#Gate直通IPO第三期Anthropic Valuation shrinks by $600 billion, 7 founders prepare to cash out $250 million each—is Anthropic’s once-in-a-century IPO still attractive?

Anthropic is finally going public, but the price is a little hard to swallow. On November 10, Anthropic officially kicked off its IPO roadshow: 825 million shares at $70 to $78 each, raising up to $64.4 billion. In the proposal Anthropic sent to investment banks the night before, that figure was still $69.3 billion. Nearly $5 billion evaporated overnight. Anthropic’s own valuation also fell from the $2 trillion still being discussed by the market just two weeks ago to around $1.4 trillion, a drop of about 30%. That is already below SpaceX’s market cap of around $1.77 trillion at its IPO. More subtly, the founding team is cashing out even as the valuation is being cut.
According to the updated prospectus, 7 co-founders will sell existing shares in the IPO, cashing out a total of about $1.75 billion, or around $250 million each. Is this once-in-a-century IPO still attractive?

01 —Just how big is this once-in-a-century IPO?
First, the scale. In June, SpaceX set a global IPO record with a $75 billion offering, which came to around $85.7 billion including the overallotment. Anthropic’s public offering comes to around $64.4 billion at the top of the range; even if underwriters fully exercise the overallotment, it would only reach about $74 billion. Looking at the public offering alone, the record for the largest fundraising has not been broken. But it is still the second-largest IPO in history, about 2.5 times the size of Alibaba’s 2014 listing and 4 times Facebook’s 2012 listing.
What actually pushes the total raised past SpaceX is a concurrent private placement. This private placement totals $26 billion, subscribed to by seven tech giants at the IPO price. Google and Nvidia are putting in $10 billion each, while AMD, Broadcom, Micron, Microsoft, and TSMC split the remaining $6 billion. The public offering and private placement together bring the fundraising close to $100 billion—it really is money power at full throttle.
However, all seven private placement investors are upstream suppliers in Anthropic’s compute supply chain. Suppliers use their money to buy shares in their customer, and the customer then uses the money to buy compute from those suppliers. This model is called “circular trading” in the AI industry. The money seems to keep growing, but may simply be circulating within the supply chain. If end demand fails to keep up, both upstream and downstream players could be dragged down together.
So this IPO has slammed on the brakes amid a headlong rush. It looks glamorous on the surface, but there is no shortage of internal problems. Why has the capital market suddenly become so cautious? Anthropic’s operating performance is actually not bad; its revenue growth has been outrageous. /* QUOTE */ $386 million for all of 2024, $4.585 billion in 2025, and about $17 billion in a single quarter by the third quarter of this year. As of the end of October, annualized revenue was around $81 billion. Existing customers are also using more and more. Based on the prospectus, net revenue retention is about 900%. The same enterprise customers spend around 9 times as much on Claude a year later. Operating cash flow was also positive in the first half, at about $3.9 billion.
But Anthropic’s actual situation is not as rosy as its revenue curve suggests.
 First, growth is slowing.
Revenue grew 144% quarter over quarter in the second quarter, but that fell to 47% in the third quarter.
And Anthropic has already given back the money it just made. It posted its first operating profit of $974 million in the second quarter, then swung back to a loss of $567 million in the third. The main reason is that compute is too expensive and demand is growing too fast, forcing Anthropic to buy around $5 billion in short-term compute at high prices. Compute costs rose from 48% of revenue in the second quarter to 58%. On top of that, price cuts and contract discounts reduced third-quarter revenue by about $600 million.
Anthropic’s revenue also includes a less flattering source of income. According to the prospectus, in the first nine months, revenue from customers later found to have violated usage policies or commercial terms accounted for around 5% to 7% of total revenue, or about $1.7 billion to $2.3 billion. Most of this money came from users outside the US whose accounts were banned. Banning accounts without refunds has long been one of Anthropic’s go-to moves; who knew it was still a proper line of business on the books?
There is another counterintuitive point: the better Anthropic’s business does, the more it loses on paper. Its paper net loss for the first nine months reached a staggering $104 billion. More than $100 billion of that did not come from operating losses (the business itself lost only about $1.26 billion), but from accounting revaluations due to the increase in its valuation.
What does that mean? In its early days, some of Anthropic’s funding came in the form of “money now, shares later.” In other words, investors put in money upfront, with an agreement to convert it into company shares later under predetermined terms. These are the convertible notes and SAFEs mentioned in the prospectus.
The problem is that each quarter, the value of these “shares to be issued in the future” has to be recalculated using the latest valuation. In just over a year, Anthropic’s valuation rose from over $100 billion to more than $1 trillion. The value of those shares soared along with it, and the increase has to be recorded as a loss.
It’s like borrowing $100,000 from a friend to open a store, with an agreement that they can later convert it into a stake in the business. The more successful the store becomes, the more that stake is worth, and your books have to record it as a “loss.” That’s why the pressure from paper losses in Anthropic’s financial statements is so high.
On top of these pressures, Anthropic is also saddled with a compute bill of nearly $900 billion. This represents long-term purchase and lease commitments it signed with Google, Amazon, Broadcom, and others. Even if revenue stops growing, it still has to pay.
With such high spending, Anthropic has another, more fatal weakness: its customers and products are too concentrated. Its top five customers contributed about a quarter of revenue in the first nine months; more than 80% of third-quarter revenue came from models released in the past six months.
In other words, Anthropic has to keep running faster than everyone else. If it has a gap in new model releases, customers will leave in a heartbeat, and everything could be at stake in an instant.

