In a tentpole moment for the industry, Anthropic filed to go public on June 1. We can’t read the filing. That’s not a complaint about access; it’s the whole point of the move. The company submitted its draft S-1 to the SEC confidentially, which means the document everyone actually wants to see, the one with the audited financials, the risk factors, and the cap table, is sitting under review where only Anthropic and its bankers can read it. What the public got was a single sentence published under Rule 135 of the Securities Act, confirming that a draft registration statement exists and that an IPO might follow.

So there isn’t much to analyze in the document itself. The decision to file it, this way and this early, is another matter. Two months ago, I wrote about Anthropic’s “Great Reversal”, and argued that its enterprise-first focus had carried it past OpenAI on revenue in just over two years. The S-1 is where that bet gets tested in front of an audience that doesn’t take run-rate on faith. The most telling detail about June 1 is that Anthropic chose to open its books first. That’s only an advantage if the books can survive being read, which is the real signal here, and it runs straight out of the business model.

What a confidential filing actually is

Confidential submission is a standard option available to any issuer, not a signal about Anthropic’s scale or stage. What it buys is two things: privacy during the revision process, with comment letters and disclosure changes handled before anyone outside the deal reads a word, and timing flexibility. A confidential draft only needs to go public 15 days before the roadshow, so a filing now points to a public S-1 and roadshow months out, plausibly this fall.

It also means the whole thing is reversible. A draft can be amended again and again, or pulled outright, and companies have filed confidentially only to never go public, with the rest of us none the wiser. June 1 doesn’t tell us Anthropic is going public. It tells us Anthropic bought the option to, on its own clock, and wanted that known.

The numbers that made bankers comfortable

We can’t see the financial statements, but we can triangulate from what Anthropic and its investors have already disclosed, and the picture explains why underwriters were willing to price this where they did.

Start with revenue. The run-rate reportedly hit roughly $47 billion in May 2026, up from about $10 billion a year earlier. That continues the curve I traced in April, when annualized revenue had gone from around $87 million at the start of 2024 to $30 billion. If $47 billion holds, Anthropic has kept up something close to 10x growth for a fourth straight year. There is no more important sentence a banker can put in front of a public-market investor.

The valuation marks are just as aggressive. The filing follows a $65 billion Series H led by Altimeter, Dragoneer, Greenoaks, and Sequoia, which lifted Anthropic’s post-money valuation to roughly $965 billion. That figure does real work: it puts Anthropic ahead of OpenAI’s reported $852 billion private mark and makes a trillion-dollar debut the base case rather than the stretch goal, assuming markets cooperate. Early, unconfirmed reports also have Anthropic tracking toward a roughly $559 million operating profit on $10.9 billion in revenue for the quarter ending June 30. Treat that one as a rumor until an audited number replaces it, but if anything close to it holds, the profitability question that has hung over every frontier lab gets far easier to answer in a prospectus.

I’d attach the same caveats I used in April, and the public S-1 will drag all of them into the light. Run-rate is the best recent month annualized, not what actually landed over twelve months, and for these companies the gap between the two has been wide. Anthropic also books reseller revenue from AWS and Google Cloud on a gross basis, counting the full end-customer spend and treating partner payouts as a cost, which flatters the top line against peers who report net. One good quarter of operating profit is not durable free cash flow. The whole discipline of an S-1 is that audited GAAP numbers replace run-rate headlines, so when this goes public, the pages worth reading won’t be the growth chart. They’ll be the footnotes that reconcile it.

First to the SEC, in a field of giants

What gave the filing its weight was the timing relative to everyone else. Anthropic is the first of the AI-era giants to reach the SEC, and it did so in what’s shaping up to be the most crowded IPO window anyone can remember.

SpaceX is reportedly set to start its roadshow as early as June 4 and price around June 11, chasing a valuation of $1.8 to $2 trillion and a raise of up to $75 billion. OpenAI, off its $852 billion private mark, is reportedly aiming to list as soon as September. Add the three together, and the targeted valuations approach $3.6 trillion, with roughly $200 billion in proceeds raised over a matter of months. That combined number is in the range of France’s annual GDP, hitting the public markets in a single stretch.

That concentration is the core of the bear case. The money that goes into three mega-listings has to come from somewhere, and the fear on Wall Street is that it gets pulled out of existing positions, drives tech’s weight in the S&P 500 to a record, and hands the public an unprecedented amount of late-stage, pre-profit risk near the top of a cycle. The banks running the deals, JPMorgan, Goldman Sachs, and Morgan Stanley among them, argue 2026 has plenty of liquidity, that global investors are actively hunting AI exposure, and that the window can absorb all three. They might both be right. A deal can clear cleanly and still mark the top.

