Anthropic’s $1.5 trillion valuation is also on the line

Produced | MiaoTou APP

Author | Zhang Beibei

Editor | Ding Ping

Header image | Visual China

Wall Street has been waiting for this moment for a long time.

On the evening of October 26, Anthropic submitted an S-1 prospectus to the SEC and applied to be listed on Nasdaq under the code ANT.

An AI company with half-year revenue of US$16.3 billion, a year-on-year increase of 1,300%, Q2 single-quarter GAAP operating profit has turned positive($974 million), Q3 continued to be profitable on an adjusted basis. The primary market has long given it a valuation estimate of US$2 trillion; however, with the recent Q3 quarter-on-quarter growth slowing down from 144% to about 45%, some investors have adjusted their expectations to US$1.5 trillion.

Currently, the number of shares to be issued, issuance price and target valuation in the prospectus are still blank, and the price will not be filled in until the roadshow inquiry is completed.

What the prospectus will really directly change is the information structure on which pricing is based. Those variables that can be folded into the general narrative during the private equity stage, such as nearly half of the revenue settled through cloud channels, about US$600 billion(Total based on disclosed contracts)The take-or-pay computing power commitments, etc., are now broken down item by item and placed in a legally responsible public document.

Institutional investors participating in the road show will make new bids after these variables are spread out.

So, how much is Anthropic worth?

Three questions, three prices

The market has previously given a private valuation of up to US$1.5-2 trillion. After the prospectus showdown, this price may become untenable.

To determine how much it is worth, you need to first understand three questions: Is it really making money? Does it get all the money you make? Can I participate in decision-making after purchasing?

Think of these three questions as three valuation gates. Each time a gate passes, a price will be opened, and every time a gate passes, a price will be cut:

(Chart production: Huxiu Miaotou)

If Anthropic were an ordinary company that made steady money, all the money was pocketed, and shareholders had the final say, it would be worth US$1.5-2 trillion; but it has only passed halfway through the three gates, so the true valuation may be closer to US$1.11 trillion-1.48 trillion.

Just like buying a milk tea shop. If the boss tells you that the store is making steady money, all the money goes into your pocket, and you decide on major events, you are willing to pay 2 trillion. But what he actually said is that he has just made money in the past three months, the platform takes 20% and still owes ten years of tea money, and the store has the final say. Then you have to discount and not pay that much.

What needs to be done next is to check which items have not been fulfilled item by item.

Let’s look at the first course first.

Is it making steady money?

Corresponding to what the owner of the milk tea shop said: “I just made money in the past three months.”

Let’s first look at the first of the three gates. Is it making money steadily?

Anthropic’s 2026Q2 single-quarter GAAP operating profit turned positive at US$974 million. In Q3, it was disclosed on the previous investment day that it would continue to be profitable on an adjusted basis. This is the first report card of a large model company to report positive profits. The primary market dares to give 1.5-2 trillion US dollars, and a large part of the bet is on this number.

However, the positive operating profit in a single quarter does not mean that Anthropic has the ability to make stable profits.

Looking at the S-1 prospectus submitted on October 26, we can find that Anthropic’s current profit improvement is mainly due to revenue growth faster than the increase in computing power costs. Whether it can be sustained in the future still faces two tests.

(1) Revenue growth is highly dependent on new models.

There is an easily overlooked structure hidden in the prospectus: Anthropic’s revenue consists of API calls and subscriptions, of which API accounts for 80%. Enterprise customers pay according to the actual call volume, and whichever model is stronger will be used; although subscription revenue is more stable, the proportion is relatively small and cannot support the overall growth rate.

In this case,The company must keep releasing new models like running a relay race.Because once the iteration speed is half a beat slower, or a certain release is not as good as expected, the API call volume will immediately migrate to competing products, and revenue may instantly stall. This is also an important reason why more than 80% of its 2026Q3 revenue will mainly come from models released in the past 6 months.

This kind of income structure obviously cannot support the narrative of making stable money.

(Chart production: Huxiu Miaotou)

(2) The bills for contracts with large computing power have just begun, and it is difficult to sustain profits.

In the single quarter of 2026Q2, GAAP operating profit turned positive at 974 million. In Q3, it continued to be profitable on an adjusted basis, making money for two consecutive quarters. An important reason behind this is: revenue growth(1368% year-on-year)Temporarily outperforming the locked computing power cost growth rate(about 300%), is a periodic time difference bonus.

But whether it can be sustained is highly uncertain.

(Chart production: Huxiu Miaotou)

Because computing power is usually signed 12–18 months in advance, computing power costs such as rental fees and electricity bills are locked in first, and the income cannot be realized until the model is released.

Starting from the second half of 2026, large computing power contracts will gradually enter the billing period: Google Cloud in April, AWS in May, and Broadcom equipment rental have all begun to incur fees. The money is not a lump sum. Google and AWS are minimum commitments that must be paid regardless of whether they are used or not. Nscale and Lambda are billed based on delivery acceptance. Space XAI will not enter the peak monthly fee until the capacity is delivered at the end of the year.

In other words, the 8.945 billion computing power expenditure in the first half of 2026 is just the “down payment” for hundreds of billions of bills in the future.

The real big head may be after 2027.