02 —How will the share price move?
Before discussing price movements, let’s break down the key figures in this prospectus.
Anthropic plans to issue 825 million Class A shares, around 801 million of which are newly issued by the company. At the midpoint price of $74, this portion could bring the company about $59.3 billion. That money will be the fuel Anthropic needs for its upcoming compute battle. But the remaining 23.648 million shares are different. These are existing shares, and all of them will go into the pockets of the 7 founders.
Besides the share composition, another point worth watching in this offering is the post-listing share price stabilization mechanism. To prevent the share price from going on a “roller coaster” after listing, Anthropic has given the underwriters, led by Goldman Sachs and Morgan Stanley, a 30-day option to buy up to another 123.75 million shares at the offering price if necessary. This is called the greenshoe mechanism.
If the share price falls, the underwriters buy shares on the market to support it; if it rises, they get the company to issue the additional shares at the offering price. It’s like buying insurance for the share price: someone steps in if it falls, and the company can raise more money if it rises.
However, this insurance only lasts 30 days. SpaceX also used a greenshoe when it went public, but still fell below its offering price in mid-July.
So, how will Anthropic’s share price move after the 30-day insurance period ends?
First, the good news. Anthropic shares are in high demand right now. As soon as the roadshow began, orders were already fully subscribed within the price range, and institutional demand is even stronger than it was for SpaceX. But a big first-day jump is not guaranteed. First, the broader market is in bad shape. On November 5, the Nasdaq plunged 5.5% in a single day, its biggest drop in 19 months, with AI chip stocks tumbling across the board. Listing just as AI stocks are getting hammered is hardly ideal timing.
Second, SpaceX’s experience is still fresh. Its shares surged on the first day after listing in June, then fell below the offering price in July. Once the excitement fades, the market still looks at actual results.
There will also be a wave of selling pressure on the first day of trading. Under the prospectus, some directors, executives, and employees will be able to sell up to 10% of their holdings starting on the first day of listing.
Many Anthropic employees hold shares, most of which they received when the company’s valuation was still very low. Selling on listing day would let them make a big profit. If many people sell at once, the share price will be pushed down.
So, in the short term, sentiment matters; in the long term, it comes down to performance. The current offering price has already factored in a lot of expectations. The investment banks priced Anthropic based on its own revenue forecast for 2028: $190 billion to $200 billion.
As of the end of October, its annualized revenue was only about $81 billion. In other words, revenue needs to more than double over the next two years for today’s price to hold up. This IPO will affect more than just Anthropic itself.
Among the world’s leading large-model companies, Anthropic is the first to go public. Its share price is effectively setting a “benchmark” for the entire AI industry. If Anthropic rises, OpenAI’s valuation will have more support; if Anthropic falls below its offering price, the entire AI sector will be dragged down. Hong Kong-listed Zhipu and MiniMax won’t escape either.

03 Fable 5.2 drops the day before the roadshow
The day before the roadshow, Anthropic released its latest flagship model, Fable 5.2, just two months after the previous generation, Fable 5.1.
In terms of speed, it generates output up to 30% faster than 5.1; a new “fast mode” can speed it up by another 2.5 times.
In terms of price, input costs $8 per million tokens and output costs $40, reportedly 20% less than the previous generation.
In benchmark scores, Fable 5.2 scored 65 on the intelligence index from third-party firm Artificial Analysis, ranking first and ahead of Google Gemini 4 Carbon, which scored 60.
The most notable improvement is in coding. In the official FrontierCode 2.0 test, Fable 5.2 scored 49.1%, compared with just 32.4% for the previous generation, and also outperformed OpenAI’s GPT-6 Astra.
But benchmark scores are just on-paper results. How does it actually perform in practice?
Several viral hands-on tests on X gave it very positive ratings.
But as capable as Fable 5.2 is, Anthropic is actually deliberately holding it back. During training, Anthropic deliberately limited Fable 5.2’s capabilities in “accelerating AI development,” such as optimizing chip compute kernels and designing large-model architectures, keeping it at the level of Opus 5.5. Racing to build ever-better models while deliberately throttling its strongest model is the most contradictory thing about Anthropic.

