The AI IPO Pipeline: What OpenAI and Anthropic Will Have to Disclose

In the first week of June 2026, two of the largest private companies ever built filed confidentially with the SEC within days of each other. Anthropic went first. OpenAI followed on 8 June, working with Goldman Sachs and Morgan Stanley, at a private valuation reported around $852 billion. Anthropic's most recent round was reported at approximately $965 billion.

Days later, SpaceX listed and became the largest IPO in history. Five weeks after that it traded below its issue price.

Those three events are related, and the relationship is the most useful thing to understand about this pipeline. This guide covers what a confidential filing actually means, what the registration statements will be required to disclose that private rounds never demanded, what SpaceX established as the reference point, and what to watch when the documents become public.

1. What a Confidential Filing Actually Is

The headlines said both companies “filed for an IPO.” That overstates it, and the distinction matters.

Eligible issuers may submit a draft registration statement to the SEC confidentially, beginning the review and comment process without publishing anything. The document is not public. The company receives comments, revises, and only makes a public filing when it intends to proceed — typically at least fifteen days before the roadshow.

What that means in practice:

• It is a preparation step, not a commitment. A company can complete confidential review and then wait indefinitely, or abandon the process entirely, without the reputational cost of a publicly withdrawn filing.
• It starts a clock that is largely invisible. SEC comment rounds take months, and doing that work privately means a company can be ready to launch inside a narrow window when conditions are favourable rather than beginning a six-month process after the window opens.
• It preserves optionality on timing. Reports that OpenAI may push its debut into next year are entirely consistent with having filed — that is precisely the flexibility the confidential route buys.

So the correct reading of a confidential filing is: this company has decided to be ready. It is not: this company is going public on a known date. Our guide to the two-year preparation timeline explains why the filing is close to the last step rather than the first.

2. The Valuation Problem

Both companies carry private marks near or above a trillion dollars — set by investors in negotiated rounds, on terms and information not available to anyone else.

A public listing changes the mechanism entirely. A private round requires one investor to accept a price. A public offering requires a book of institutions to accept it simultaneously, with a prospectus in front of them, knowing they will be marked to market daily afterwards.

Those two processes do not reliably produce the same number, and the gap is not always in the direction founders expect. Private marks can understate a company — a strategic investor accepting a lower price for access — or overstate it, particularly where a round included structure such as liquidation preferences or ratchets that make the headline valuation misleading. Our guide to liquidation preferences covers why a headline private valuation and the economic value of common stock frequently diverge.

The specific complication in this pipeline is that neither company has a clean comparable. Existing public software companies do not have their cost structure; existing public infrastructure companies do not have their growth. That is precisely the condition that produced a public disagreement over SpaceX's value spanning more than a factor of two — and it will recur here.

3. What the S-1 Will Force Into the Open

This is the section worth reading closely, because it is where narrative valuations meet audited disclosure. A registration statement requires things private rounds never demanded.

Gross margin, and what sits inside cost of revenue

The central open question about frontier AI economics. Every model inference has a marginal compute cost, and how that cost is classified — cost of revenue versus research and development — determines the reported gross margin. Investors will read this line first, because it decides whether these are software businesses with software multiples or something structurally closer to capital-intensive infrastructure.

Training costs and how they are treated

Whether the cost of training a model is expensed as incurred or capitalised and amortised has an enormous effect on reported profitability, and the accounting policy note will be scrutinised harder than the income statement. Related: research and development spend that is genuinely creating a durable asset is viewed very differently from spend required simply to stay competitive.

Revenue quality and mix

Consumer subscription, enterprise contracts, API consumption and any large partner arrangements have completely different durability. Consumption-based revenue can decline without a customer churning, simply because usage fell. Investors will want the split, the retention data by cohort, and the contracted versus consumption breakdown — which is the same analysis described in our guide to the metrics investors underwrite.

Customer and partner concentration

Filings must disclose customers above a materiality threshold. In a market where a small number of very large enterprises and cloud providers account for a substantial share of activity, this disclosure is likely to be more revealing than any other single item.

Compute commitments

Multi-year purchase obligations for infrastructure appear in the contractual commitments table. These are effectively fixed costs stretching years forward, and their size relative to revenue is a direct measure of operating leverage in both directions.

Governance and control

Both companies have unusual structures by conventional standards. Registration statements require the control arrangements, board composition and any dual-class provisions to be set out plainly, with the associated risk factors. For companies whose governance has itself been a subject of public dispute, this is a section that will be read with unusual attention.

Legal and regulatory exposure

Pending litigation, intellectual property claims relating to training data, and regulatory developments across multiple jurisdictions all become disclosed risk factors. Nothing concentrates the mind on legal exposure like having to describe it in a document that carries liability.

4. What SpaceX Established as the Reference Point

SpaceX is the most relevant precedent available, and not for the reason usually cited.

The lesson is not that a category-defining company can raise an enormous sum — it demonstrated that emphatically, raising $75 billion. The lesson is what happened over the following five weeks: a 19% first-day gain, addition to the Nasdaq-100, and then a slide through the index-inclusion flow to below the $135 issue price by mid-July.

Three transferable observations, covered in full in our SpaceX IPO case study:

• Scarcity of access drives the debut; disclosure drives everything after. Day one prices the excitement of being able to buy at all. The subsequent months price the filing.
• Segment economics change the story. SpaceX's S-1 revealed that its AI division consumed more in 2025 than its only profitable business earned. That was not visible in any private round.
• Index inclusion is a flow, not a bid. It generates one wave of mechanical buying and then stops.

