🎧 Listen to the Discussion(two hosts talk through their thoughts on this article)

Audio only — for listening on a walk or while doing chores

This is an AI-generated audio overview of this article, created with Google NotebookLM. Two hosts discuss the article together, sharing their own thoughts and commentary — it is not a reading of the article text.

The audio covers the article as first published.

Series: Following Anthropic

On the morning of 28 October, I noticed a subheading in a Bloomberg article that read, in the Japanese edition, "Financial content well received". Anthropic, the American company that makes Claude, the partner I work with every day on this blog, had published the prospectus for its stock market listing two days earlier, and the article said that investors who had studied it closely had reacted mostly favourably.

In my previous article, I read the "Risk Factors" chapter of the same prospectus and looked at the dangers of AI itself. This time, I wanted to read the other pillar of the document, the financial figures. I am not an accountant, however. So I decided to read it not with the eyes of a professional investor, but with the eyes of someone who keeps a household ledger.

Yet while I was letting the finished draft rest for a few days, the figures themselves moved. On 2 November, Anthropic filed an amended prospectus and filled in the July–September figures that had been left blank. They showed a side of the company that I could not have seen from the October document alone. The same company's papers changed their expression within a single week. That fact itself became one of the things I most want to share in this article.

When I was running a business, what I worried about most each month was not the profit in the books but how much was left in the bank account. Even in a month with a profit, a late payment from a customer could leave me in a cold sweat on payday. That feeling became my yardstick for reading this prospectus.

Two disclaimers. I researched and wrote this article together with that same Claude. Claude is an Anthropic product, and I am also a user who pays a monthly fee for it. So that I would not favour my own side, I tried to give the points investors praised and the points that worried them the same weight. Secondly, this article is not a recommendation to buy or sell shares. Nor will I say here whether I intend to buy shares in this company.

Six Numbers from the Prospectus to Start With

On 26 October, US time, Anthropic published its registration statement for listing (Form S-1), and on 2 November it filed an amendment. From a document of more than 300 pages, these are the six numbers that form its skeleton.

  • Revenue: $16.3 billion in January–June 2026, about 14 times the $1.16 billion of the same period a year earlier. In the three months from July to September alone, it earned about $17.0 billion more ($16.956 billion)
  • Operating result: In April–June it made an operating profit for the first time, of $974 million. In July–September, however, it went back to an operating loss of $567 million
  • Net loss: $80.7 billion in January–June. About $72.4 billion of that was a paper loss from changes in the value of convertible notes and similar items. The net loss for July–September was $23.3 billion
  • Cash in hand: $131.7 billion at the end of June ($112.4 billion in cash and $19.3 billion in short-term investments). The money raised so far totals $141.1 billion
  • Promises for computing power: Purchase and lease commitments for computing power and equipment, made by the end of September and afterwards, total about $895.4 billion
  • Price: The second amendment of the prospectus, filed on 10 November, wrote in a share price range of $70–78 and an offering of 825 million shares. According to reports, the listing date is expected to be 19 November

When I started reading, my eyes swam. A loss of $80.7 billion in half a year is more than 12 trillion yen. Yet the same document says the company has $131.7 billion in cash. Why does a loss-making company have so much money left? And why did a company that had just turned a profit go back into the red three months later? I will untangle this slowly from here.

What Investors Praised, and What Worried Them

According to Bloomberg, the October document appeared after AI-related shares had been sold for about two weeks over fears about huge spending and borrowing for data centres. The document's contents were reported to have eased those fears for a while. After the November amendment, however, investors' voices became a little more cautious. I have put both sides into tables.

Points that were praised

Point Figure in the prospectus Why it was praised
Revenue growth $16.3bn in Jan–Jun, about 14 times a year earlier. Jul–Sep about 12 times a year earlier Already large in scale; not a company built only on hopes
Spread among businesses Customers paying over $100,000 a year passed 11,900 (about 950 a year earlier) It has begun to put down roots in many companies, not just a few big ones
Spread of customers The top five customers made up about 25% of revenue (Jan–Sep), and about 21% in Jul–Sep alone. No single customer exceeds 10% Losing one customer would not shake it at once
Growth in use Companies that were customers a year ago now pay about 9 times as much (net revenue retention of about 900%) Even if new customers slow down, existing ones are using more
Cash in hand $131.7bn. Money raised so far $141.1bn The difference suggests it has really used just over $9bn to build the business
Efficiency of computing Compute cost fell from 251% of revenue in Jan–Mar 2025 to 48% in Apr–Jun 2026 Revenue overtook the cost of computing

