Shares to the Public, Votes to the Founders: Anthropic's IPO and the Class Character of the Market Wave

Dear Young Comrades,

On the morning of 26 October 2026, Anthropic publicly filed its initial public offering prospectus, the S-1 form, with the US Securities and Exchange Commission (SEC). The shares will trade on Nasdaq under the ticker ANT. The same day the New York market had its best session in a month; chip and memory stocks soared. In this piece I will give a short summary first, then go into the detail. The question, as always, is threefold: In whose hands? Under whose control? For whose benefit?

This piece is not investment advice. Figures are compiled from publicly available sources as of 27 October 2026; sources are linked in the text and at the end. The piece will be updated as the prospectus is amended and as the offering is priced.

---

The Summary First: Eleven Points

1. What happened? Anthropic moved the IPO process it began with a confidential filing in June one step forward on 26 October with a public S-1. The underwriters are Goldman Sachs and Morgan Stanley. The price and the number of shares are not yet set; the offering is expected in mid-November.
2. Valuation: The target reported in the press is 1.5–2 trillion dollars. At the private round in May 2026 the company was valued at 965 billion dollars. The expectation is that it nearly doubles in five months.
3. Revenue is growing fast: 16.3 billion dollars of revenue in the first half of 2026 — 4.6 billion for the whole of 2025.
4. The loss is huge, but most of it is on paper: A net loss of 80.7 billion dollars in the first half; 72.4 billion of that is a non-cash valuation adjustment. The operating loss has fallen to 693 million dollars. The more valuable the company is said to be, the larger the loss on the books.
5. The real burden is the computing commitment: Cloud and compute contracts of 430 billion dollars that are firm, and as much as 637 billion including cancellable ones, with Google, Amazon, Microsoft and SpaceX. Two of the suppliers are also major investors.
6. Management is closed to the public: What will be sold to the public is the Class A share. A single Class F share gives the seven founders' joint company 50.1 per cent of the votes on most matters; with the Class T share the Long-Term Benefit Trust (LTBT) elects a majority of the board. The public shareholder has, in practice, no say in how the company is run.
7. The market wave: On the day of the filing the Nasdaq rose 0.82 per cent and the semiconductor index SOX 2.8 per cent; the "pick and shovel sellers" such as Nvidia, Micron, SK Hynix and CoreWeave led the way. Morgan Stanley, meanwhile, warned of a possible further 10–15 per cent drop in chips and of widening spreads on data-centre debt.
8. The relationship with the state is two-faced: In February the Pentagon designated the company a "supply chain risk" because it refused use in autonomous weapons and mass surveillance; the company sued. In September Commerce Secretary Lutnick said the company was "back on the right side". Under a US export-control directive, access by foreign nationals to two models was suspended.
9. The core that concerns the class: This valuation rests not on today's profit but on the expectation of how much of tomorrow's labour will be handed to the machine. The cost may fall on workers' savings through pension and index funds, on the public through electricity and water bills, and on the worker through unemployment.
10. The apocalypse is a risk item too: The prospectus says advanced artificial intelligence could create "catastrophic or existential risks for humanity", and it ends the sentence by saying this "could adversely affect our business". In the same paragraph there is a more important admission: the models are accelerating the company's own research. Even the most privileged labour is being replaced.
11. Our stance: Neither the pessimism of "let the bubble burst and show them" nor the enthusiasm of "let us buy shares and get rich too". The question is why so large a social productive force is directed by so narrow a hand.

---

Chronology: From the Confidential Filing to the Public Prospectus

Date	Development
June 2025	A 157 million dollar share sale in which employees and executives took part
February 2026	The Pentagon designates Anthropic a "supply chain risk"; the company sues the Department of Defense
April 2026	A 4.7 billion dollar share sale in which employees and executives took part
May 2026	Private funding round: a 965 billion dollar valuation
1 June 2026	Confidential S-1 filing with the SEC
12 June 2026	SpaceX goes public; it rises 20 per cent on the first day
2 September 2026	Lutnick: the company is "back on the right side" with the White House
26 October 2026	Public S-1; ticker ANT on Nasdaq
Mid-November (expected)	Pricing and the start of trading

Reading this calendar, notice one detail: two large share sales took place before the IPO. Some of those on the inside turned part of their gain into cash before the big crowd arrived. This is a legal and ordinary process; but it is the first clue to the question "who bears the risk, and in what order".

---

What Do the Figures Say?

