Within one week, both frontier labs slid an S-1 across the SEC's desk
Anthropic filed a confidential draft S-1 on June 1, OpenAI on June 8. The frontier race has reached its capital-markets phase, and the real motive is finding a funding pipe deeper than private rounds for an exploding compute capex curve.
Summary
Within seven days, the two strongest frontier AI labs handed the same document to the same agency. Anthropic announced on June 1 that it had confidentially submitted a draft S-1 to the U.S. Securities and Exchange Commission; OpenAI followed on June 8. Both notices used almost identical language: this is only a draft registration under Rule 135, the number of shares and the price have not been set, and whether an offering actually happens will depend on market conditions and other factors. Neither said it was about to IPO. Each one quietly pushed open the first door of the process.
Put the two events on one timeline and the signal is clear. The frontier race is shifting from “whose model is stronger” to “who can pull a steady stream of cash out of the public markets.” This is not coincidence; it is two companies driven to the same move by the same cost curve. The compute bill for training and inference has grown too large to keep filling one private round at a time, so both are laying pipe for the next stage of financing — and the public market is the only pool deep enough to absorb that size.
Worth saying upfront: OpenAI was blunt about it. “We recently submitted a confidential S-1. We expect it to leak so we’re just announcing it.” It even stressed that “we have not decided on timing yet; it may be a while because there are things we want to do that are likely easier as a private company.” That line is the part of this story most worth reading correctly. Filing an S-1 is not the same as an imminent IPO. It is holding the option to go public at any time, not nailing a listing date to the calendar.
The move
What actually happened is not complicated. Anthropic put out its notice on June 1, saying it had confidentially submitted a draft Form S-1 as a PBC, preserving the option of a proposed IPO of common stock. That came just days after its May 28 announcement of a $65 billion Series H, a round that pushed its post-money valuation to $965 billion; the same announcement noted run-rate revenue had crossed $47 billion that month. Per multiple financial outlets, Anthropic also told investors it expects its first profitable quarter in Q2 2026 — a card it played out loud, ahead of OpenAI.
OpenAI followed on June 8 with the same confidential draft filing and the same restrained wording. It currently carries a private valuation of roughly $852 billion; its CFO Sarah Friar reportedly said as early as April that the company was already “acting with the good hygiene of a public company.” One precondition that gets overlooked: the dismissal of Musk’s lawsuit this May cleared the last structural obstacle to OpenAI’s conversion into a PBC, the architecture that makes a public listing legally viable in the first place. So this move was not a sudden impulse. It is the natural next step after the org restructuring, financial discipline, and legal risk were checked off one by one.
The shared feature of both notices is that they leave room. The whole point of a confidential submission is that the financials — gross margin, losses, customer concentration — stay private, seen only by the regulator, until the company actually launches a roadshow and the prospectus goes public. Both companies leaned fully on that mechanism: signaling “we’re ready” to the market while refusing to flash a single number ahead of time.
The real motive
The stated reason is “give us the option to list.” The real motive is buried in the compute bill.
These two companies’ cost structures have parted ways with traditional software entirely. They are not burning server rent; they are burning industrial-grade scarce inputs — power, chips, data centers. While filing its S-1, Anthropic was simultaneously locking in a ten-year, $100-billion-plus AWS compute commitment and negotiating gigawatt-scale next-generation TPU capacity with Google and Broadcom. The peer comparison is starker still: per SpaceX’s own prospectus, xAI rents Colossus-cluster compute from it at roughly $1.25 billion a month (about $15 billion a year), under a contract running through May 2029 — a vivid measure of just how steep the frontier labs’ compute-spend curve has become. On OpenAI’s side, it is the enormous capex of the Stargate data centers. None of this is the kind of spend a private round or two can keep feeding. It is a curve that only climbs.
