Anthropic Filed Its S-1: The Two Biggest AI IPOs Ever Are Coming

On June 1, 2026, Anthropic confidentially filed a draft S-1 with the Securities and Exchange Commission — the formal first step toward what will almost certainly be the largest technology IPO in history. The company enters the process with a revenue run-rate of roughly $47 billion as of May 2026 and a post-money valuation of $965 billion following its $65 billion Series H. OpenAI is widely expected to file its own registration, setting up the two largest AI listings of 2026 and arguably the two largest IPOs ever. But the size is not the story. The story is disclosure. For three years, the AI bubble debate has been conducted with leaked numbers, investor spin, and selectively quoted run-rates. An S-1 is different: audited financials, mandatory risk factors, and a legal regime where misstatements carry liability. The argument about whether AI economics actually work is about to be settled — or at least conducted, for the first time, with real numbers.

The Filing: What We Actually Know

A confidential draft S-1 is exactly that — confidential — so the public record is limited to what the company and its investors have disclosed. Three numbers frame the offering. First, the run-rate: approximately $47 billion in annualized revenue as of May 2026, a figure that would have sounded like a typo eighteen months ago, when the company was tracking toward single-digit billions. Second, the valuation: $965 billion post-money after the $65 billion Series H, which makes Anthropic the most valuable private company ever to approach the public markets. Third, the timing: a June 1 confidential filing typically implies a public flip of the registration statement in late summer and a listing in the fall window, assuming the SEC review and market conditions cooperate.

The valuation math is worth a beat of attention. At $965 billion against a $47 billion run-rate, Anthropic would price at roughly 20x forward revenue — rich by software standards but not absurd by them, and dramatically more grounded than the multiples that prevailed across AI's private market in 2024–2025. The question public investors will actually interrogate is not the multiple on revenue but the multiple on gross profit — and that depends entirely on a number no outsider has ever seen audited: what it costs, in compute, to generate a dollar of Claude revenue.

Why file now? Three pressures converge. The first is arithmetic: at nearly a trillion dollars of private valuation, the remaining pool of private capital deep enough to fund the next compute cycle is approximately one — the public markets. The second is obligation: employees holding seven years of paper wealth, early investors with funds reaching end-of-life, and strategic partners whose own shareholders want the position marked by something more rigorous than the next private round. The third is the window itself. IPO windows are weather, not climate — they open when public-market appetite, rates, and sector sentiment align, and they close without notice. The AI trade has dominated public equities for three years; an AI lab that waits for a better window is betting that enthusiasm survives whatever its own S-1 reveals about the industry's economics. Filing first, before a rival's disclosures or a macro shock can reframe the category, is the conservative play dressed as the bold one.

Why OpenAI Almost Has to Follow

OpenAI's expected filing is less a choice than a convergence of pressures. The company raised $110 billion in February 2026 toward a valuation approaching $1 trillion, and capital at that scale has a limited set of exits: the public markets are effectively the only pool deep enough. Its investors — including strategic partners whose own balance sheets carry the relationship — need a liquidity path. And there is a competitive-narrative problem: if Anthropic lists first, it sets the comparables, frames the disclosure standards, and absorbs the institutional demand for "the public AI lab" allocation. Letting a rival define the category's public valuation is not a position OpenAI's board will accept passively.

The competitive backdrop sharpens the urgency. ChatGPT's share of AI web traffic has fallen to 54.7%, down from 76.5% in February 2025, while Google's Gemini has climbed to 27.4% and continues to gain. Dominance eroding at that pace changes what an S-1 needs to prove. A company with stable share can sell a growth story on the category's expansion alone; a company losing twenty points of share in fifteen months must show — with cohort data, retention curves, and enterprise contract metrics — that its revenue is durable rather than a receding tide's high-water mark.

What the S-1s Will Finally Reveal

The reason these filings matter beyond finance is that the AI industry's central empirical questions have been unanswerable from outside. Private companies disclose what flatters them. Run-rates get announced; cost structures do not. Revenue gets characterized as "annualized" at the steepest point of the curve; churn never gets characterized at all. An S-1 ends that asymmetry in four specific places.

The third item deserves particular attention, because it connects to the structural question we examined in our analysis of the AI boom's circular deals: how much of the industry's revenue is the same capital recirculating between vendors, investors, and cloud providers? S-1 related-party disclosure rules are precisely designed to surface this. When Anthropic's filing goes public, the footnotes will show what fraction of its revenue comes from parties that are also its investors or compute suppliers — and the same scrutiny will apply to OpenAI's relationship with Microsoft. Whatever the numbers turn out to be, the era of arguing about them from leaks is over.

The fourth disclosure — compute purchase commitments — may be the sleeper. Frontier labs have signed multi-year, take-or-pay agreements for chips, data-center capacity, and cloud services whose aggregate size has been reported in fragments but never totaled under audit. These commitments function like off-balance-sheet leverage: they are fixed obligations that must be paid whether or not revenue growth materializes to absorb them. A company with $47 billion of run-rate and, hypothetically, several hundred billion dollars of committed compute purchases is making a very specific bet about demand growth, and the contractual-obligations table will let analysts price that bet for the first time. If the AI demand curve flattens, the commitments do not. That asymmetry — growth optionality against fixed compute obligations — is the real risk architecture of the frontier-lab business model, and it has never been visible.