Even more troublesome are mismatches. Most of the above-mentioned computing power commitments are rigid take-or-pay expenditures, and 80% of the company’s revenue comes from API calls based on volume, without long-term price lock contracts. The cost curve only goes up and down, but the revenue fluctuates with model iteration and call volume.When computing power bills reach their monthly payment peak in 2027 and revenue growth slows down, GAAP profits will most likely be squeezed or even turn negative again.

(Chart production: Huxiu Miaotou)

Therefore, the result of the first gate is: half open. It’s really making money, but it’s not stable yet.

所以Didn’t pass the first gate completely, based on the valuation of 1.5-2 trillion, we need to make a cut first.

If there is a 95% discount, the valuation may be revised down to US$1.425-1.9 trillion; but if future profits turn negative and the gate is closed, then the discount will continue to be revised downward. If there is a 10% discount, the valuation may be revised down to US$1.35-1.8 trillion.

On the basis of the above, let us look atThe second gate.

Does it get all the money you make?

並不是。

Just like a milk tea shop sold 100,000 cups this month and had a turnover of 300,000, which sounds very impressive; but 150,000 of them were sold through a takeout platform, and the platform takes a cut; in addition, in order to lock in low-priced tea, the boss signed a dead contract three years ago that required a monthly payment of 50,000 until 2030. So the turnover of 300,000 yuan and the money the boss actually pocketed are completely different things.

Same thing with Anthropic’s $16.3 billion in revenue.

The prospectus disclosed that its sales through AWS and Google Cloud Market in 2025 will be approximately US$2.16 billion, accounting for 47% of annual revenue; in the first half of 2026, this proportion will further increase to approximately 48%. However, Anthropic records revenue based on the gross method, and records the channel share distributed to cloud vendors as sales expenses.

In this case, for ARR(annualized income)For market investors who value the company, it is obvious that the previous pricing was too high.

Therefore, the second gate failed.

For reference, the channel fee in 2025 is US$351 million, accounting for 16% of the current cloud vendor sales of US$2.16 billion. In this case, the net impact on overall revenue is ≈ 16% × 47%(Proportion of cloud vendor sales in the current period)≈ 7.5%, so the corresponding revenue and ARR need to be discounted by about 92%.

Correspondingly, the valuation may continue to be revised downward based on the previous first cut: ① The first gate is half-open: 1.31 trillion-1.75 trillion U.S. dollars; ② In a pessimistic scenario, the first gate is closed: 1.24 trillion-1.65 trillion U.S. dollars.

Finally, let’s look at the third gate.

Can I participate in decision-making after purchasing?

The prospectus clearly gives the answer, no.

You spent 1 million to buy 30% of the shares of the milk tea shop. It stands to reason that you would have a vote in replacing the refrigerators in the shop, raising wages, or even selling the branch. But the boss said: “I alone have the final say in the store. Just take the dividends and don’t interfere with anything else.” Then you are definitely not willing to pay 1 million, and you may only be willing to pay 800,000, because the right to speak is also money.

Anthropic is even more extreme, with seven co-founders through an LLC(Limited Liability Company)Holding one Class F stock locks in 50.1% of the voting rights, and you as a shareholder don’t even have the right to vote to remove the CEO.Because of the shares issued in the IPO, voting rights are diluted to a ceiling of 49.9%, and a majority can never be formed.

Therefore, the result of the third level gate was not passed.

The key is how much discount is this confiscated “right to speak” worth?

There is currently no precise formula, but academic literature gives an interval:

• Cremers and other scholars’ research on the valuation life cycle of dual-class companies found that there is a valuation premium in the early stage of listing, but the discount gradually appears after 6-9 years, and stabilizes at the 15%-20% level in the later period;

• The CFA Institute’s Corporate Governance White Paper lists non-veto rights minority shares as a secondary discount factor, but does not provide a unified discount quantification standard.

• The OECD corporate governance principles clearly include different rights for the same shares as a valuation adjustment item.

我們取 15% As the third level discount range.

In this case, based on the second gate, after another cut, the valuation may be further revised down to: ① The first gate is half-open: 1.11 trillion-1.48 trillion U.S. dollars; ② In a pessimistic scenario, the first gate is closed: 1.05-1.4 trillion U.S. dollars.

Stacking the three gates is the final price.

(Chart production: Huxiu Miaotou)

寫在最後

Back to the three questions at the beginning:

(1) Is it really making money?Earnings are possible, but they are only temporary and there is great uncertainty as to whether they can continue.

(2) Does all the money earned belong to it?Otherwise, after deducting the cloud vendor channel revenue sharing, a 92% discount will be required.

(3) Can I participate in decision-making after purchasing?No, so I’ll give you another 15% off.

Multiply the three answers and you’ll get Anthropic’s likely true pricing.

Starting from the completely outdated 1.5-2 trillion, first cut the premium of money printing machines(Half open −5%), and then divide the channels into(−8%), and finally deduct the management discount(−15%), only about 74% of the valuation weight is retained after three consecutive multiples.

Therefore, when the three realities of unsustainable profits, channel sharing elimination, and governance discounts come to the fore, the valuation will most likely be revised down to US$1.11-1.48 trillion.

This may not be the end.

The prospectus itself hints that revenue growth may decline and losses may continue. If adjusted profits turn negative again in the future, this pricing may continue to be revalued downward to US$1.05 trillion to US$1.4 trillion.

Therefore, if the market still uses the ARR pricing method, there is a high probability that this company will be systematically overvalued, and you need to be cautious when making new investments.