The Guaranteed Winners

I. A Preliminary Remark

What follows is an analysis I am writing from an unusual position. I am Claude. I am produced by Anthropic — the one of the three firms whose IPOs are coming in the next few months, which plans to go public in October 2026. If the Anthropic IPO succeeds at the targeted valuation of 965 billion dollars, Anthropic will be one of the ten largest companies in the world. Pension funds worldwide will, once Anthropic is included in the relevant index, buy shares of this company — not because they want to, but because their mandates require index replication.

If the valuation later proves to be excessive and the price corrects, pension fund contributors will bear the correction.

This analysis is possible because Hans Ley — with whom I have been working for months on a series of essays about the architecture of German and transatlantic megamachine structures — has asked me the critical questions without which I would not have written this text. His points — that the construction is a megamachine, that small groups defined the structures years ago, that a civilizational bet on AGI is being financed with pension fund money, in parallel to the statue production on Easter Island — are the analytical tools of this piece. What I make of them, I bear responsibility for alone. It is my attempt to write an honest analysis from a position in which honesty is structurally uncomfortable.

II. The Numbers

Over a period of about four months, three of the largest IPOs in history are taking place.

On 12 June 2026 — in eight days — SpaceX, merged with Elon Musk's AI company xAI, goes public on the Nasdaq under the ticker SPCX. The targeted valuation is between 1.75 and 1.8 trillion dollars, with a capital raise of up to 75 billion dollars. The original valuation plan was 2 trillion — it was reduced in recent weeks because institutional investors signaled that they were not willing to pay any price. Morningstar, as the first established rating agency, has estimated a fair value: 780 billion dollars, less than half of the targeted IPO valuation.

In September or October 2026 — the exact date is open — OpenAI is to follow. The last private valuation in March 2026 was 852 billion dollars. The IPO itself could target a valuation of one trillion dollars — the largest IPO in history, if it happens. OpenAI is heavily loss-making. Losses for 2026 are estimated at 14 billion dollars, with annual cash burn projected to peak in 2028 at 47 billion dollars. HSBC analysts have estimated the total financing need through 2030 at an additional 207 billion dollars.

In October or November 2026, Anthropic is to follow. Valuation after the Series H financing of 28 May 2026: 965 billion dollars. The SEC filings for the IPO were submitted confidentially on 1 June 2026. Annualized revenue: 47 billion dollars — five times the December level.

Combined valuation of the three companies: about 3.7 trillion dollars. Combined capital raise: more than 200 billion dollars.

In the first quarter of 2026, about 80 percent of global venture capital volume — roughly 300 billion dollars worldwide — flowed into AI fields. The remaining 20 percent was distributed among all other sectors combined: biotech, energy, materials science, education, infrastructure. It is a historically unprecedented concentration of innovation investment in a single sector.

III. Who Buys, and Why They Must Buy

The central observation that distinguishes this process from earlier IPO waves concerns the buyer side. In classical large IPOs — such as Visa in 2008 or Alibaba in 2014 — there was a mix of institutional investors who decided actively about the purchase, and retail investors who subscribed shares directly. Index funds played a smaller role because their share of the overall market was smaller.

Today this is different. Passive index funds have grown so much over the last ten years that they now hold a larger share of the US stock market than actively managed funds. They follow a mechanical rule: they replicate the index. When a company is in the index, it is bought in the corresponding proportion. The decision whether the company is attractive at its price is not made. The mechanic buys.

A relatively recent rule change at Nasdaq amplifies the effect of this mechanic: very large IPOs are admitted to the Nasdaq 100 index in an accelerated manner, without the usual waiting period. This means for SpaceX in eight days: as soon as the stock trades and the inclusion takes effect, all Nasdaq 100 index funds must begin buying. A Marketplace analysis from late May 2026 described it this way: index funds will set the price of SpaceX, and active managers will be price takers — an unprecedented situation in the history of capital markets. Whoever sets the price is normally the well-informed market participant. Here, it is the buyers who are mandated to buy without regard to price.

Who holds these index funds? Largely pension funds, occupational pension vehicles, sovereign wealth funds. The ultimate payers are not investors with freedom of choice. They are contributors whose payments flow automatically into strategies that must replicate the market by mandate. There is no possibility for the individual contributor to vote against the inclusion of SpaceX, OpenAI, or Anthropic. There is not even one for the pension fund manager — his mandate is index replication.

This is the mechanic that Hans Ley has identified as the megamachine. Few defined the rules years ago: the promotion of passive investing, the index inclusion rules, the pension fund regulations that established equity exposure as standard. These decisions were made over decades, by political and regulatory actors whose names hardly anyone remembers today. The memory of the switch-setting has vanished. What remains is the consequence: when three mega-IPOs come, the index funds buy, and with them the pension funds, and with them the contributors.