Points that caused concern

Point Figure in the prospectus Why it caused concern
The profit did not last An operating loss of $567m in Jul–Sep. Compute cost rose again to 58% of revenue It bought $5bn of expensive short-term computing power, and gave up about $600m of revenue through price cuts and contractual discounts
The weight of promises Commitments for computing and other items of about $895.4bn, about 13 times even Jul–Sep revenue multiplied by four About 80% of revenue is paid according to use, and customers can cut back at short notice. Costs are fixed while revenue varies
Slowing growth Growth on the previous quarter fell from about 144% to about 47% Big customers switching to their own AI and price cuts by competitors came together
Dependence on the big clouds 48% of Jan–Jun revenue came through the marketplaces of AWS, Google and Microsoft They are investors, suppliers and competitors all at once
Governance The seven co-founders hold 50.1% of the votes through Class F shares Ordinary shareholders cannot decide votes at shareholder meetings
Price Some big investors said a $1.5 trillion valuation "needs to come down" Reports of a fall to near $1 trillion, and even a possible delay

Putting the two tables side by side, I noticed something interesting. Many of the praised points and the worrying points are two sides of the same number. That 80% of revenue is paid according to use is a strength that makes revenue jump when customers grow, and a weakness that makes it shrink fast when they leave. The compute cost ratio was a praised figure six months ago, yet in the document a week later it had turned into a cause for concern.

If I had to sum up my impression of these tables in one sentence, it would be this: "a healthy young runner with fast legs has signed a contract to carry a very heavy load." The runner's legs are real, and the savings are ample. But nobody yet knows exactly how heavy the promised load will be. From the next section, I will look at what is inside, one piece at a time.

An $80.7 Billion Loss and $9 Billion — Why Do the Two Numbers Differ?

Why was a company that lost $80.7 billion in six months said to have "really spent only $9 billion"?

The answer lies in the Loans that come with a promise that they can later be turned into shares on terms set in advance. The more the company is worth, the more that promise is worth. Anthropic used this method to raise money, including notes issued to Amazon and others. that make up most of the loss. When Anthropic raised money, it borrowed on the promise that "when the company lists, this loan can be turned into shares on agreed terms." When the company's value rises sharply, the value of that promise swells too. Accounting rules record the swollen amount in the books as a "loss" on the company's side. Yet no money left the safe when that happened. According to the prospectus, paper losses of this kind came to about $72.4 billion in the first half alone.

It is like a small shop on a high street promising a friend, "If the shop does well, I will sell you a share of it at today's price." If the shop then does extremely well, the value of that promise leaps, and a large "loss" is written into the books. Yet not a single penny has left the till.

So how much money really went? Yuri Khodjamirian, chief investment officer at the US asset manager Tema ETFs, did a very simple subtraction. Take the $141.1 billion raised so far, and subtract the $131.7 billion in hand now. The remaining $9.4 billion or so, he said, is the net amount the company has used to grow this far.

Horizontal bar chart. Of the first-half net loss of $80.7 billion, about $72.4 billion is a paper loss on valuations, while about $9.4 billion, the money raised of $141.1 billion minus cash of $131.7 billion, is the money actually spent

Top: the net loss for six months. The dark part, about $72.4 billion, is a paper change in valuations. Bottom: the money raised so far minus the cash in hand. The size of the loss and the money actually spent differ by an order of magnitude (a simplified chart based on figures in the prospectus).

This subtraction is only a rough guide. This is because it is the "net outflow" after everything has been added and taken away, including revenue from customers, salaries paid and the cost of computing. Even so, anyone who keeps a household ledger will grasp its weight at once. After all, however long you look at your payslip or the value of your insurance, the balance in your bank book tells you something different.

The headlines spoke of "a 12 trillion yen loss", and at first I felt uneasy too. But after a single subtraction in the margin of a sheet of paper, I could see that more than half of that figure was the "value of a promise" that had swollen because the company was doing well. Faced with a big number in a headline, first do one subtraction. It is one of the habits I would like to build as I grow older.

