under the ticker ANT. At the top of the range that is about $64 billion, the figure the filing gives as the maximum public raise. An overallotment option adds almost $10 billion. Google, Nvidia and Microsoft have separately committed a combined $26 billion through a private placement at the IPO price. Together that reaches the $90.4 billion ceiling the company disclosed on Tuesday.
Bankers at Goldman Sachs and Morgan Stanley now want total proceeds "well above $100 billion." The ordinary arithmetic shows how hard that push is. If the $26 billion placement is a fixed dollar commitment, the public tranche plus the greenshoe would have to bring in more than $74 billion, against roughly $74 billion at $78. Getting "well above" $100 billion therefore requires pricing materially above the range, adding shares, or both. This is an inference from the disclosed figures, not a stated plan. Either route has the same effect for a buyer: a higher entry point or a larger supply of stock at debut.
The valuation numbers point the same way. The roughly $1.4 trillion fully diluted target at the top of the range implies around 18 billion fully diluted shares, again by division rather than disclosure. On that base the 825 million shares on offer are under 5% of the company. Even with the greenshoe and the strategic placement, the tradable float at listing sits in the high single digits. A small float makes a heavy book easy to manufacture and thin trading easy to move. Jack Selby, who runs Copper Sky and is a managing partner at Peter Thiel's family office, pointed to the other end of that structure on Bloomberg TV: "just imagine where it's going to go after the lockups expire." The float that is scarce on day one will not stay scarce.
What a Multiple-Times Book Actually Measures
A book "multiple times" oversubscribed on a $64 billion deal implies indications well into the hundreds of billions. Very few institutional allocators can actually absorb an order that size. Much of that figure reflects a familiar habit. Accounts expect to be cut back on hot deals, so they inflate their orders to secure a meaningful allocation. The headline multiple therefore often measures expected scarcity more than real willingness to hold the stock at the price.
The deal's own history during the roadshow supports that reading. The "fair value" range discussed around the deal slid from $2 trillion to $1.5 trillion, then to $1.2 trillion, as the roadshow began. Bankers are now pushing back toward north of $1.5 trillion on the strength of the book. A valuation that falls by 40% during marketing and then gets bid back on order indications is being set by allocation dynamics, not by agreement on what the business is worth.
The recent record of oversubscribed deals shows that this gap matters after the bell. Klarna and StubHub both drew heavily oversubscribed books and both traded below their offering prices.
Cerebras was roughly 20 times covered and Jersey Mike's about 10 times, and those names also show that coverage ratios and later returns sometimes diverge sharply. Follow-on offerings from Alphabet, Intel, Meta and Strive drew strong demand, and most of those stocks now trade below the offering price. In credit, oversubscribed bond deals from Oracle, SpaceX, Paramount Skydance and Amazon later traded at higher yields, meaning lower prices for the buyers who fought for allocations. That last point deserves attention because it involves debt. Even in the asset class where book coverage is supposed to say the most about clearing levels, a crowded book has not protected secondary holders.
The $26 Billion That Comes Back as Compute
The private placement is the part of the order book that gets the least scrutiny. A $26 billion commitment at the IPO price from Google, Nvidia and Microsoft looks like strategic conviction. These three are also, in different ways, suppliers or infrastructure partners to a company that has just said its priority is funding compute. Capital that comes from the providers of chips and cloud capacity, and that the issuer says will go toward computing power, partly cycles back to those same providers as revenue.
The filing excerpts available here do not show how much of the proceeds is earmarked for any one vendor, and the link should not be overstated. Still, the structure matters for how the demand signal is read. A portion of the "demand" in the headline proceeds is partly a vendor-financing loop. Placing it at the IPO price, whatever that ends up being, also means these buyers are not price-sensitive the way a long-only fund is. Their returns come partly through the commercial relationship. That makes the placement a weak anchor for the fair-value question public buyers face.
The same logic applies to the broader tape. Nasdaq futures rallied Wednesday after CoreWeave reported strong sales growth, lifting Taiwanese and Korean equities overnight. The rebound followed last week's Democratic sweep, which raised fears of nationwide data center moratoriums. The AI complex is using one infrastructure company's results to steady itself ahead of a listing whose proceeds are destined for infrastructure. Sentiment and fundraising are leaning on the same pillar.
Where the Oversubscription Record Cuts Both Ways
The strongest objection is that the record of oversubscribed deals is mixed, not uniformly bearish. SpaceX was about four times covered and trades above its offer. Strive is the other exception in the follow-on group. A "graveyard" that includes notable survivors cannot prove that a hot book forecasts weak returns. It only shows that a hot book does not forecast strong ones.
Outside reference points also sit above the bankers' numbers. Morningstar's estimated fair equity value is $1.7 trillion. Hyperliquid's pre-IPO Anthropic contract implied about $1.8 trillion on Wednesday morning. If those marks hold, even a price north of $1.5 trillion leaves room for an opening gain. On that reading the bankers are not squeezing buyers but closing part of a genuine valuation gap.
Both of those anchors have limits. Morningstar's figure is an analyst estimate, not a clearing price. The Hyperliquid contract trades in a thin, crypto-native venue with no obligation to deliver shares, and it is exactly the kind of instrument that inflated pre-listing enthusiasm tends to inflate further. They are useful as sentiment gauges. They do not show how institutions will behave once they hold the stock and the float starts to trade.
There is a second objection: an issuer raising maximum cash for compute could be the right decision for long-term holders, even at the expense of day-one buyers. That may be correct for the business. It does not help the person paying the upsized price. The FT's account makes the issuer's priority explicit, and it is not the buyer's priority.
The Print That Settles It
Three observable figures will decide the read before and shortly after next Thursday's debut.
- How the raise is enlarged. Pricing is set for Wednesday night. A higher price on an unchanged share count says demand is price-insensitive. An increase in share count says the company is choosing size over price, which fits the FT's account and adds supply into a small float.
- The gap between the final price and the outside marks. A price that implies more than $1.5 trillion narrows the gap to Morningstar's $1.7 trillion and Hyperliquid's roughly $1.8 trillion. The smaller that gap, the less the "oversubscribed" premium is available to new buyers.
- Whether the offer price holds as support. Klarna and StubHub show that a deal can clear easily and then trade below the offer. If ANT closes its first sessions under the offer price despite a book several times covered, that confirms the book measured allocation competition, not conviction. Selby's warning then moves from rhetoric to a live risk for the wider AI trade.
The thesis fails if Anthropic prices above the range without adding shares, and the stock holds a premium to the offer through the first week of trading. Until that happens, the cleaner reading of the evidence is this: a company that says it is maximizing cash, a book being used to justify a higher raise, and a float small enough to make both look stronger than they are.