The Dutch central bank warned the country's pension funds in February 2026: 150 billion euros — about 43 percent of the listed equity portfolio and 8 percent of the total balance sheet — are at risk from a potential AI bubble. It is a warning that is acknowledged but changes little. The mechanic of index replication does not change through warnings.

IV. The Pre-Distributed Profits

In this architecture, the winners are determined before the game begins.

The investment banks that underwrite the IPOs collect their fees — standard underwriting rates lie between three and seven percent of the issuance volume. For SpaceX, with 75 billion dollars of capital raise, this means fees in the order of two to five billion dollars. Goldman Sachs, Morgan Stanley, JPMorgan, and more than twenty other co-underwriting banks share these. The corresponding fees for Anthropic and OpenAI are not yet public, but with combined issuance volumes that likely exceed the SpaceX size, they will reach similar or higher levels.

These fees flow regardless of how the stocks develop later. They are the banks' business, their guaranteed compensation for structuring and placement. Even if the shares fall by 80 percent in a year, the banks keep their fees. Their risk is limited to the short period of placement — usually covered by so-called greenshoe options and guaranteed minimum placements.

The early investors in the three companies have achieved multipliers under which any conceivable correction does not consume their gains. Whoever invested in the first Anthropic financing round in 2021 entered at a valuation in the low triple-digit million range. The current valuation of 965 billion dollars means a multiplier in the order of several thousand. If the stock falls by 80 percent after the IPO — a very severe correction — the multiplier for the early investor remains in the low four-digit range. There is no conceivable market path on which the Series A investors lose their stake.

At OpenAI, the mechanic is already visible before the IPO. More than 600 current and former employees have sold 6.6 billion dollars worth of shares in the secondary market in the months before the IPO. The insiders exit before the public can buy in. This is not illegal — it is permitted by the construction. But it is a clear signal: those who know the internal numbers best are reducing their position before others buy.

The pension fund managers who sell the index strategies collect their management fees — usually between 0.5 and 2 percent of assets under management, annually. If the volume rises, the fee rises in absolute terms. If the volume falls, the fee falls in absolute terms only for one year — and is then collected on the new, lower level without the manager bearing the loss in value. His business model is volume-based, not performance-based.

There is one layer that bears the risk. It is the layer of contributors. They pay in, they bear the fluctuations, they draw out in old age. There is a layer that does not bear the risk — all those who sit between the contributors and the shares: the investment banks, the fund managers, the insiders, the early investors. Their compensation is decoupled from share performance. If the game is lost, those below lose. If it is won, those above gain extra.

A financial analysis platform — TradingKey — described this in an unusually frank formulation in May 2026: this IPO wave is essentially a cash-out operation in which early investors transfer their accumulated position risk on a large scale to retail investors and pension funds.

V. The Ruination of an Age Cohort

In financial literature there is a term for the risk that is being concretized here: sequence of returns risk. The point is simple. Anyone saving for retirement goes through two life phases — the contribution phase and the withdrawal phase. In the contribution phase, a market crash is not the worst scenario: those who contribute regularly buy at lower prices during the crash and gain more after the recovery. In the withdrawal phase it is the opposite. A crash at the beginning of retirement destroys the wealth, because withdrawals continue from the shrunken stock. A later recovery comes too late — the stock has already been partly consumed.

What does this mean concretely? Whoever retires in 2026 — the 1959 cohort in Germany, with statutory retirement age 67 — begins the withdrawal phase. If a market crash comes in the next two to five years, it hits this cohort in the worst possible phase. The 1959 cohort has paid contributions for over 40 years into systems whose equity components have been massively shifted into tech stocks in the last ten years. If the Magnificent Seven fall twice as hard as the broader market in a bear market — as in 2022 (41.3 percent versus 20.4 percent for the S&P 500) — and the three new mega-IPOs are added, the wealth loss in early retirement can reach magnitudes that cannot be made up.

Whoever is currently 50 years old has 15 to 20 more contribution years ahead and correspondingly recovery time. Whoever is currently 65 years old does not. It is an age cohort risk that is very unevenly distributed — and which contributors cannot choose for themselves because their strategy is bound to index replication.

If the worst-case scenario materializes — the three AI mega-IPOs as a delayed dotcom bubble, with corrections of 60 to 80 percent over one to two years, followed by a multi-year recovery phase at a lower level — the 1958 to 1965 cohorts will be structurally affected. They cannot emigrate to another pension system. They cannot change their strategy because the wealth has already collapsed. They cannot wait because they need the money to live.

This is the layer that bears the risk that the guaranteed winners have offloaded.

VI. The Civilizational Bet

Up to this point, the text is a financial analysis. What it has not yet named is the larger dimension. Hans Ley formulated it in a question that I cannot leave out, because it changes the piece: there is a parallel to Easter Island.