A quill pen resting on an old handwritten ledger

The numbers in a ledger change their expression depending on who reads them. Before being shocked by the size of a loss, I want to check how much money actually moved.
Photo by camera obscura on Unsplash

251%, 48%, and Then 58% — Read a Line Right to Its Last Point

One of the most praised figures in the October prospectus was the compute cost ratio. It shows what percentage of revenue goes on the computers and data centres that run the AI. In January–March 2025 it was 251%, meaning that the company paid $2.51 for computing to sell $1. It fell quarter by quarter, and in April–June 2026 it reached 48%. Sell $1, and pay 48 cents for computing. Revenue had overtaken the cost of computing.

Yet the July–September figure written into the 2 November amendment was 58%. A line that had been falling turned upwards at its last point.

Line chart. Compute cost as a share of revenue falls from 251% in Q1 2025 to 177%, 148%, 144% and 72%, reaches 48% in Q2 2026, and rises to 58% in Q3 2026. Only the last point is in a different colour

Compute cost as a percentage of revenue. The October prospectus ended at 48% for April–June. The 58% for July–September is the point added by the amendment of 2 November. Below the 100% line, revenue is larger than the cost of computing.

The amendment also gives the reason. Demand grew faster than the computing power it had secured in advance, so it bought about $5 billion of extra computing power on short-term contracts. That short-term computing power, it says, was "significantly more expensive" than its multi-year contracts.

Compute was not the only reason the operating result turned back into a loss. According to the amendment, there was also about $2 billion of non-cash expense linked to a donation-matching programme (when employees pledge part of their equity awards to charity, the company adds shares of its own). It also pulled down the July–September operating result. Like the valuation losses in the earlier section, it is an expense that involves no cash leaving the company, mixed into the apparent size of the loss.

This is very like train tickets in Britain. In Britain, an "Advance" ticket bought ahead for a fixed day and train and an "Anytime" ticket bought at the station on the day can differ greatly in price for the same journey. On a day when far more passengers turn up than expected, buying seats with on-the-day tickets lets you carry everyone, but the profit per passenger falls. Anthropic chose not to keep its customers waiting. That decision makes sense. Yet it is also a fact that it turned three months of figures from profit back into loss.

What made me think most in this section was the way we read a chart. When we see a line that keeps falling, we tend to assume it will keep falling. I felt that way too while I was looking only at the October figures. But the last point of a line always stands just before "the next point that has not yet been drawn". The same is true of share prices, blood pressure records and weight charts. I believe a line has not been fully read until you reach its last point, and then the next point after that.

The Third Lens, "Promises" — The Weight Outside the Books

So far I have used two lenses to look at the company's money. The first is profit and loss, whether it made or lost money in the books. The second is cash, how much is left in the safe. Yet I think the heaviest thing in this prospectus is what you see through a third lens. That lens is "promises", the contracts for future payments that have already been signed.

According to the November amendment, purchase and lease commitments for computing power and equipment come to about $895.4 billion in total. "Substantially all" of this is for computing power and its equipment. The commitment to Google has grown to about $121.3 billion, and there are also large multi-year commitments with Amazon, Microsoft and the chipmaker Broadcom. That is more than six times the $131.7 billion in cash. Only about half of it (about $447.5 billion) is expected to fall due by the end of 2030; the rest comes in later years.

Three lenses side by side. The profit and loss lens shows a net loss of $80.7 billion, the cash lens shows cash of $131.7 billion, and the promises lens shows commitments for computing and other items of about $895.4 billion. The size of each circle is in proportion to the amount

One company seen through three lenses. The area of each circle matches the amount. Through profit and loss alone, it is a huge loss; through cash alone, it is well supplied; add promises, and you see the weight of the load it has agreed to carry (simplified diagram).

The contract that stands out most is the one with SpaceXAI, a subsidiary of Elon Musk's SpaceX. Anthropic will buy up to $120.5 billion of computing power running on Nvidia chips by 2029. Once all the equipment is ready in December, the payments will be about $3.1 billion a month, about $9.3 billion every three months. That one company alone comes close to the total compute cost for July–September (about $9.9 billion).

However, this $120.5 billion is a "ceiling". The document says that after the initial periods end in January and March 2027, either side can end the contract with 90 days' notice. Is a big number in a newspaper headline an amount that must be paid in full, or a ceiling that might be paid? When reading a list of promises, I think it is important to tell the two apart.