The Risk-Factor Section Will Be a Genre Unto Itself

Beyond the financials, the S-1's risk-factor section will be the most consequential piece of AI-industry writing published this year — because securities law requires companies to enumerate, with specificity, everything that could materially hurt them. For a frontier lab, that list is extraordinary: regulatory regimes on three continents that could restrict model deployment; the possibility that open-weight models commoditize the product entirely; dependence on a small number of chip suppliers and cloud partners who are simultaneously competitors; safety incidents that could trigger liability or deployment pauses; and key-person concentration that venture investors tolerated but index funds will price. Companies routinely under-disclose in marketing and over-disclose in risk factors, because the liability asymmetry points in opposite directions. Read the risk factors as the only document where a frontier lab is legally incentivized to argue the bear case against itself.

Public-Market Discipline vs. Private Hype

The deeper shift is epistemic. Private AI valuations are set by the most optimistic marginal investor in a competitive round — a mechanism that structurally selects for belief. As we argued in our analysis of OpenAI's valuation and the bubble question, a private company can sustain an enormous gap between narrative and economics for years, because no one with standing to challenge the numbers has access to them. Public markets invert every one of those properties: prices set continuously by buyers and sellers including skeptics, quarterly audited reporting, short sellers with financial incentive to find the weak footnote, and analysts paid to model unit economics rather than to win allocation in the next round.

It is worth remembering how violently public markets react to new information about AI economics. When DeepSeek's R1 demonstrated that frontier-adjacent capability could be trained at a fraction of assumed cost, Nvidia lost roughly $600 billion of market capitalization in a single session — an episode we dissected in our analysis of DeepSeek R1 and Nvidia's market cap loss. That repricing happened to a company adjacent to the question. A public Anthropic or OpenAI will be the question — every research release, every pricing change, every competitor benchmark will mark to market in real time. The labs are trading the comfort of private narrative for the deepest capital pool on Earth, and the price of admission is living inside that volatility forever.

History offers a calibration point. Google's 2004 IPO was received as the moment a speculative category — search advertising — would finally be audited, and the audit revealed a business far better than skeptics believed, repricing the entire internet sector upward. Facebook's 2012 listing ran the opposite script: a broken first day, a year underwater, and then vindication as mobile revenue materialized. Coinbase in 2021 marked the exact top of its cycle. The pattern across all three: the IPO did not cause the outcome, it revealed which story had been true all along. The AI labs' listings will do the same. If frontier economics are sound, the S-1s will prove it and the bears will capitulate. If they are not, no roadshow narrative will survive four quarters of audited reporting.

What It Means for Enterprise Buyers

If your company builds on Claude or GPT-family models, these IPOs are procurement events, not just market news. The first effect is positive: vendor stability. The persistent background worry of enterprise AI adoption — that your foundation-model vendor is a private company burning billions with opaque finances — gets replaced by quarterly visibility and a fortress balance sheet. Public companies also behave differently in contract negotiations: revenue durability starts mattering more than revenue optics, which historically pushes vendors toward longer-term enterprise agreements with real SLAs rather than land-grab pricing.

The second effect cuts the other way: margin pressure becomes pricing pressure. A public company that has promised investors expanding gross margins has three levers — cheaper inference, higher prices, or steering customers toward higher-margin products. Buyers should expect all three, and negotiate accordingly: lock multi-year pricing where you can, keep your architecture portable across model providers, and treat the newly published cost disclosures as negotiating intelligence. For the first time, you will know roughly what your vendor's inference actually costs. Use it.

What It Means for Startups

For the rest of the AI ecosystem, the IPOs create the thing the private market has lacked for three years: a real benchmark. Every AI startup's valuation conversation currently happens in a vacuum, anchored to whatever the last comparable private round paid. A publicly traded Anthropic and OpenAI replace that vacuum with daily-priced comps — revenue multiples, margin expectations, growth-rate premiums — that will propagate down through every Series B term sheet within a quarter. If the IPOs price strongly and trade well, the funding environment for credible AI companies loosens. If they price strongly and then break, the correction that follows will be swift and indiscriminate. Either way, the era of AI valuations untethered from any public reference point ends this year.

There is also a talent dimension that founders under-appreciate. A public Anthropic and OpenAI can offer something no private AI company can: liquid equity at a price the employee can verify every morning. For the past three years, frontier labs and well-funded startups competed for researchers with paper valuations of similar unverifiability. After the IPOs, the comparison becomes asymmetric — a senior engineer weighing a startup's options package against a public lab's RSUs is comparing a lottery ticket to a brokerage balance. Startups will have to pay for that asymmetry, either in cash, in deeper equity, or in genuinely differentiated missions. The second-order effect: expect a wave of senior departures from both labs twelve months after listing, when the first lockups expire and a generation of AI researchers becomes independently wealthy enough to found the next cohort of startups. That is how every platform era has seeded its successor.

Conclusion: The Audit the Industry Needed

The two biggest AI IPOs ever are coming, and their significance has little to do with their size. The AI buildout has been the largest capital deployment in technology history, conducted almost entirely outside the disclosure regime that public markets impose. That was always going to end — the capital requirements guaranteed it — and June 1, 2026 is the date it formally began ending. Within a year, the central questions of the AI economy will have audited answers: what the margins are, who the customers really are, how much of the revenue is circular, and what has actually been promised to whom. Bulls and bears alike should welcome it. Whatever the AI economy turns out to be, we are finally going to see its books.