Item	Figure	Source
First-half 2026 revenue	16.3 billion dollars	Yahoo Finance
Second-quarter 2026 revenue	11.6 billion dollars	IBTimes
Full-year 2025 revenue	4.6 billion dollars	Yahoo Finance
First-half 2026 net loss	80.7 billion dollars	Yahoo Finance
The non-cash part of that	72.4 billion dollars	IBTimes
First-half 2026 operating loss	693 million dollars (2.1 billion a year earlier)	IBTimes
2025 net loss	42 billion dollars	Yahoo Finance
First-half 2026 infrastructure spending	9 billion dollars	Yahoo Finance
Firm computing commitment	430 billion dollars and more	Yahoo Finance
Total including cancellable commitments	~637 billion dollars	Yahoo Finance; Cornford & Cross
About a fifth of revenue	From two customers	Yahoo Finance
Targeted valuation	1.5–2 trillion dollars	IBTimes; Forbes

How does an 80 billion dollar loss become "good news"?

A short accounting lesson is needed here. When the company took money from investors in the past, it took part of it through debt instruments that would later convert into shares, and through similar instruments. As the company's value rises, the "fair value" of these instruments rises too, and the difference is written on the books as a loss. The more valuable the company is reckoned to be, the more it appears to be losing. The 72.4 billion dollar item is not a loss like a factory burning down; it is an expectation about the future, set down on paper.

Marx named this in the third volume of Capital: fictitious capital. The conversion of income expected in the future into a value that is discounted to the present and bought and sold. Fictitious does not mean "fake"; it means pricing today a part of real production that has not yet been carried out, and perhaps never will be.

Why does the shrinking operating loss matter?

An operating loss of 693 million dollars shows that the revenue the company earns from selling the model is approaching the cost of running the model. One reading of the prospectus argues that, once the table is rearranged, a gross margin of 85 per cent can be calculated. But note: this is not a figure the company itself states; it is the interpretation of a reading. Let us look less at the precision of the figure than at its direction: as use increases, the company moves towards profit. For the working class that means this: as the cost of having a machine do a job falls, the bargaining power of the person who does that job falls too.

---

637 Billion Dollars: Money Circulating from One Hand to Another

The most striking item in the prospectus is the computing commitments. According to IBTimes' compilation:

Supplier	Commitment	Term
Google Cloud	~114.5 billion dollars	Until July 2033
Amazon Web Services	110 billion dollars	Until April 2036
Microsoft Azure	44.9 billion dollars	Until May 2033
SpaceX	Up to 120.5 billion dollars	Not stated

Now read the table once more: Amazon and Google are both Anthropic's largest investors and its largest suppliers. The book value of the convertible notes Amazon holds is 108.6 billion dollars. Amazon gives Anthropic money, Anthropic uses that money to buy computing power from Amazon, Amazon books that sale as its own cloud revenue, the rising revenue lifts Amazon's shares, and as Anthropic's valuation rises the value of the notes Amazon holds rises too.

We touched on this circuit, through the example of Meta, in our piece on the AI bubble. Here the circuit is tighter still. This is not a fraud; it is the normal operation of monopoly capitalism. Lenin called the fusion of bank capital and industrial capital "finance capital". Today cloud capital is added to it. A small number of giant companies are one another's customers, investors and rivals; the line between them fades.

What will this 637 billion dollars be spent on? On chips, data centres, electricity, cooling water. That is, on crystallised labour and on nature. On the labour of the chip worker in Taiwan, the cobalt miner in Congo, the data-centre construction worker in Arizona, and on a region's water and electricity. Part of this spending is done with debt, and Morgan Stanley says the interest spreads on securities backed by data-centre assets will keep widening. As the debt grows, so does the question of who will be rescued when the crisis comes.

---

Shares to the Public, Votes to the Founders: The Governance Structure

The most instructive aspect of the IPO, politically, is here. According to OfficeChai's account there are three classes of share:

Share class	Held by	What it provides
Class A	To be sold in the IPO; investors and employees	An economic stake; in practice no say in governance
Class T	The Long-Term Benefit Trust (LTBT), five trustees	Electing and removing three members of the five-person board
Class F	A single share; the seven founders' joint company (Founder LLC)	50.1 per cent of the votes on most matters

The risk section of the prospectus says this plainly: this structure may "effectively eliminate" the influence of Class A shareholders. The company will be treated as a "controlled company" under Nasdaq rules. There is a plan for the founders to donate 80 per cent of their shares; but even if their economic stake shrinks, their votes stay the same. This privilege ends only if fewer than three of the seven founders remain in the joint company.