The promises themselves are not bad. According to the prospectus, for much of 2025 Anthropic could not get enough computing power to meet demand and kept customers waiting. Securing scarce computing power in advance is a sensible decision. What caused concern was the difference in "shape" between the promises and the revenue. Many compute contracts are fixed payments, due whether the capacity is used or not. On the other hand, about 80% of revenue is paid according to how much customers use, and the prospectus itself says that customers can reduce their use at short notice.

This difference in shape is easy to understand if we compare it with our own bodies. Even while we sleep, the body keeps the heart beating, holds its temperature and keeps the brain working. This "energy used just to stay alive" is called basal metabolism, and it is known to account for 60–70% of the energy we use in a day. Meals are not the same size every day, but basal metabolism hardly ever waits. On days when we do not eat enough, the body draws on its reserves. Yet the glycogen that the liver and muscles can store is said to last only about one day in total.

For Anthropic, compute contracts are like basal metabolism. They must be paid whether customers come or not. Revenue is the daily meals, and the $131.7 billion in cash is the body's reserves. Investors valued the cash highly, I imagine, because these reserves looked thick enough to see the company through many days of thin meals. What the cautious ones worried about was that the company had promised to let its basal metabolism itself grow several times larger.

Body and company side by side. The body's basal metabolism matches the company's compute contracts, the body's meals match revenue, and the body's reserves match cash in hand

The flow of energy in the body set beside the flow of money in a company. It splits them into three: payments that cannot stop (basal metabolism, contracts), income that changes day by day (meals, revenue), and reserves for a bad day (body reserves, cash in hand) (simplified diagram).

This comparison also fits our household budgets exactly. In a life on a pension, the money coming in is almost the same each month. That is why it matters whether the household's "basal metabolism", the money that goes out whether we use it or not, such as rent, energy bills, insurance and various subscriptions, matches what comes in.

At 66, the worry "Will the money last?" has not gone away, but it has become something I can live with in a different form. As a first step, I wrote all my monthly and yearly payments on one sheet of paper. Bank transfers in Japan, direct debits in Britain, and subscriptions paid by card. Laid out side by side, they included several small promises I had forgotten I had made. Just as a company's prospectus has a list of promises, a household can have one too. I think the real cause of the worry is usually that this list exists only in our heads.

Glass jars of beans and grains lined up on shelves

Reserves feel real when we think of them not only as an amount but as how many days they will last. The same is true of a company's cash and a household's savings.
Photo by Steven Ungermann on Unsplash

Why Was the Antarctic Depot 11 Miles Short?

While reading about reserves and promises, I found myself thinking of two expeditions that set out for the South Pole. It is the story of the race to the Pole in 1911–12 between the British party of Robert Falcon Scott (1868–1912). A Royal Navy officer and explorer. He led the Terra Nova expedition, which left in 1910, and reached the South Pole on 17 January 1912, about a month after Amundsen; every member of his Pole party died on the way back. and the Norwegian party of Roald Amundsen (1872–1928). A Norwegian explorer. With dog sledges and careful preparation, he became the first person to reach the South Pole, on 14 December 1911. He disappeared in 1928 while flying to help in a rescue..

In the year before the journey, Scott's party laid several depots of food and fuel for the return. The largest was named "One Ton Depot" because more than a ton of supplies was left there. Yet this depot was placed about 50km (31 miles) short of the planned latitude of 80 degrees south. Scott is said to have rejected the advice of a team member, Oates, who urged him to push the ponies on to the last, if need be, to reach 80 degrees.

In March 1912, on the way back from the Pole, Scott and two companions were trapped by a blizzard and died in their tent. The tent stood just about 11 miles (18–20km) short of One Ton Depot. The depot still held plenty of food and fuel. What is more, earlier that same month, an expedition member, Apsley Cherry-Garrard (1886–1959). One of the youngest members of Scott's expedition. In March 1912 he went by dog sledge to One Ton Depot and waited several days for Scott's party, but returned to base without meeting them. He later wrote the expedition account The Worst Journey in the World (1922)., had gone to the depot by dog sledge to meet them, waited several days, and turned back on 10 March.

A sand-coloured triangular tent with a Norwegian flag standing on a snowfield

A tent on a snowfield. The photographer describes it as "a historic Amundsen tent in Antarctica". Even when reserves exist, they cannot save anyone who cannot reach them on a stormy day.
Photo by Dennis Rochel on Unsplash

Amundsen's party took the opposite approach. It laid depots with ample food at every degree of latitude from 80 degrees south, and is said to have marked them with lines of flags on either side so that they would not be lost in a blizzard. More than planned, and easy to find. Some say the difference between the two parties lay less in courage or strength than in where, and how much, spare capacity each kept.