Every bank pitching the AI deals is modelling that aftermarket. Every institution deciding whether to take an allocation is looking at it. The reference point for pricing these offerings is not SpaceX's debut — it is SpaceX at sixty days.

5. The Retail Question

OpenAI's CFO has indicated the company intends to allocate IPO shares to retail investors, which is a meaningful departure from the norm and worth understanding on both sides.

Why an issuer would do it. A large, engaged consumer user base is a natural shareholder base, and allocating to it converts users into advocates. It also broadens the register beyond institutions that will trade the position on quarterly results.

Why it is complicated. Retail allocations in heavily oversubscribed deals are typically small per investor, which produces disappointment. And a retail investor who receives allocation at the issue price is in a genuinely better position than one who buys in the aftermarket — the two get conflated in coverage, and the second group is the one that historically fares worst.

The honest framing for individual investors: an allocation and a market purchase on day one are completely different transactions. The first participates in the underpricing discount. The second pays it.

6. What to Watch When the Filings Go Public

A practical checklist for reading either S-1 when it appears, in the order the numbers actually matter.

1. Gross margin, and the accounting policy note explaining what sits in cost of revenue. Everything else follows from this.
2. The revenue split between consumer, enterprise and API, with retention by segment.
3. Contractual compute commitments as a multiple of current annual revenue.
4. Customer concentration, and whether any disclosed large customer is also an investor or infrastructure partner — related-party revenue is treated very differently by analysts.
5. Training cost treatment, and what reported profitability looks like under the alternative policy.
6. The share count and structure, including any dual-class arrangement and the full diluted number rather than the basic one.
7. Use of proceeds. Whether the money funds compute, general corporate purposes, or selling shareholders. A large secondary component means existing holders are taking money off the table.
8. The risk factors nobody drafted casually — skip the boilerplate and read the ones written specifically for this company.

7. What This Means If You Are Not Buying the Stock

Most readers of this will not be allocating to a mega-cap IPO. The pipeline still matters, for three reasons.

It sets the comparable set for private rounds. Once these companies trade publicly, every AI company raising privately is valued partly against an observable multiple rather than against a negotiated private mark. That is a substantial change, and for most companies it will be a downward one — public multiples are almost always lower than late-stage private ones.

It unlocks liquidity that recirculates. Employees and early investors at these companies will eventually sell, and a meaningful share of those proceeds returns to the ecosystem as angel investment and new fund commitments. Large listings are historically followed by a wave of new managers and active angels — which our guides to angel portfolio construction and raising a first fund both address.

It tests whether the category has an exit path at all. A generation of AI companies has been funded on the assumption that the largest players would eventually be worth extraordinary sums in public markets. These listings are where that assumption gets its first real examination.

Frequently Asked Questions

Does a confidential filing mean an IPO is imminent?

No. It means the company has begun SEC review privately and wants to be ready. Companies routinely file confidentially and then wait months or longer, and some never proceed. The public filing, roughly fifteen days before a roadshow, is the signal that a deal is actually happening.

Can I invest before the IPO?

Only through the private secondary market, and with real caution. Shares in these companies trade through brokers and structured vehicles, frequently at prices disconnected from any recent primary round, sometimes through layered structures where you own an interest in an entity rather than the shares themselves. Understand exactly what you would own, what fees apply, and whether the company permits the transfer — our guide to secondaries and tender offers covers the mechanics and the traps.

Why would a company at a trillion-dollar private valuation need to list?

Liquidity for employees and early investors, principally. A company more than a decade old with heavily vested staff faces genuine retention pressure if nobody can sell. Access to permanent capital at scale, and a publicly traded currency for acquisitions, are the other conventional reasons.

Which is likely to list first?

Unknowable from outside, and both companies have preserved the flexibility not to answer it. Anthropic filed first and OpenAI a week later; reports since have suggested OpenAI's timeline may extend. Filing order tells you very little about listing order, because the decisive variable is market conditions in the window each chooses.

Should I buy on the first day?

The structural point is the same one the SpaceX sequence illustrated: day-one buyers transact in the most information-asymmetric session a stock will ever have, against sellers who received allocation at a discount. Waiting for the lockup expiry and two quarters of reported results costs you the pop and removes most of the asymmetry. That is an observation about market structure, not a prediction about these companies.

The Bottom Line

Two confidential filings weeks apart signal that the largest AI companies intend to be ready, not that a date exists. The interesting moment is not the filing — it is the day the S-1 becomes public and the gross margin line, the compute commitments and the customer concentration stop being private.

SpaceX already demonstrated what the sequence looks like: enormous demand for access, followed by a market that prices the disclosure. Watch the sixty-day chart, not the debut.

Global Capital Network connects founders, private shareholders and investors across the transition from private to public markets. See upcoming events or get in touch.

Valuations and filing status are as reported at early August 2026 and change frequently. This article is analysis and general information, not investment advice. Verify current facts before relying on any of them.

Sources: CNBC: OpenAI confidentially files for IPO · TechCrunch: OpenAI files confidentially, following Anthropic · CNBC: OpenAI to allocate IPO shares to retail investors · CNBC: OpenAI IPO timeline reportedly delayed