Anthropic’s early IPO and the subsequent secret listings of AI giants reveal the underlying battles and strategies.

Are you ready? Starting this June, we may witness the biggest wave of IPOs in human history.

On May 20, SpaceX filed its S-1 filing with the SEC, aiming to list on Nasdaq on June 12. Its valuation exceeds $1.8 trillion, with a fundraising target between $40 billion and $80 billion. On the same day, news broke that OpenAI was secretly preparing to file its IPO prospectus, with Goldman Sachs and Morgan Stanley jointly underwriting, aiming for a September listing. Its latest round of financing corresponds to a valuation of $852 billion.

Late last night, Anthropic officially announced that it had filed its S-1 document with the SEC. Last week, they completed a $65 billion Series H funding round, causing their valuation to surge from $38 billion in 2024 to $965 billion, surpassing OpenAI to become the world's most valuable private AI company.

Combined, these three IPOs this year could raise more funds than all venture-backed IPOs in the US since 2000. For reference, data from Renaissance Capital shows that the total amount raised in the US IPO market in the past five months this year was only $28.4 billion.

Secret IPO, we'll fix the books before showing them to you.

The original meaning of IPO is "public," but among these three giants, Anthropic and OpenAI both chose to file confidentially. SpaceX also used the same method in April, only disclosing it to the public in May.

In 2012, the JOBS Act, signed by President Obama, opened a green channel for eligible "emerging growth companies": draft prospectuses could be submitted confidentially to the SEC. All reviews, feedback, and revisions proceeded normally, but the S-1 form was not immediately made public; it was only required to be disclosed 15 days before the roadshow.

An S-1 form is like a company's most naked X-ray. SEC regulations require that the registration statement include audited financial statements, complete business and risk disclosures, management's analysis of the company's performance, details of executive compensation, and the use of proceeds. In other words, revenue, cost structure, customer information, and even how much the founders are paid will all be laid bare in this form.

For a company preparing to go public, the S-1 form is its trump card. Therefore, the secret filing channel became one of the most important institutional reforms in the US capital markets in recent decades, and the most popular star policy among companies in the JOBS Act.

There are at least three benefits to secretly submitting your cards in order to protect them.

First, the negotiations with the SEC were conducted privately and kept entirely confidential. The SEC would review the draft proposal round by round and provide feedback, to which the company would respond and make revisions. Which figures needed to be added, which statements posed legal risks—the process was repeatedly refined until both parties were satisfied before it was released to the public.

Secondly, it allows for information warfare within the fierce competition among AI vendors. As a competitor, Anthropic would never want OpenAI to see its true cost structure, and vice versa. The computing power expenditure, inference cost ratio, and reliance on major clients in S-1 are nuclear-weapon-level business intelligence for competitors. Secret submission means that the company can hold onto this information for as long as possible.

Finally, it provides an escape route. A secretly submitted draft does not constitute a formal registration statement until it is made public. The JOBS Act allows companies to terminate the process at any time during this stage, simply by not proceeding. This allows for a quiet exit when market conditions are unfavorable, avoiding headlines about a "failed IPO."

They're all good cards, but you can't look at them.

So here's the question: to maximize the benefits of a secret commit, one should remain completely silent and make a fortune without saying a word, yet Anthropic posted an announcement on its website. Why are these companies all announcing their "secret commits" and doing the opposite?

The submitted content was confidential, but the act of submitting itself was public, even high-profile, creating a carefully designed information asymmetry. Publicly declaring that a secret submission had at least several effects:

First, it stimulated valuation. Anthropic's Series H funding round and confidential filing were almost simultaneous. It was impossible for the investors not to know about their upcoming IPO. The expectation of an IPO is itself part of the pricing, which directly boosted the valuation of the private round, pushing it to a high of 965 billion, surpassing OpenAI.

The IPO news also effectively reassured clients and the team. Anthropic relies on businesses for 80% of its revenue, and major clients need to know that their finances are stable and healthy, and that they can cooperate long-term. For the team, the promised stock options now offer a glimmer of hope for realization.