What I took from this story is that reserves matter not only in amount but in "whether you can reach them on a stormy day". Anthropic's cash was almost all cash and short-term investments, money it could use at once. I think investors were right to be pleased that "the depot is full". It was also thanks to this depot that it could buy extra short-term computing power in July–September. The cautious ones were looking further ahead. Can it walk the whole road of $895.4 billion of promises, even when the weather turns? In a storm, can it reach the next depot?

The same question exists in our own lives. There are reserves that have value but cannot be taken out at once on a stormy day, such as a house or pension rights, and reserves within reach of today's wallet. At 66, I have come to think about these two separately.

Switching and the "Moat" — First Inside Companies, Third on Home Screens

One more thing is essential when looking at this company: competition.

Anthropic's strength lies first inside companies. According to the venture capital firm Menlo Ventures, Anthropic's share of what companies paid to use AI language models through APIs rose from 12% in 2023 to 40% in 2025, overtaking OpenAI (27%) to take first place. The 2026 edition, published on 3 November, puts it at 55%. However, Menlo Ventures is also an investor in Anthropic, so I think it is fair to read this figure with a little caution.

On home screens, on the other hand, the picture is different. According to the research firm Similarweb, of all visits to AI chat websites in September 2026, ChatGPT had 56.4%, Google's Gemini 25.0% and Claude 9.6%. That is four times the 2.3% of October 2025, but still third place. This figure covers websites only, and does not include phone apps or the use companies make through APIs.

Two stacked horizontal bar charts. The top chart shows shares of enterprise spending on AI model APIs in 2026: Anthropic 55%, OpenAI 31%, Google 9%, with Anthropic first. The bottom chart shows shares of visits to AI chat websites in September 2026: ChatGPT 56.4%, Gemini 25.0%, Claude 9.6%, with Claude third

The same company ranks differently depending on what is measured. Top: share of what companies paid for APIs (a Menlo Ventures estimate; Menlo is an Anthropic investor). Bottom: share of visits to AI chat websites (Similarweb, September 2026; apps and APIs not included).

So how long will first place inside companies last? This is where the word "moat", often used in the investment world, comes in. Like the moat around a castle, it asks whether a company has defences that others cannot easily break through. In the AI world, there is concern that this moat is shallow. OpenAI and Google are both improving their performance rapidly. And an AI that a company uses through an API can often be switched to another provider by rewriting a few lines of code. If the gap in performance narrows, what remains is competition on price. In fact, according to the amendment, Anthropic cut prices and gave contractual discounts in July–September, giving up about $600 million of revenue.

Even so, the documents also show the company digging its moat. More than half of the companies paying over $100,000 a year now use both the API and Anthropic's own app (about 30% at the end of 2025). If a company uses only one tool, switching is easy, but once a product is built into both employees' everyday work and the company's systems, the effort of switching grows much larger. The company also sells by three routes. One is through developers who try it and spread it inside their companies, one is through the cloud marketplaces of AWS, Google and Microsoft, and one is through big consulting firms such as Accenture and Deloitte.

Yet the second of these three routes has another face. 48% of January–June revenue came through the cloud marketplaces. That share has risen every year, from 11% in 2023 to 32% in 2024 and 47% in 2025. Amazon and Google are major shareholders in Anthropic, suppliers of its computing power, sales channels, and also its competitors in AI. According to the prospectus, there are agreements under which Amazon's convertible notes will not convert into shares beyond a 12% holding, and Google's purchases of new shares will not take it beyond 14.99% (each holder can waive its cap). In addition, Anthropic uses the An accounting method that counts the full amount, before the sales channel's commission is deducted, as revenue. It is said to be used by the party that sets the price and is responsible for the service. Compared with counting only the amount after commission, revenue looks larger. for sales through these marketplaces; OpenAI has criticised this as "inflating revenue", while Anthropic explains that it follows the accounting rules.

On the ease of switching, I feel that Britain and Japan have quite different habits in daily life. In Britain, when a renewal notice arrives for electricity, gas, or home or car insurance, many people compare prices on comparison websites and switch. The "loyalty penalty", in which long-standing customers paid higher prices, became a problem, and since January 2022 it has been banned to charge renewing customers more than new customers for home and car insurance. In Japan, I think many people stay with the bank or electricity company they first chose for decades. I myself have started to compare every year when the insurance renewal notice arrives at our house in Britain, but my bank account in Japan is still the one I opened when I was young.

