Anthropic, the artificial intelligence lab behind the Claude assistant, set the terms this week for what would be the largest first-time share sale in history by valuation. The company is offering 825 million shares at an indicated range of $70 to $78 each, with pricing scheduled for Wednesday, 18 November 2026 and trading due to begin on Nasdaq the following day under the ticker ANT.

At the top of that range the sale would raise $64.35 billion, with a further $9.7 billion available if the underwriters, led by Goldman Sachs and Morgan Stanley alongside twenty other banks, exercise their option to sell additional stock. The implied valuation is roughly $1.3 trillion to $1.4 trillion on a fully diluted basis, or about $917 billion to $1.02 trillion counting only shares currently outstanding.

The timing is awkward. In the week before the roadshow began, the Nasdaq fell 5.5 percent and erased close to $2.5 trillion of market value in a single session. Then, on Thursday 12 November, Nvidia chief executive Jensen Huang resigned following reports of extramarital affairs, including one with a woman that US intelligence reportedly believes is a Chinese spy. The bellwether of the entire AI trade lost 12.75 percent over that week, its sixth consecutive weekly decline and its longest losing streak since 2012.

In this article we explore how Anthropic's offering is structured, what its valuation multiple actually measures, why an $80.7 billion reported loss is less alarming than it sounds, what oversubscription does and does not prove, and whether the deal can be read as a verdict on AI economics.


The Shape of the Deal

An initial public offering is the first sale of a private company's shares to public investors. Two numbers matter and they are often confused. The first is how much money the company raises, which depends on how many shares are sold. The second is what the whole business is said to be worth, which depends on the price per share multiplied by the total share count.

Anthropic is selling a comparatively thin slice. At roughly $1 trillion for shares outstanding and 825 million shares on offer, the public float works out at somewhere near six percent of the company. That is deliberate. A small float with heavy demand tends to support the price in early trading, because there simply is not much stock available to sell.

The $9.7 billion underwriter option is what the market calls a greenshoe. Banks are permitted to sell more shares than the company formally offered, and if the stock trades well they exercise the option to cover that short position. If the stock trades badly they buy shares in the open market instead, which provides a floor. It is a stabilisation mechanism dressed up as an extra tranche.

The gap between the fully diluted valuation of up to $1.4 trillion and the roughly $1.02 trillion counting outstanding shares matters too. Fully diluted means every share that could exist once employee options, restricted stock and convertible instruments are counted. The difference, roughly $380 billion, is a measure of how much of this company has already been promised to people other than the buyers of the IPO.


The $26 Billion Order From the Supply Chain

Running alongside the public offering is a private placement of $26 billion at the IPO price, taken up by Google, Nvidia, AMD, Broadcom, Micron, Microsoft and TSMC Partners. Crucially, that stock carries no lock-up, meaning those buyers face no contractual period during which they are barred from selling.

This is the most revealing feature of the transaction. The buyers are, with one exception, Anthropic's suppliers and cloud partners: the companies that sell it chips, memory, networking silicon, foundry capacity and compute. Money raised from an AI lab's investors flows to the chip makers, and some of it flows back as equity in the AI lab.

That circularity is not in itself fraudulent or even unusual in infrastructure build-outs. Vendor financing has a long history in telecoms and heavy industry. But it does mean the $26 billion cannot be read as a clean, independent price signal. These are strategic buyers with a commercial interest in Anthropic's continued spending, not dispassionate allocators of capital. The absence of a lock-up makes the commitment softer still.


What 17 Times Revenue Actually Measures

At the top of the range, the valuation is roughly 17.3 times Anthropic's $81 billion annualised revenue run rate as of 31 October 2026. A revenue multiple divides what the market says a company is worth by its annual sales. For context, the median public software company currently trades at roughly 3.4 times revenue, with mega-cap SaaS names at a median of about 14.6 times.

The word doing the heavy lifting is run rate. Anthropic defines it by annualising consumption-based fees over a trailing 28-day period and subscription fees for a single month. In plain terms: take the most recent four weeks of usage revenue, multiply up to a year, and add twelve months of the latest month's subscriptions. It is a forward-looking extrapolation, not a record of money received.

The actual recorded figures are smaller and still extraordinary. Revenue for the first nine months of 2026 was $33.24 billion, up roughly 1,200 percent year on year, with first-half revenue of $16.29 billion representing a 1,304 percent jump. The run rate stood at $65 billion at the end of July and $81 billion three months later, a $10 billion increase in October alone.

So the multiple cuts both ways. Against trailing revenue the valuation looks far richer than 17 times. Against the run rate it looks defensible only if growth of that magnitude persists, because a run rate built on 28 days of consumption can contract as quickly as it expanded if enterprise customers throttle usage.


The $80.7 Billion Loss, Unpacked

Anthropic's filing disclosed a net loss of $80.7 billion for the first half of 2026 against $16.3 billion of revenue. On its face that is one of the largest half-year losses ever reported by a company seeking a public listing.

Most of it is not cash leaving the building. Roughly $72.4 billion came from non-cash accounting adjustments tied to the rising value of convertible debt and other financial instruments. When a private company issues instruments that convert into equity, accounting rules often require it to revalue them each period. If the company's own valuation soars, the liability attached to those instruments soars with it, and the increase is booked as a loss. Paradoxically, the better the business does, the larger this charge becomes. Once the instruments convert at listing, the effect disappears.