Finally, there's the warm-up effect: the headline alone, "Trillion-Dollar AI Company to IPO," is enough to get the underwriting firms and institutions moving. They don't even need to look at the S-1 certificate; the roadshow hasn't even started yet, and the atmosphere is already buzzing.

In the game of capital, "disclosure" has always been a conditional, structured, and tiered disclosure. What is disclosed is not the information itself, but the hierarchy of information. Who sees it first, who can understand it, and who can act on it are all predetermined from the moment it is secretly submitted.

Not making money, or the definition of "making money" is very flexible.

Of the three most anticipated companies preparing for an IPO, SpaceX and OpenAI are not yet profitable; OpenAI, for example, is projected to generate $13 billion in revenue by 2025 but will still be operating at a loss. Anthropic is the only one that claims to be profitable.

According to an exclusive report by the WSJ, Anthropic expects to achieve revenue of $10.9 billion and operating profit of $559 million in the second quarter of 2026, compared with revenue of $4.8 billion in the previous quarter. This would be the company's first profitable quarter since its inception.

It sounds like good news, making its IPO seem like a natural progression, but a closer look at the timeline reveals many intriguing details.

Anthropic signed two new large computing power contracts this year, one from SpaceX and the other from xAI.

The first two months of these two contracts were at a discounted price, a so-called ramp-up period, with full-price billing of $1.25 billion per month starting in July. If we trace back the profit cycle, these two discounted months exactly constitute the cost basis for Q2's "profitability." Once the discounts end, computing power demand will drive up expenses, and whether profitability can be maintained is uncertain.

Moreover, this is a non-GAAP calculation. Private companies are not bound by the financial reporting standards of publicly traded companies, and their profit calculations are more flexible. Tech blogger Ed Zitron did the math: if all the computing power contracts with AWS, Google Cloud, and SpaceX were calculated at full price, the monthly cost could reach as high as $3.75 billion, which would be close to $45 billion a year—the $559 million operating profit is almost negligible in the face of this figure.

The timing of the leak is also worth noting. Anthropic's profit forecast was leaked on the same day NVIDIA released its earnings report. On one hand, the largest seller of AI infrastructure said chip demand was still rising; on the other hand, the largest buyer said it had already started making profits. These two pieces of news, overlapping, conveyed the same message to the market: the money invested in AI is not wasted. For the upcoming IPO roadshow, there couldn't be a better prelude.

There is another set of questionable figures. ARR, annualized recurring income, is the annualized estimate obtained by multiplying the current monthly income by 12, a metric that once caused Manus to stumble.

Anthropic's annualized revenue growth trajectory is unprecedented: approximately $1 billion at the end of 2024, reaching $14 billion in February 2026 (Anthropic itself disclosed in its Series G announcement: "our run-rate revenue is $14 billion"), surpassing $30 billion in April, and approaching $44 billion in May.

The four figures form a nearly vertical curve. However, CFO Krishna Rao stated in sworn testimony in court this March that the company's cumulative total revenue "exceeds $5 billion." If the ARR figures are accurate, then Q1 alone generated $4.8 billion in revenue, meaning the company has barely made any money in the past few years and relied entirely on this year's Q1 growth.

Given Claude Code's phenomenal start to the year and Codex's recent rise, the surge in Q1 revenue is understandable. However, OpenAI consistently boasts over 800 million weekly active users , a number far exceeding Anthropic's. Anthropic's valuation of $965 billion, higher than OpenAI's $852 billion, begs the question: what kind of pricing are investors assigning?

But in any case, investors don't want to wait any longer.

From 2022 to 2024, the US IPO window was practically closed for three years. The combination of interest rate hikes and market panic led to high-valuation tech companies experiencing immediate price drops after listing, a truly dismal situation. These three years also coincided with the peak of the AI ​​bubble, with paper valuations increasing several times, even tenfold.

Paper wealth remains just paper wealth. VC funds have a cycle; LPs invest money and expect returns upon maturity. Fund duration is typically 10 years, and most early-stage AI investments were completed between 2018 and 2021, meaning 2026 is already approaching the settlement period.