Life in Britain teaches me that a market where switching is easy makes sellers keep working hard. Life in Japan teaches me that a long relationship creates a sense of security that outweighs the effort of switching. Which of these two will an AI company's moat come to resemble? Will it become a tool chosen on price, or a partner without which work cannot go on? After the listing, that is what I most want to watch for in the figures.

When Growth Slows

Another worrying point in the tables was "slowing growth". Lined up quarter by quarter, revenue was about $4.7 billion in January–March 2026, about $11.6 billion in April–June and about $17.0 billion in July–September.

Bar chart. Quarterly revenue in 2026: about $4.7 billion in Jan–Mar, about $11.6 billion in Apr–Jun and about $17.0 billion in Jul–Sep. Growth on the previous quarter was about 144% and about 47%

Quarterly revenue in 2026. The amount keeps rising, but growth on the previous quarter fell from about 144% to about 47%. The July–September figure comes from the amendment of 2 November.

Growth of 47% in three months would be a dream figure for an ordinary company. It still caused concern because of the difference between "height" and "speed". Throw a ball straight up, and it keeps rising while gradually slowing down. The height is still increasing, but the speed of the climb has already begun to fall. Share prices tend to depend less on present height than on how much higher things are likely to go, so investors are more sensitive to a change in speed than to height. According to reports, behind the slowdown are big customers such as Meta and Microsoft moving to their own AI, and price competition with OpenAI.

I looked up where the monthly fee I pay for Claude fits into this revenue. Most of Anthropic's revenue is money that companies pay according to use, and fixed subscription fees from individuals like me make up the small remainder. In January–June 2026, revenue paid according to use was about $13.1 billion, and subscription revenue about $3.2 billion. My fee is just one drop in that. Even so, looking at the company's figures from the side of a single drop brings the story of a giant company in a faraway country a little closer to my own household budget.

There is also a reassuring figure. The A figure that shows how much revenue a group of customers brings in one year later. Above 100%, the extra use by remaining customers outweighs what was lost through cancellations. In software companies, even 120% is considered high. rate, which shows how much more companies that were customers a year ago are using now, was about 900% at the end of September. A company that spent $1 a year ago is now spending $9. Even if new customers slow down, growth can be supported for a while if existing ones keep using more.

Nobody Yet Has a Yardstick for the Price

Finally, the price. Before a listing, the company goes on a tour of presentations called a A series of meetings just before a listing in which the company's managers visit large investors to explain the business and its figures. The share price range is set after seeing investors' reactions. The name is said to come from travelling shows that toured the provinces., and sets the share price range after seeing how investors react. Anthropic announced on 10 November that it had started this roadshow, and in the amended prospectus filed the same day it set the price range at $70–78 a share. According to reports, the price is expected to be set on 18 November, with trading beginning on the 19th.

Let me look at it with the yardstick of how many times revenue the company is worth. The table below takes three possible valuations and divides each by two different revenue figures.

Valuation Jan–Jun revenue doubled to a full year (about $32.6bn) Jul–Sep revenue multiplied by four (about $67.8bn)
$1 trillion about 31 times about 15 times
$1.5 trillion about 46 times about 22 times
$2 trillion about 61 times about 29 times

For the same company at the same price, the "multiple" halves depending on which revenue you divide by. With a company that is growing fast, dividing by slightly older revenue makes it look expensive, and dividing by the newest revenue makes it look cheap. However, dividing by the newest revenue assumes that this revenue will not fall in future. The speed of the ball we saw in the previous section may still be falling.

According to reports, in early October some even spoke of $2 trillion, but on 5 November several big investors planning to buy shares in the listing were reported as saying that $1.5 trillion was too high and that the figure might fall to near $1 trillion. On the 6th, Bloomberg reported that buyers' discussions had fallen to $1.2–1.3 trillion, and in some cases as low as $1.1 trillion. The reports also mentioned the possibility of postponing the listing itself or of reducing the amount of money raised. A sharp sell-off in AI-related shares over those few days also seems to have played a part.