The operating loss, which measures the actual cost of running the business against revenue, was $693 million, narrowed from $2.1 billion a year earlier. That is the number to watch. On a revenue base of $16.3 billion in the half, it implies a business approaching operating breakeven at remarkable speed.

The longer history is messier. Full-year net losses grew from $8.3 billion in 2024 to $41.9 billion in 2025, against a $12.9 billion loss in the first half of 2025, which means the deterioration was concentrated in the second half of last year. Investors buying into the operating-leverage story are betting that the trend of the first half of 2026 is the real one.


Oversubscription and the Limits of the Signal

The deal has been reported as oversubscribed, meaning investors have asked for more shares than are available. Bloomberg put the figure at about six times, later reports at nearly seven, and Altimeter Capital chief executive Brad Gerstner said on Monday 16 November that demand was running at ten to twenty times.

That spread is itself informative. Oversubscription numbers are not audited. Investors routinely inflate orders knowing they will be scaled back, and banks have an interest in the number being large. A book six times covered and a book twenty times covered are very different states of the world, and the fact that no one can say which it is should temper any conclusion drawn from it.

Bankers also told prospective investors that existing backers would be prioritised in allocation, meaning some new buyers will receive less stock than they asked for. Prioritising existing holders is a sign of relationships being honoured, but it also reduces the number of genuinely new, price-insensitive buyers in the aftermarket.


The Climbdown From $2 Trillion

The most concrete evidence of cooling sentiment is the price range itself. In early October, Wall Street estimates for Anthropic's listing valuation ran as high as $2 trillion. The range now on the table tops out at $1.4 trillion fully diluted.

During the market turmoil of the week beginning 9 November, prospective buyers reportedly discussed figures of $1.2 trillion to $1.3 trillion, with some considering as low as $1.1 trillion. CNBC reported that some shareholders described even $1.5 trillion as a stretch.

Seen from the other direction, the deal is still a substantial markup on the $965 billion valuation Anthropic announced in a May 2026 funding round, achieved in roughly six months. And it sits below the approximately $1.75 trillion SpaceX sought when it listed in June 2026, which gives some sense of the scale at which the very largest private companies are now coming to market.


The Nvidia Shadow

No IPO is priced in isolation. Nvidia's week was close to a worst case for sentiment. After Huang's resignation on 12 November, the shares fell more than four percent on the day having been down as much as fourteen percent intraday, and the market capitalisation closed below $4 trillion for the first time since July 2025. Bernstein halved its price target from $400 to $200. The stock fell a further 4.12 percent on Monday 16 November to its lowest close since 1 July 2025, and is down more than twenty percent over the past month.

There is no public evidence that bankers explicitly cited Nvidia's situation when setting Anthropic's range, but the two situations are directly linked. Nvidia is one of the seven buyers in the $26 billion concurrent private placement, and Anthropic agreed to pay Nscale up to $44.6 billion through 2033 for Nvidia-powered computing capacity, part of $900.9 billion in disclosed purchase and lease commitments. They also share a shareholder base and a narrative. When the sector's most visible stock is in a six-week decline, the marginal buyer of a trillion-dollar AI listing demands a discount.


Who Controls the Company

Anthropic's seven co-founders, all former OpenAI employees including chief executive Dario Amodei and president Daniela Amodei, will retain 50.1 percent of voting power on general shareholder matters after the offering. The chief financial officer is Krishna Rao, and the company was founded in 2021.

That structure means outside shareholders, however much capital they contribute, cannot outvote the founders on ordinary matters. Dual-class voting has become standard in large technology listings, and buyers accept it in exchange for access. It is worth being clear about what it means in practice: public investors are providing capital and bearing price risk without a meaningful mechanism to force change if strategy or spending disappoints.


Key Takeaways for Investors

  • The headline multiple of about 17 times revenue rests on an annualised run rate extrapolated from 28 days of consumption, not on audited trailing sales. Against the $33.24 billion recorded over nine months, the valuation looks considerably more demanding.
  • The $80.7 billion half-year net loss is dominated by non-cash revaluation of convertible instruments. The $693 million operating loss, narrowed from $2.1 billion, is the figure that describes the business.
  • The $26 billion private placement comes largely from suppliers and cloud partners with commercial exposure to Anthropic's spending, and it carries no lock-up. It is strategic capital rather than an independent valuation endorsement.
  • The move from $2 trillion estimates in early October to a $1.3 trillion to $1.4 trillion range, with buyers discussing figures as low as $1.1 trillion, is the clearest measure of how much sentiment has shifted in six weeks.
  • Oversubscription claims range from about six times to twenty times. Unverifiable order books tell you demand exists; they do not tell you what price that demand clears at once shares trade freely.
  • A float of roughly six percent and prioritised allocation to existing backers both support the price initially and limit what early trading reveals about genuine market-clearing value.

Conclusion

The interesting question is not whether Anthropic's shares rise on their first day. With a small float, a stabilisation option and a book several times covered, a first-day gain is close to engineered. The question is what happens when the supply of stock expands and the strategic holders, who face no lock-up, decide what to do.

This deal is being described as a referendum on AI investor appetite, and in one narrow sense it is. But the more precise reading is that the market has already voted, and it marked the asking price down by roughly a quarter from where it stood six weeks ago while the underlying business grew its run rate by about $10 billion. Revenue accelerated and the valuation fell. That is not a loss of faith in AI economics. It is the beginning of a negotiation about what those economics are worth, conducted for the first time in public, with a price printed every second.