With the interest rate environment improving, market sentiment recovering, and the window reopening, this is almost the best time.

Whoever is in charge of the company, whoever bears the consequences.

However, Anthropic is not an "ordinary company".

It is a public interest corporation registered in Delaware, abbreviated as PBC. PBC's board of directors is legally required to balance shareholder interests with its public mission. Anthropic's public mission, enshrined in its bylaws, is "to responsibly develop and maintain advanced AI for the long-term benefit of humanity." Based on this, the company has also designed a long-term interest trust mechanism called LTBT, where independent trustees elect a portion of the board seats. This proportion will gradually increase to a majority on the board, theoretically ensuring that the security mission is not overshadowed by commercial interests.

This governance structure has never been tested in the public market, and the first challenge it will face is cost.

Once the full switch takes place next month (July), the combined monthly computing power bill from AWS, Google Cloud, and SpaceX could reach as high as $3.75 billion. According to a January report by The Information, Anthropic's inference costs were 23% higher than expected.

This discrepancy ultimately has only two possible outcomes: either it can be passed on to users, meaning price increases, reductions in free credit limits, and lower quality models for free tiers; or it can be passed on to financial statements, as the company switches from non-GAAP to GAAP accounting after going public, eliminating any room for flexible accounting and requiring the company to report gross margins to the public market every quarter. However, this means pressure on profit margins and stock price volatility.

Currently, Claude Pro costs $20 per month, and Max costs $100 to $200 per month. After its launch, these prices will be influenced by two forces: users' willingness to pay and Wall Street's expectations for gross margin. Historical experience suggests that the latter often has a stronger influence.

For paying users, the public release of S-1 is a noteworthy milestone. The prospectus discloses core data such as total revenue, detailed revenue costs (including computing power expenditures), and user scale. For the first time, analysts can use this data to calculate the computing power cost per user and the profit margin. Questions such as where the pricing ceiling is and how much room there is for price increases now have a basis for calculation.

Two paths, two products, the future

Going back to the previous question: Why does Anthropic have a much smaller user base than OpenAI, yet its valuation is higher?

These are two companies with drastically different business models. Anthropic derives 80% of its revenue from enterprise clients, and eight of Fortune's top ten clients are its customers. Claude Code contributes over $2.5 billion in annualized revenue. While its user base is small, the value of each individual user is extremely high—this is the valuation logic behind an enterprise services company.

OpenAI's logic is exactly the opposite. With 800 million weekly active users, primarily consumers, and a much lower ARPU, it follows the valuation logic of a consumer internet company.

The two valuation logics will rapidly diverge into two product paths after the IPO. Anthropic's enterprise customers account for 80% of its revenue, and its product roadmap will continue to tilt towards the needs of large clients: stronger stability and compliance capabilities, and a more conservative pace of feature releases (the restricted release of Claude Mythos reflects both the aforementioned security mission and aligns with enterprise clients' requirements for predictability). Individual users will not be abandoned, but their priority will gradually shift downwards.

The OpenAI roadmap means monthly active users are the core asset, putting greater pressure on monetization through consumer spending. Prior to Codex, they considered adult content modes, the Sora app, and advertising, and it's possible that more aggressive monetization methods will be launched soon. After all, when user volume is your biggest story, you need to give users a reason to continue paying.

The divergence between these two paths is already underway, and IPOs will accelerate this process with quarterly reports. Users choosing which AI ecosystem to embrace is essentially choosing which business model to pursue.

The S-1 will eventually be released. For investors, it's financial data; for users, it's a product roadmap. Whether subscription prices will increase, whether the free tier will be scaled back, and which way the balance between security and growth will tip—the answers are all in those hundreds of pages. It's worth taking the time to read it when it's released. Even if you're not using it for stock trading, you can still understand what might happen to the tools you use every day.

#Welcome to follow iFanr's official WeChat account: iFanr (WeChat ID: ifanr), where more exciting content will be presented to you as soon as possible.