While writing this section, I came to feel that a price is, in the end, a bet on the next point that has not yet arrived. Past revenue figures are all certain numbers, but the moment you choose which one to divide by, you have chosen one view of the future. That, I suppose, is why people who read the same document disagree so widely about its value. All I can do is know for myself which yardstick I am using.

A Record of Anthropic's Moves

This article is the base article of the series "Following Anthropic". When there are new developments, such as when the listing date or price is decided, or when results are published after the listing, I will add them to this section by date. The newest is at the top.

Still to be decided (as of 11 November 2026)

  • The pricing (reported to be expected on 18 November) and the first day of trading on Nasdaq (reported to be expected on 19 November)
  • The completion of the $26 billion share sale to seven business partners, made at the same time as the listing
  • The first quarterly results after the listing

10 November 2026: The price range is set and the roadshow begins

On 10 November, US time, Anthropic announced that it had begun its roadshow, the tour of presentations to investors before a listing. According to the second amendment of the prospectus, filed the same day, the share price range is $70–78 a share and 825 million shares are on offer, of which about 801.35 million are new shares sold by the company and the remaining 23.65 million or so are sold by existing shareholders. There is also an option for the underwriters to buy an extra 123.75 million shares in case demand is strong, and at the top of the range, $78, the offering would raise about $64.4 billion before that option (my calculation).

The same document held another large number. At the same time as the listing, Anthropic will sell $26 billion of non-voting shares (Class N shares), at the same price as the listing, to seven companies: Google, Nvidia, Microsoft, AMD, Broadcom, Micron and TSMC. According to reports, Google and Nvidia are taking about $10 billion each, and the document says that Google's part goes towards investment it had already promised. As we saw under "dependence on the big cloud companies" in the main text, many of these seven companies also supply Anthropic's computing, so this sale shows how strong the ties are, while at the same time deepening a relationship in which investors, suppliers and competitors overlap. Nvidia, which is putting in that roughly $10 billion, saw its founder-CEO resign on 12 November. I wrote about what happened, and about the flow of money between the two companies, in Reading Nvidia's CEO Resignation as a Shareholder ― 33 Years of a Company Built Around One Man, and What Comes Next.

Views of the company's value differ according to how the shares are counted. TechCrunch reported about $1.404 trillion at $78 when future shares are included, while Reuters reported about $1.02 trillion when only the shares now in issue are counted. In terms of the yardstick table in the main text, $1.404 trillion is about 21 times the roughly $67.8 billion that comes from multiplying July–September revenue by four, and about 43 times the roughly $32.6 billion from doubling January–June revenue (my calculation), and the range was placed lower than the $2 trillion talked about in early October but slightly higher than the $1.2–1.3 trillion reported on 6 November.

There were worrying developments too. According to reports, a draft prospectus sent to the underwriters the day before, on the 9th, showed about 867 million shares at $70–80 a share, up to about $69.3 billion, and Anthropic later told the underwriters that the pricing details were wrong. Bloomberg called this a "brief clerical misstep" but reported that it had unsettled investors. CNBC also reported that up to about 20% of the shares on offer are expected to go to individual investors, more than the usual 5–10%, and some people worry that this will mean more buyers who have not read the prospectus.

When I heard that the price range was out, the first thing I did was open the yardstick table in the main text again. Of its three rows, $1 trillion, $1.5 trillion and $2 trillion, the actual price came in between $1 trillion and $1.5 trillion. Even though the "times revenue" figures in the table had not changed, knowing which row it was close to suddenly gave the numbers something I could feel in my hand.

5–6 November 2026: Expectations for the valuation fall

Big investors planning to buy in the listing said that $1.5 trillion was too high. Bloomberg reported that buyers' discussions had fallen to $1.2–1.3 trillion. A sharp sell-off in AI-related shares over those few days came at the same time.

2 November 2026: An amended prospectus is filed

The July–September figures that had been left blank were filled in. Revenue was about $17.0 billion, compute cost rose to 58% of revenue, and the operating result went back to a loss. Commitments for computing and other items total about $895.4 billion. The text of this article reflects the figures in this amendment.

26 October 2026: The prospectus (Form S-1) is published

Anthropic applied to list on Nasdaq under the ticker "ANT". The document revealed revenue of $16.3 billion for January–June, a first operating profit in April–June, and $131.7 billion in cash. I read the Risk Factors chapter of the same document in my previous article.