Even the deepest private market has a ceiling. Valuations of $965 billion and $852 billion are nearing the limit of what a single private investor or syndicate can absorb — Anthropic’s Series H pulled in nearly every large institutional investor on the planet. When a company’s financing need is large enough to mobilize the entire private market in one round, the public market becomes the only place left with headroom. The real motive behind filing an S-1 is to attach a thicker, more durable feeding tube to that compute capex curve than any private round can be.
There is a second motive: the narrative prize. “First publicly traded frontier-AI pure play” is a scarce label, and whoever grabs it first gets to define how the category is valued on the secondary market and which metrics matter. Anthropic filed a week early and floated “Q2 profitability” first; OpenAI answered with sheer scale — its CFO noting the valuation would rank among the fifteen largest in the S&P 500. This is positioning for who gets to be the “first AI stock.” The winner takes liquidity, pricing power, and a stretch of the market’s attention.
Who is threatened
The first group hit is the second-tier labs still surviving on private money alone. xAI and Mistral are still chasing private rounds one at a time, and when the two leaders pull the private market’s attention and capital toward themselves and open the deeper public channel first, the financing window for the chasers gets meaningfully squeezed. Capital is finite, and so is the institutional appetite for the compute narrative. The leaders listing first means they can keep refueling from public markets, while the followers grind through ever pricier, ever more crowded private rounds to fund the same compute curve. This is a structural crowding-out.
The second group hit is the industry’s “narrative valuation” itself — the leaders included. A confidential filing delays the reckoning; it does not cancel it. The moment an IPO is real, the prospectus must disclose gross margin, training and inference costs, revenue mix, and customer concentration. This is the first time the frontier-AI industry is forced to show a public set of financial cards. Until now, valuations have rested on growth stories and run-rate revenue. After that, cash flow, unit economics, and the absolute size of the losses get tested line by line. Anthropic floating “about to be profitable” first shows it knows this gate is unavoidable, so it put its best-looking number on the table early. For companies with weaker books, public disclosure will be a physical exam.
The third group is anyone treating “staying private” as a competitive edge. OpenAI’s line — “some things are easier as a private company” — reminds everyone there is a price for listing: disclosure duties, quarterly-earnings pressure, shareholder-suit risk, diluted patience for long-horizon research. But once both leaders decide to pay that price for the depth of public capital, the pitch “we stay private, so we stay free” loses some of its shine.
And for anyone who actually wants to read this business rather than ride the headlines, the signal to wait for is just as clear: not the act of filing, but the moment the prospectus flips from confidential to public — that is when verifiable gross margin and unit economics finally appear. Until then, run-rate revenue and the “about to be profitable” talking points are selective disclosure. The number to watch is not the valuation but whether they are willing to lay the cost side bare alongside it. Waiting for that public financial statement gets you closer to the truth than racing to read any single “submission.”
What to ignore
The first thing to ignore is reading “confidential submission” as “imminent IPO.” This is where financial headlines most easily lead you astray. OpenAI said it itself: timing undecided, possibly a while yet. A draft S-1 is the start of the process, not the end; it grants an option, not a schedule. Treating the act of filing as a valuation anchor to trade on misreads what the thing is.
The second thing to ignore is conjuring up the numbers that have not been disclosed. The entire point of a confidential filing is that the details stay private — the actual gross margin, losses, and customer structure are not visible to anyone right now. The valuations floating around ($965 billion, $852 billion) are private-round post-money figures and reporting, not the offering price in the S-1; the price and share count are still unset. Any calculation that treats those private valuations as an IPO anchor is telling a story with undisclosed data. When unsure of a figure, label it “reportedly” rather than state it as fact.
The third thing to ignore is over-reading the near-simultaneous filings as some conspiracy or quiet pact. They did not coordinate — Anthropic was a week ahead, and OpenAI went out of its way to explain it was announcing because “we expect it to leak.” What pushed them into the same move is the same compute-cost curve and the same private-market ceiling, not collusion. See that clearly and you won’t mistake an inevitability driven by the physics of capital for a scripted joint performance.