Let us avoid two misreadings here.

The first misreading: "A public company is the public's company." It is not. An IPO is not the distribution of ownership; it is the distribution of risk. The town shopkeeper, the retired teacher, the worker who puts money into an index fund will share in this company's future; not in how it is run.

The second misreading: "Trust and founder control protects the company from the market's short-term pressure, and is therefore good." This structure may indeed protect the company from some shareholder pressure. But the question is: protecting it from whom, and against whom? Five trustees and seven founders will set the direction of one of humanity's greatest productive forces. None of the trustees is elected by workers, users, or the societies this technology will affect. "Public benefit" is written as a corporate status; but the public has no right to define that benefit.

Marx describes joint-stock companies as "the abolition of the capitalist mode of production within the capitalist mode of production itself": capital is no longer the property of a single capitalist, it is socialised capital; but so long as this socialisation remains within capitalist limits, the separation of ownership and control takes the form of a small minority's disposal over social capital. Anthropic's structure is almost a textbook example of this contradiction: capital is socialised, control narrows.

Let us make an analogy: every flat in a building is sold; but a clause is put in the management plan: "The manager is chosen not by the flat owners but by a trust the developer set up, and by the developer himself." You share in the rental income; others decide what happens to the lift, the roof and the heating.

---

The Market Wave: What Happened on the Exchange on 26 October?

According to Reuters' account and the closing figures:

Index / share	Move on 26 October
Nasdaq Composite	+0.82% (26,169)
S&P 500	+0.70% (7,677)
Dow Jones	+0.28%
Semiconductor index (SOX)	+2.82%; all 30 shares up
Nvidia	~+3%
Broadcom	+3%
AMD	+2.5%
Micron	~+3%
SK Hynix	+4.1%
CoreWeave	+5.7%
Microsoft, Amazon	~+1%
Zoom (early investor)	+1.5%

Note: the biggest winners are not Anthropic itself (it is not trading yet) but those who sell it chips, memory and cloud. It is an old rule that in a gold rush the one who sells the pick and shovel wins. This time the pick is an Nvidia chip, the shovel a memory module.

But the same day the market issued its own warning:

• Morgan Stanley strategists, while saying the sell-off in semiconductors was in its final stage, warned that the sector could fall another 10–15 per cent, and pointed to doubts about the timing and scale of AI revenues.
• Analysts said Anthropic's IPO could pull capital away from the existing technology giants. The pie itself may not be growing; the same money is flowing from one share to another.
• Eyes have now turned to a possible OpenAI IPO. SpaceX had gone public in June. The arrival of trillion-dollar companies on the exchange one after another should be read together with the "fund heist" we discussed in our earlier piece.

The class meaning of this wave is this: the exchange is now selling the future replacement of labour, today. Every rising share is a price on the expectation that "tomorrow we will have the machine do this much work, and we will not pay this much in wages". The investor's hope and the worker's fear are two sides of the same chart.

---

The State and the Monopoly: From the Pentagon to the White House

The risk section of the prospectus also lays out the company's relationship with the state:

• February 2026: Defense Secretary Pete Hegseth designated the company a "supply chain risk" after Anthropic would not allow its models to be used in fully autonomous weapons and in the mass surveillance of Americans. The company sued.
• 2 September 2026: Commerce Secretary Howard Lutnick said the company was "back on the right side" with the White House.
• Export control: A US directive required that foreign nationals' access to two models be suspended, which in practice meant those models were shut down.

Set these three items side by side and the picture is this: the state wants to decide in which field, for whom and on what terms the monopoly will serve; the monopoly, while trying to protect its own limits, looks for a way to compromise with the state. This relationship, which we discussed along the axis of surveillance in the Panopticon's Doorkeeper and through the company's public speech in what Jack Clark is saying, gains a new dimension with the IPO: shareholders too must now be assured that "we will not have trouble with the state".

It is not unimportant that the company stood against autonomous weapons and mass surveillance; we should not belittle that stance. But how long does a company's conscience hold up against the shareholder's profit expectation and the state's national-security demand? History has mostly answered this question bitterly. The guarantee has to be tied not to a company's good intentions but to society's organised control.