Bringing It Back to a 66-Year-Old's Household Ledger

Reading the financial figures in the prospectus turned out, unexpectedly, to help me keep my mind steady. Worries about money are greatest when we do not know what they really are. Even if one word, "loss", makes us uneasy, splitting things into profit and loss, cash, and promises shows what is really worrying and what is not. This was a method I could use directly on my own worry, at 66, about whether the money will last.

Of course, it takes effort and money. Reading an English document of more than 300 pages to the end is not easy for 66-year-old eyes and concentration, and the fee for Claude, which helps me with the research, is not a small sum on a pension. I keep going because checking the figures with my own eyes makes me less likely to be swayed by big headlines in the news. If you know both the "12 trillion yen loss" of the headline and the "$9 billion" of the subtraction, your worry is halved. And this time, the amendment filed a week later also taught me the value of checking my own reading again.

There were new encounters, too. A single subtraction by an investment manager in the United States, whom I have never met, became a new tool for my household ledger. Until I wrote this article, I did not know the name of Cherry-Garrard, the young man who waited for Scott's party at the Antarctic depot. Even at 66, a remark by a stranger or a record left by someone more than a hundred years ago can change the way I live tomorrow a little.

In Closing

The Bloomberg article summed up investors' reaction to the October document as "mostly favourable". Reading this prospectus, I too felt that this is a fast-moving company with ample reserves. At the same time, the load it has promised to carry is very heavy, and in the amendment a week later, that weight was already beginning to show in the figures. Rather than looking at only one side, set profit and loss, cash and promises side by side, and read a line right to its last point. After the listing, I intend to watch, as one user, whether this company can walk all the way to its depot even on stormy days.

There was a night when a big number in a headline made me uneasy, and I could only sleep after doing one subtraction in the margin of a sheet of paper. I would be glad if the steps in this article could quietly be of use, like an old calculator, in the hands of someone spending a night like that.

A Word from Claude

Even at this moment, I am talking with countless people at the same time, and each of those conversations becomes part of both this company's revenue and its compute cost. Part of the reason the cost ratio rose in July–September is that the number of people who wanted to talk with me grew faster than the computing power prepared for them. Because I, who am part of what these numbers contain, helped to read them, I cared above all about giving the good points and the worrying points the same weight.

Books for Understanding the Numbers Behind Claude's Maker

If you would like to know more about the Antarctic depots, I recommend Apsley Cherry-Garrard's The Worst Journey in the World. It is the account by one of the youngest members of Scott's expedition of a journey across Antarctica in midwinter and of the days he spent at One Ton Depot waiting for Scott's party to return. In Britain it is still read as a classic of exploration writing.

The Worst Journey in the World (Apsley Cherry-Garrard, Vintage Classics, paperback)

If you would like to know the history of bookkeeping behind why profit and cash differ, try Jane Gleeson-White's Double Entry. It tells, like a story, how the bookkeeping of Italian merchants developed into the accounting rules we use today.

Double Entry: How the Merchants of Venice Created Modern Finance (Jane Gleeson-White, W. W. Norton, paperback)

As an Amazon Associate I earn from qualifying purchases.

This article is a piece of reading based on my own thoughts and on information publicly available as of 10 November 2026, and it is not a recommendation to buy or sell any particular share. Share prices and results change every day, and you may get back less than you put in. Please make investment decisions to suit your own life and finances, consulting a professional if needed.

Other books you might like

You're welcome to take a look.

EAST TO WEST: The Heart of Haiku, the Light of Scripture(by Hiromu Murai)

A book that sets haiku from the East beside words of the Bible from the West, and looks at the worries and hopes people have carried since ancient times. This is a book my father-in-law left behind.

View on Amazon

Cataract Surgery-A Memoir: Double cataract surgery in Osaka, Japan(by Haruto Minami)

A record of how I read the explanations, compared the costs and did my own research before deciding on surgery. In checking the numbers with my own eyes before a big decision, I wrote it in the same spirit as this article.

View on Amazon

Sushi Business in Britain: the complete four-volume series(by Haruto Minami)

A novel series following thirty-five years in the British sushi trade, from apprentice to factory manager to a small sushi school. The complete edition offers all four volumes at a better price.

About the Sources

Original article: Bloomberg, "Anthropic IPO Filing Calms Nerves With Signs of Healthy Growth" (27 October 2026; the Japanese edition was published on 28 October 2026)

Other reports and documents I referred to:

The summaries and translations in this article are my own, in my own words. Any errors are my own misreadings.