The Observer's Position: How Anthropic Is Preparing for a Trillion-Dollar IPO

In June 2026, Anthropic filed a registration statement with the U.S. Securities and Exchange Commission ahead of its public listing. The document is confidential: no public version exists yet, and this article was written before one appeared.

Chairs in the park

The company now at the center of what many are calling the largest IPO in history was assembled at the height of the pandemic, when in-person meetings were out of the question. Once the team had grown to around twenty people, they began meeting once a week for lunch at Precita Park in San Francisco, with everyone bringing their own chairs.

OpenAI alumni siblings Dario and Daniela Amodei founded the project, becoming respectively CEO and President. Tom Brown, Jack Clark, Jared Kaplan, Sam McCandlish, and Christopher Olah joined them. As Clark revealed to journalist Alex Kantrowitz during the summer of 2025, the team picked “Anthropic” from a spreadsheet that also included “Aligned AI,” “Generative,” “Sponge,” “Swan,” “Sloth,” and “Sparrow Systems.” They made the decision with a single note: “We like the name it is good.”

Five years

The first major check arrived in 2021 from Sam Bankman-Fried, founder of the crypto exchange FTX. He paid roughly half a billion dollars for a 13.56% stake in the company but received no voting shares or board seat; Dario Amodei would later say that the new investor had made the Anthropic team uneasy from the very start. (FTX collapsed a year and a half later, and a jury convicted Bankman-Fried of fraud.)

Anthropic introduced its first language model, Claude, in March 2023. Amazon had taken a stake in the company by autumn, with Google following shortly after.

From there, the numbers kept climbing, barely pausing to settle. The annualized run rate (the revenue over a month) stood at around $1 billion at the start of 2025 and roughly $9 billion by year’s end. By May 2026, it had surpassed $47 billion, a fivefold increase in six months.

The Series H round in May raised $65 billion and valued Anthropic at $965 billion, pushing it past OpenAI (which had been worth $852 billion when it closed its own round in March).

On June 1, 2026, Anthropic confidentially filed for a public listing. The company had gone from chairs in the park to nearly a trillion dollars in five years.

Who owns

Anthropic PBC is a private company. It has no obligation to publish its shareholder registry, so verified information is scarce, and what little exists is measured in different units.

Google. The corporation (which also develops Gemini, Claude’s direct competitor) owns 14% of Anthropic with a contractual ceiling of 15%: it’s simply not allowed to buy more. Neither company disclosed this figure, which surfaced during an antitrust case against Google and leaked to The New York Times. It reflects a 2025 snapshot. (The massive Series G and Series H rounds diluted everyone’s stakes.) Google has no voting rights or board position.

Amazon. Here the situation is reversed: the data is current, but percentages are absent. In its summer 2026 filings, Amazon valued its share of Anthropic at $190.4 billion (convertible notes and non-voting shares). Only $13 billion of that is actual cash invested. An additional $50.5 billion in gains from a single quarter is pure paper appreciation — a markup on an asset that got more expensive. Amazon doesn’t disclose its actual ownership percentage, and it’s working on its own AI while simultaneously selling Claude through its cloud servers.

The founders. Forbes estimates each co-founder’s stake at just over 1.8%. Combined, they hold roughly 12.5%: less than Google alone.

Only fragments of knowledge exist about the remaining investors. Microsoft, NVIDIA, sovereign wealth funds, and heavyweights like Sequoia Capital all participated in various rounds. Only data aggregators publish exact figures, and those figures differ everywhere you look. All sources agree on one thing: no one holds a controlling stake.

Google and Amazon are in a different category altogether, functioning as investors, competitors, distributors, and landlords all at the same time. (Non-voting shares were not invented for the deals with these companies, however. That structure has been in place since the very first check, back in 2021.) Amazon’s $50.5 billion in paper gains represents an appreciation of its stake in a company that uses investor money to rent computing infrastructure from Amazon itself. That gives Google’s lack of voting rights an added layer of meaning. To understand the implications, you need to understand who actually makes the decisions.

Who decides

Ownership at Anthropic does not mean decision-making authority. The company’s founding documents separated the two from the very beginning.

Anthropic is incorporated as a public benefit corporation. This means its board of directors is legally entitled to place the company’s mission on equal footing with investor interests. Above the board sits the Long-Term Benefit Trust. The Trust holds special Class T shares, which carry no financial value whatsoever. They exist for exactly one purpose: to elect directors.