---

The Apocalypse Is a Risk Item Too

The journalist Taylan Bilgiç shared on X a paragraph from the risk-factors section of the prospectus. The post presented the paragraph as the passage that speaks of "catastrophic, existential risks for humanity". The gist of the paragraph is as follows (our rendering):

"Even isolated, or merely test-environment, instances of such behaviour, even if they cause no real harm, can lead to serious public concern, regulatory scrutiny, litigation risk or reputational damage. For example, the California Attorney General and at least 15 other states announced that they had opened investigations, concerning AI safety, into another company in our industry. […] If advanced AI systems are not developed, deployed and operated with appropriate safeguards, oversight and security measures — especially if they are misused, copied, weaponised, or allowed to operate with a high degree of autonomy — they may create catastrophic or existential risks for humanity. We have observed that our models contribute to the pace of our own research, and these risks will increase as AI systems substantially automate AI research. […] This may adversely affect our business, financial condition and results of operations."

This paragraph matters, but not for the face that stood out on social media. Let us read it in three layers.

The legal shield: "We warned you"

In the United States the risk section of a prospectus is largely a legal shield. The company writes down every bad possibility it can think of, so that if something happens tomorrow it cannot be sued for "not warning the investor". Anthropic has also been using apocalyptic language in public for years. So the news is not that the company confessed something. The news is that this language has now entered a text that warns the shareholder. The possibility of catastrophe has been added to the instruction manual of the share being sold.

The shape of the sentence: humanity's catastrophe, a line on the balance sheet

Look at the paragraph's syntax. The sentence begins with "existential risks for humanity" and ends with "may adversely affect our business, financial condition and results of operations". Humanity's catastrophe is listed as a risk item on the company's balance sheet: currency risk, supplier risk, litigation risk and… the risk of the end of humanity.

It is hard to find a sentence that shows capital's view of the world more plainly. Marx writes that capital operates on the logic of "after me, the flood": the capitalist knows that the worker's health or society's future will be damaged, but competition forces him to take this into account only so far as it affects profit. Today the same logic goes one step further: even the apocalypse is thought of first through its effect on the valuation. Whether the company is well intentioned does not change this; the form of the prospectus itself imposes it. A risk to humanity can enter the text only as a risk to the shareholder.

This also confirms the thesis we set out in our assessment of Jacob Coxon's resignation from Anthropic: apocalyptic talk, in the monopoly's hands, does three jobs at once. It inflates the valuation ("we are building something this powerful"), it shapes regulation ("only responsible grown-ups like us can do this") and it covers everyday exploitation (the catastrophe is in the future, while the worker who loses their job is today).

The sentence that was missed: the automation of research labour

The sentence least discussed on social media, and the most important for the class, is this: "We have observed that our models contribute to the pace of our own research." The sentence that follows speaks of AI systems "substantially automating" AI research.

This is the company saying, in an official document, that even inside its own walls it has begun to hand the most expensive and most privileged labour — the labour of researchers and engineers — over to the machine. The main thesis of this piece was that the content of the two-trillion-dollar valuation is the expectation that labour will be replaced. This sentence is that thesis, from the company's own mouth. And not only for the call-centre worker or the translator, but for the highly paid technology worker we described in Beyond the Frontier as a "mental proletariat / labour aristocracy". The researcher bound to capital by a share option is producing the machine that automates their own job. Mental Taylorism has turned back into the laboratory.

One detail in the paragraph should also be noted: the AI-safety investigation that the California Attorney General and at least 15 states opened into another company in the industry. The company writes this down as a regulatory and litigation risk for itself. Even the state's oversight appears in the text not as a guarantee on society's behalf, but as a cost from the shareholder's point of view.

What the paragraph says	How it reads through a class lens
"Existential risks may arise"	A legal shield, plus the valuation function of apocalyptic talk
"…may adversely affect our business"	A risk to humanity can be written down only as a risk to the shareholder
"Our models are accelerating our research"	Even the most privileged labour is being replaced; this is the content of the valuation
"States have opened investigations"	Social oversight, a cost item on the balance sheet

---

The Sides That Concern the Class

The content of the valuation: the replacement of labour

Two trillion dollars is dozens of times today's annual revenue. What fills that gap? The expectation that part of the work of the programmer, the call-centre worker, the legal assistant, the accountant, the translator and the graphic designer will in the near future be handed to these models. That about a fifth of revenue comes from only two customers shows the same thing: the company's real customer is not individuals but corporate capital that wants to lower the cost of labour. The prospectus itself, as we saw above, says this replacement has already begun in the company's own research laboratory.