This is where going first stops being a generic race-win and turns into a statement about the business. Filing first means going into the audit first, on the public record, while competitors watch. That is only a position of strength if your revenue is the kind that survives GAAP: enterprise contracts billed to real companies, not a consumer free tier that costs money to serve and books no revenue.

The enterprise-first model I described in April is precisely what lets Anthropic treat opening its books as an edge rather than an exposure. The risk runs the other way, too. Whoever prices first sets the comparable that everyone else is measured against, and volunteers to be the market’s test of whether an AI company at this scale can be priced at all. Anthropic is wagering that its fundamentals are good enough to want that test instead of fearing it.

Why now, and why quietly

The logic of filing confidentially in June 2026 reads as a bet on two windows, one opening and one closing.

The open one is demand. AI is the theme the public markets unambiguously want, and the revenue and valuation marks have never looked stronger. Casting its filing into that appetite, ahead of two rivals fishing for the same dollars, is the obvious play. The closing window is a different kind of urgency, it’s pressure. A company worth nearly a trillion dollars on paper, with employees sitting on fortunes they can’t sell and a secondary market already pricing the stock, can’t stay private forever. An IPO answers the liquidity question and the price-discovery question at once.

Doing it confidentially is about control on top of all that. Anthropic gets to hash out its hardest disclosures, its compute commitments, its customer concentration, and its unusual governance, with the SEC before any competitor or customer reads them. It keeps its options open if the SpaceX or OpenAI deals sour the mood. And it gets to plant the “first to file” flag without committing to a date.

What the document will have to say out loud

The most useful thing you can do with a confidential S-1 is anticipate the disclosures it will eventually be forced to make, because that’s where the real tension in Anthropic’s story lives.

The biggest is concentration, on both sides of the ledger. Anthropic’s compute is bound up with Amazon and Google, who are at once its largest infrastructure providers, major investors, and, through their own models, competitors. The same reseller arrangement that pads the top line also deepens the dependency. A prospectus has to spell all of that out as a risk factor in plain English. Seeing it quantified (how much revenue runs through how few partners, how much compute sits with how few clouds) will be far more clarifying than any run-rate number.

Then there’s governance, which is what makes Anthropic genuinely strange as an IPO candidate. It’s a public benefit corporation, and its board answers in part to a Long-Term Benefit Trust charged with weighing safety and the public interest against shareholder returns, sometimes above them. For the first time, public investors will have to price a company that has formally reserved the right to put their interests second. The S-1 will have to say, in so many words, exactly how much control they’re buying and how much they aren’t.

Then there’s the safety-versus-commercialization tension I flagged in April, now with real money on it. Anthropic’s Responsible Scaling Policy v3.0, from February 2026, walked back its most distinctive promise, the explicit pledge to pause development if its safety measures fell short. In a prospectus, “we may voluntarily slow or halt our most valuable product line for safety reasons” is a textbook risk factor. The brand depends on investors believing Anthropic means it. The valuation depends on them believing it almost never comes to that. Watching the company reconcile those two in the dry language of an S-1 will be one of the more revealing reads of the year.

None of this is a reason the IPO falls apart. It’s why the public S-1 will be a better read than the confidential one, and why the real story plays out in the gap between “filed” and “priced.”

What the filing tells us

Strip it down and June 1 is an act of confidence with a hedge built in. Anthropic believes its books can withstand SEC scrutiny now, and that the demand environment pays whoever moves first. The confidential route is the hedge: it claims the first-mover story without committing to a date, and leaves a path to retreat if SpaceX or OpenAI sour the market before Anthropic has to price.

What June 1 can’t settle is whether that confidence is warranted. The figures we’re working from are run-rate and private marks, not audited GAAP. And if the deal comes, it prices into a window already crowded with the two largest private companies on earth, in a market that can’t decide whether this is the start of a re-rating or the end of one.

The through-line from The Great Reversal is that Anthropic’s choices on the product side keep resurfacing later as leverage somewhere else. Building on enterprise revenue instead of consumer scale is what now lets it open its books before anyone else and call that a strength rather than a risk. Whether public markets reward that the way private investors have is still open, and filing confidentially is how Anthropic gets to pose the question without yet having to hear the answer.

Sources: Anthropic’s confidential S-1 announcement; Anthropic Series H announcement; TechCrunch; NPR; Fortune; Yahoo Finance; IG, indmoney, and TradingKey coverage of the 2026 SpaceX/OpenAI/Anthropic IPO race. Financial figures come from public reporting and private funding disclosures, not from the confidential S-1, whose contents are not public.