The trustees hold no equity in the company. They do not share in its profits and are compensated only for their time. They select their own successors, merely consulting the company in the process. Terms last just one year.

Neel Buddy Shah, head of the Clinton Health Access Initiative, currently serves as chair. The other trustees are Richard Fontaine, head of a Washington think tank; Mariano-Florentino Cuéllar, former California Supreme Court justice; Novartis CEO Vas Narasimhan; and, as of July 2026, Nobel laureate and former Federal Reserve Chair Ben Bernanke.

The Trust’s authority grows in step with each funding round: the more money raised, the more board seats it appoints. On April 14, 2026, when Narasimhan joined the board, Trust representatives held an absolute majority for the first time, crossing a threshold written into the founding documents. That was seven weeks before the company filed to go public.

The closest analogue to this model is that of OpenAI. There, following its 2025 restructuring, a nonprofit sits above the commercial entity and appoints its entire board. The difference is that the nonprofit holds roughly 26% of the equity (approximately $130 billion at the time of the restructuring). Anthropic’s trustees own nothing.

However, the full text of Anthropic’s trust agreement has never been made public. For years, the Trust did not exercise its authority: it could have appointed three directors by the end of 2024, but added only one. Additionally, a qualified shareholder majority can amend the Trust’s powers without the trustees’ consent. The very mechanism designed to resist the pressure of capital can be rewritten by a sufficient consolidation of it.

Anthropic will have to lay all of this out in the risk factors section of the prospectus. For anyone buying shares, the inability to influence the company’s direction is a risk.

Warming up before the start

May 2026: Anthropic is valued at $965 billion, an official figure agreed upon with investors.

Early July: secondary market trades are being conducted at an implied valuation of $1.2 trillion. That is the price at which existing shareholders are selling to one another.

August: the press reports expectations of $2 trillion and above. Outlets are careful to note that the figure is coming from shareholders. Anthropic has not confirmed the valuation. Around the same time, Bloomberg reports that the annualized revenue run rate has broken through $65 billion — seven times higher than at the end of last year.

Three months, three valuations, each with its own source: the company itself, the secondary market, prospective investors. None of these numbers was pulled out of thin air. All of them are real; each simply follows its own math.

Anthropic isn’t the only LLM provider warming up. On June 8, OpenAI announced it had “recently” filed its own confidential S-1. The company didn’t provide a date.

The June 2026 IPO of Space Exploration Technologies Corp., or SpaceX, shows how these warm-ups can change the process of going public. Pricing its shares at $135, SpaceX raised $75 billion (nearly $85.7 billion after the underwriters exercised their over-allotment option) at a valuation of $1.77 trillion. That represented approximately 4% of shares issued. (A typical IPO brings 10–20% of shares to market.)

Instead of building an order book, the company announced a fixed price up front and allocated roughly 30% to retail investors, about six times the norm. Demand came in at a quarter of a trillion dollars, yet less than 5% of the stock was actually sold: there simply is not enough free capital in the market to absorb that much at once. Limited supply creates scarcity, scarcity drives up the price, and that price is multiplied by a whole that nobody actually buys.

This principle cuts both ways. On its first day of trading, SpaceX shares closed at $160.95, 19% above the listing price, pushing the company’s market cap past $2 trillion. On July 16, the stock fell below the listing price for the first time. By July 23, it had dropped to $118.25, implying a valuation of $1.51 trillion. By that point, roughly 185 million short positions had been opened, approximately 29% of the public free float.

SpaceX’s prospectus also revealed that Anthropic pays Elon Musk’s firm $1.25 billion a month for computing capacity in an agreement that runs through May 2029. The market’s go-to benchmark turned out to be the company’s landlord. Everyone here is playing several roles simultaneously.

Measures of uncertainty

Anthropic called its model Claude after Claude Elwood Shannon, who in 1948 defined information entropy as a measure of uncertainty in a message. The less predictable a message, the more information it carries: if you can already guess what it contains, the odds are against those contents being anything new.

Five years ago, people brought their own chairs to a San Francisco park and discussed an AI project. Today, their company is valued at more than $2 trillion — a figure put forward by the very entities that intend to buy in. The Anthropic IPO is expected as early as October, but the company hasn’t named a date, price, size, or exchange.

Anthropic means “relating to humanity.” It shares its root with the anthropic principle: the universe appears to us as it does because it must permit the existence of an observer. The group from Precita Park liked the name, and the principle holds here just as elegantly. How someone sees the company depends not only on the existence of the observer but also on where they’re standing.