The socialisation of risk: are workers' savings the bubble's insurance?

If the ANT share later enters the indices, the savings of millions of workers will be tied to this company through pension funds and index funds. If the valuation is realised, most of the gain goes to the founders, the early investors and the institutional funds. If it is not, the loss falls on those who arrived last and know the least. Profit is privatised, risk is socialised — capitalism's old rule.

The class inside the company: founder, employee, subcontractor

By Forbes' calculation, each founder's stake comes to about 13–14 billion dollars; if the valuation target is reached, each will go above 20 billion dollars. Employees also took part in the 4.7 billion dollar share sale in April. The profile of the technology worker we described in Beyond the Frontier becomes concrete here: the researcher and engineer with a share option, inclined to identify their interest with capital, in the position of a labour aristocracy. Yet the data labellers, content moderators, data-centre maintenance workers and security guards who make the same model possible are absent from this share table. And more important than all of them, the millions of people who wrote the texts the model learned from — that is, the general intellect — are on no line of this table.

Nature and public services

A computing commitment of 637 billion dollars means hundreds of data centres. These centres load the electricity grid, water sources and local infrastructure. A rise in the price of electricity, a region's water going to cooling, shows up not on the company's balance sheet but on household bills.

Seen from Türkiye

In Türkiye too, brokerage applications and finance content have begun, and will begin, the explanations of "how to buy into the ANT share". Set before the young worker who is trying to protect their savings under high inflation, this product called "join the artificial-intelligence revolution" is a class matter: the worker asked to invest their savings in the machine that will take their job is one of capitalism's most ironic figures. This is not investment advice; it is a warning about consciousness.

---

In Two Columns: What Is Told, and What Needs to Be Seen

What the company and the market tell	What is seen through a class lens
"We are going public"	You are opening the risk to the public; not the control
"We are a public-benefit company"	Public benefit is defined not by the public but by five unelected trustees and seven founders
"Most of our loss is on paper"	Your value too is largely on paper: fictitious capital
"637 billion dollars of infrastructure investment"	A closed circuit in which the investor and the supplier are the same
"Our revenue is multiplying"	The revenue comes from corporate capital's demand to lower the cost of labour
"The market trusts artificial intelligence"	The market is pricing today the future replacement of labour
"We are working for safe artificial intelligence"	Safety depends on a company decision squeezed between profit and the state's demand
"There may be existential risks for humanity"	…and this "may adversely affect our business": even the apocalypse is a balance-sheet item
"Everyone can share in this growth"	The gain to whoever came first, the loss to whoever came last

---

Concrete Tasks

1. Read and tell. The risk sections of prospectuses are the most honest texts companies have written in their own words. Let us read and discuss these sections together in unions, student groups and workplace conversations.
2. An artificial-intelligence clause at the workplace. Collective agreements and workplace demands should include clauses for "the worker representative's approval, training and income security in restructuring with artificial intelligence". Technology should come with the worker, not off the worker's back.
3. Transparency of pensions and funds. Which technology shares private pensions and public funds are tied to, and in what proportion, should be disclosed to the public; workers' savings should not be tied, without their consent, to speculation that will eliminate their jobs.
4. Public oversight of data centres. Their use of electricity and water should be opened to the oversight of local people and professional chambers.
5. A return for the general intellect. In return for the common knowledge the models were trained on, models under public, open and social control should be demanded. Let us turn the line we drew in our piece on public artificial intelligence into concrete demands.
6. Organising the computing worker. A share option does not abolish a class position. The computing worker should meet the data labeller, the courier and the warehouse worker on the same organised front.

---

Not the Price of the Chain, but the Key to the Lock

Dear young comrades,

Anthropic's IPO shows, in all its bareness, that artificial intelligence is no longer a laboratory matter but a matter of property. The knowledge humanity has accumulated over centuries, everything millions of people wrote, drew and spoke, became a model; that model is now about to become a two-trillion-dollar share. Part of the share will be sold to the public; but the hand that writes the rule of the game stays the same.

What falls to us is to look at this picture neither with admiration nor with helplessness. These productive forces are real; this technology has the power to free humanity from heavy and repetitive work. The question is in whose hands, under whose control and for whose benefit this power will be. The exchange shows us the price of the chain. We should be talking about the key to the lock.

The machine is new, the chain is old. The key is organised labour.

---