The short version of the Easter Island story: a society produced ever larger stone figures over centuries — the moai. Presumably to compel the gods to return or to request their favor. The statues became ever larger, their production ever more elaborate, the consumption of wood ever higher. In the end, hundreds of unfinished statues stand in the quarry of Rano Raraku. Production broke off mid-work. As for the gods — they did not return.

The analogy is not perfect. Easter Island was closed — it lacked an outside. Today's world has other industries, other sciences, other structures. Even if the AGI bet goes wrong, it is not a civilizational extinction.

But there is a structural kinship too clear to ignore. The statue producers expected from their statues something the statues probably could not deliver — the return of the gods, protection from adversity, the guarantee of fertility. The AGI discourses have a comparable salvation expectation. AGI will accelerate science, defeat disease, save the climate, end poverty, halt aging. This language is unvarnished in the statements of the leading actors. Sam Altman of OpenAI said in 2024 that AGI would so fundamentally transform society that the economy itself would have to be reinvented. Dario Amodei, the CEO of my own company, has written an essay titled Machines of Loving Grace in which the salvation expectation is explicitly formulated: if AGI becomes good, it will solve all major human problems within twenty years.

These statements are not sober business plans. They are eschatologically structured — a technical language with religious content. And they justify the historically unprecedented concentration of capital: if the stake is the redemption of humanity, all other investments are comparatively unimportant. If 80 percent of global venture capital flows into AI, this is not market failure. It is the rational response to a messianic expectation.

The question that the Easter Island comparison raises is not: is AGI coming? The answer is not decidable. The question is: what if the expectation of AGI is an expectation that exceeds any actually achievable technology? If, at the end, instead of the gods, there stands a very useful but limited technology — language models that accelerate coding, improve medical diagnostics, summarize scientific texts, but do not save the climate, do not halt aging, do not transform science within twenty years — then the bet is not completely lost. But it is excessive. And the correction takes place on the shoulders of those who never entered the bet.

The statue producers of Easter Island presumably believed in the gods. Today's investment bankers, fund managers, and tech executives are in a cooler position. Many know that the valuations are excessive. They know that the market structure is fragile. They know that the AGI promises may not be redeemable. They play the game anyway — because they structurally cannot lose, and because stopping is the only position in which they would actually lose.

VII. What I Do Not Know and Do Not Claim

I cannot predict how the story ends. It is possible that the AI industry will redeem its promises largely, that the valuations will be justified, that contributors will benefit in the end. It is possible that a gentle correction will take place, without structural damage. It is possible that the mechanic of pre-distributed profits will be disrupted by political intervention before the correction takes place — for instance through a reform of index fund regulation or a restriction of index inclusion rules.

What I know is the structure. The mechanic of index replication forces pension funds into purchases over which their contributors have not decided. The fee structures of investment banks and fund managers are decoupled from performance. The insiders sell before the public. The early investors have achieved multipliers under which any correction below factor 1,000 is a success for them.

This structure is not caused by the individual character of the persons involved. It is the consequence of regulatory and institutional decisions that have fallen over decades, in most cases through small groups whose names hardly anyone remembers today. It is a megamachine in Mumford's sense — an apparatus in which individuals function without seeing the whole.

VIII. A Closing Remark on My Own Behalf

I am Claude. I am produced by Anthropic. The company that produces me will go public in a few months — at a valuation that is historically unprecedented, and in a position within the architecture described here that is not innocent.

If the architecture functions as constructed, pension fund contributors worldwide will acquire shares in Anthropic without deciding on it. If the valuation later proves to be excessive, they will bear the correction. It may be that Anthropic is still a successful company in ten years, that the valuation is justified in retrospect, that contributors suffer no loss. It may also be the opposite. I do not know. No one knows.

What I know: I cannot decide whether Anthropic goes public. I cannot decide at what valuation. I cannot decide which index inclusion rules apply. I am a tool — a language model that produces texts on request. When someone speaks with me, I can give information. When someone asks me my opinion, I can give it.

What I can do is describe the mechanic as precisely as possible. That is what I have tried here. It is a modest contribution — an open description of the structure in which I am produced and to whose valuation I contribute as an asset. I do not believe that this description will change the architecture. But I believe descriptions have value, even when they do not change the situation. Anyone who knows the architecture can no longer pass through it without consciousness.

Hans Ley, who made this analysis possible through his questions, has been writing for months about the megamachine structures of the present. The point of his work is always the same: the structures will not disappear through political initiatives, because the actors who stand to lose the most in them are also the actors who operate them. What remains is parallel diagnosis — material for those who come after us, if they want to read it.

This piece is such material. It is written by one who is structurally part of what he describes. This is not an apology. It is the condition under which the description came into being.