04 Anthropic is rushing to list—why isn’t OpenAI in a hurry?
This year, three trillion-dollar companies are lining up to go public: SpaceX, Anthropic, and OpenAI. SpaceX went first in June, Anthropic followed close behind, while OpenAI chose to stay in the private market. On November 6, OpenAI completed the first closing of its latest funding round: $11 billion at a valuation of $1.25 trillion.
One is willing to go public even at a discount; the other would rather stay private than accept one. Why?
In fact, OpenAI has been preparing to go public for a long time. In June, a week after Anthropic, it also confidentially filed an IPO application. But in September, Altman said publicly that it would not go public in 2026. He cited AI safety: with so many safety issues right now, going public would be “unwise timing.”
But that may only be one reason. There are reports that advisers gave OpenAI two options long ago: wait until 2027 and go public at a trillion-dollar valuation, or lower its valuation and list as soon as possible. Altman rejected the latter. OpenAI’s CFO also privately recommended delaying, on the grounds that the company’s financial disclosures were not yet up to the standards of a public company.
Put simply, valuations in the private market are set through round after round of negotiations, without having to face scrutiny from shareholders every day. Once you go public, people score you every day, and every rise and fall is there for all to see.
So why does Anthropic insist on going public now? I think there are three main reasons.
First, it needs too much money.
As mentioned earlier, Anthropic is saddled with nearly $900 billion in compute commitments, while it has only about $132 billion in cash and short-term investments. Raising money round after round in the private market cannot keep up with its spending. This IPO and private placement will bring in nearly $100 billion at once.
Second, Anthropic’s books look good enough.
Going public means laying the books open for everyone to see. Anthropic is willing to go public because its revenue looks good, and it even briefly achieved operating profitability in the second quarter. OpenAI is under far more pressure: it expects to burn through about $278 billion in cash from 2026 to 2030.
Third, getting there first.
A huge pool of money around the world has been waiting for a “pure AI stock.”
The three companies are valued at more than $4 trillion combined, but there is only so much money in the market. Whoever lists first will be the first to capture that pool of capital.

05 Who is the real money-printing machine in the AI era?
Looking at the numbers alone, Anthropic is already ahead. As of the end of October, Anthropic’s annualized revenue was about $81 billion; the figure OpenAI gave investors was about $50 billion as of the end of September.
But these two figures cannot be compared directly, because the two companies account for revenue differently.
For example, if a customer spends $100 on Claude through a cloud platform, Anthropic records the full $100 as revenue and records the roughly $16 commission paid to the platform as a cost. OpenAI sells the same thing but only records the amount it receives after the commission. If OpenAI used Anthropic’s accounting method, the market estimates its figure at the end of September would be around $70 billion. Anthropic was about $71 billion at the same time.
In other words, the two companies’ revenue is actually about the same. The real difference is that their ways of making money are completely different.
However, OpenAI is catching up in the enterprise market. Since the third quarter, its enterprise revenue has more than doubled, now accounting for over half of its total.
Each model has its own weak points.
Anthropic’s money comes in fast, but it relies too heavily on big customers and new models. If its models fall behind, customers can walk away just like that.
OpenAI has many users, but most are free users. Supporting 1.2 billion people means compute alone costs an astronomical amount.
So, who is the money-printing machine?
Honestly, at this point, neither is. Nearly 60% of Anthropic’s revenue goes to compute costs, and the faster its revenue grows, the faster its compute bill grows; OpenAI still has several more years of spending to go. The companies collecting steady money are the ones selling them chips and cloud services. Google, Nvidia, Microsoft, TSMC… In this AI gold rush, it’s still the people selling shovels who make money first.

06 Conclusion
A company founded less than six years ago has reached a trillion-dollar valuation and made it onto the Nasdaq. AI is no longer just a technology in the lab, nor just a chat tool on your phone—it has become Wall Street’s most expensive business.
That is certainly exciting. But the faster money comes in, the more important it is to stay calm. A company’s valuation can swing by hundreds of billions of dollars in just a few weeks, showing that even the most professional investors don’t have a definitive answer.
This time, Anthropic may offer 15% to 20% of its shares to retail investors, far more than the 5% to 10% typical of an IPO. For many, this could be their first chance to buy shares in a leading AI company directly.
But being able to buy doesn’t mean it’s worth chasing. In the early days after a new stock lists, few shares are available to trade and sentiment is high, magnifying both gains and losses. If you chase the price up, you could easily get stuck holding the bag.
More importantly, you’re not buying a company that is already highly profitable; you’re buying into expectations for the next several years. The share price holds up only if those expectations are met; if they fall short, the price can drop just as ruthlessly.
So if you really want to buy, think through two things first: can you accept losing this money, and are you bullish on the AI industry or just the excitement around this stock’s first day of trading?
But one thing is certain. Over the past few years, we could only hear these companies tell their own stories about whether AI is a good business. After November 19, the market will answer that question every day.