OpenAI files the biggest IPO ever attempted

The filing that ends the private-market era

OpenAI filed its confidential S-1 with the Securities and Exchange Commission on Friday, May 22, 2026, and four business days later the rest of the AI ecosystem is still cataloging what just changed. Per CNBC’s pre-filing scoop, the company submitted the draft registration statement through Goldman Sachs and Morgan Stanley with a target listing window that could open as early as September, and a valuation band that begins at $852 billion — OpenAI’s March funding-round mark — and extends past $1 trillion. Per Investing.com’s IPO test analysis, that range, if realized, would make this the largest technology IPO in financial history, eclipsing Alibaba’s 2014 listing by nearly five times. Sam Altman has spent four years insisting OpenAI’s mission is incompatible with the quarterly-earnings cycle. He filed anyway. The reason matters more than the filing itself.

The reason is compute. Per Proactive Investors’ coverage of the Wall Street IPO stampede, OpenAI has signed compute commitments totaling more than $1.4 trillion across Microsoft, Oracle, CoreWeave, Broadcom, AMD, Nvidia, and a constellation of sovereign-backed data-center vehicles. Private markets, even at $852 billion post-money, cannot underwrite that obligation. The capital calendar for 2026–2030 requires public liquidity, secondary offerings, and the optionality of a freely traded equity that can be used as currency for acquisitions and strategic compute deals. The IPO is not Altman’s choice in any meaningful sense; it is the only remaining instrument that matches the size of the bet OpenAI has already placed.

What lands hardest is the velocity. Per Sacra’s company file on OpenAI, annualized revenue moved from $2 billion in 2023 to roughly $25 billion in March 2026 — a 12.5× expansion in 36 months — and Per Searchlab’s 2026 statistics roundup, monthly revenue is now running near $2 billion with ChatGPT carrying more than 900 million weekly active users and over 9 million paying business seats. No precedent in modern capital markets matches the curve. The closest analogs — Amazon in 1997, Google in 2004, Meta in 2012 — went public with revenue between $147 million and $3.7 billion at IPO. OpenAI will print a top line on the cover of its red herring that is one full order of magnitude above the previous benchmark for a technology debut. The market has not absorbed a number like that before.

The risk profile is just as unprecedented. Per the Where’s Your Ed At analysis of OpenAI’s Q1 finances, OpenAI generated $5.7 billion in Q1 2026 against a non-GAAP operating margin of negative 122% — meaning the company lost roughly $1.22 for every dollar of revenue, and burned approximately $6.95 billion in a single quarter. The CFO blog posts will reframe that number with cost-allocation footnotes and Microsoft revenue-share adjustments, but the underlying truth is brutal: at the current trajectory, OpenAI is on pace to lose more than $30 billion in 2026 on its way to its own $30 billion top-line target. The IPO asks public investors to underwrite that gap on the conviction that the next two model generations will bend the curve. It is the largest bet on operating leverage that Wall Street has ever been asked to make.

Follow the money, find the moat

The IPO arithmetic is staggering, and once you write it down, the strategic logic comes into focus. Per the Davarion Group’s analysis of the S-1 filing, at $852 billion OpenAI prices at roughly 34× run-rate revenue; at $1 trillion, that multiple compresses to 40×. Compare those figures against Microsoft’s 13× forward-revenue multiple and Nvidia’s 25×, and the OpenAI listing is asking for a premium that has historically been reserved for companies with proven, recurring, and rapidly expanding margins. OpenAI has none of those characteristics yet. The premium is being charged on the option value of artificial general intelligence — the asserted possibility that the company will, within a decade, occupy a position in the global economy that is structurally larger than any current operating business. That is not an earnings story. It is a sovereign-bond-with-equity-upside story, and the buy-side knows it.

The cap table that emerges is its own indictment of how concentrated the AI economy has become. Per CryptoBriefing’s breakdown of the underwriting syndicate, Goldman Sachs and Morgan Stanley anchor the deal with JPMorgan in a co-lead position, and the post-IPO float will be a tiny percentage of total shares — likely under 10%, by industry comparables — leaving Microsoft, SoftBank, Nvidia, Amazon, and a constellation of sovereign funds in functional control of the equity. Per the Roborhythms breakdown of IPO mechanics, the structural quirk is that OpenAI’s for-profit subsidiary, OpenAI PBC, is what trades; the OpenAI Foundation retains majority control via super-voting shares. The combination — high valuation, low float, dual-class structure, deep insider concentration — produces a public security whose price will move violently on small order flow. Volatility is not a bug of this listing; it is a structural feature.

IPO debut	Listing year	Revenue at IPO (USD)	First-day market cap (USD)
Amazon	1997	$147M	$438M
Google	2004	$3.2B	$23B
Facebook	2012	$3.7B	$104B
Alibaba	2014	$8.6B	$231B
OpenAI (target)	2026	~$30B	$852B–$1T

The competitive context tightens the case further. Per VentureBeat’s reporting on Anthropic’s revenue surge, Anthropic crossed $30 billion in annualized revenue in April 2026, briefly overtaking OpenAI on the headline figure and adding urgency to OpenAI’s listing calendar. Per Anthropic’s own Series G announcement, the company closed $30 billion at a $380 billion post-money valuation in March, and Per proactive coverage of the listing pipeline intends to file its own IPO targeting October. The same syndicate banks that price OpenAI will be back at the desk four weeks later pricing its closest competitor. The deal arithmetic of going first is non-trivial: the lab that lists in September captures the institutional book before Anthropic competes for it, and dictates the valuation comparable that Anthropic will be measured against. The IPO is also a market-positioning weapon, not just a fundraising mechanism.

The deepest moat OpenAI is buying with public money is not algorithmic. It is physical. Per our coverage of OpenAI’s enterprise consulting subsidiary, the company has spent the last sixty days replumbing itself into a deployment business — DeployCo, the Tomoro acquisition, expanded forward-deployed engineering — and Per CIO’s analysis of OpenAI and Anthropic’s services push, the move pulls revenue closer to a Palantir-style services-and-software hybrid that carries higher gross margins and stickier customer relationships than pure API consumption. The IPO proceeds will not buy a new model. They will buy the data centers, the deployed engineers, and the sovereign partnerships that make the next model defensible. The hard part of frontier AI in 2026 is not training; it is distribution, and OpenAI is going public to buy distribution at industrial scale.

The final piece of the strategic case is talent currency. Per TechCrunch’s coverage of Andrej Karpathy joining Anthropic, the highest-leverage AI researchers are now mobile between labs in ways that would have been unthinkable two years ago, and the public-market premium attached to a listed equity is a recruiting weapon. An OpenAI common share that trades freely on the NYSE, with a transparent comparable price and a deep options market, is worth more to a candidate than an equivalent dollar of restricted private stock. The IPO closes the talent-currency gap with Google, Microsoft, and Meta and reopens the recruiting funnel that Anthropic, with its $380 billion private mark, has been hammering for two quarters. Listed equity is liquid; private paper is hope, and at frontier-AI talent levels, the difference is measured in eight-figure compensation deltas.

The ways this bet could blow up

The first failure mode is the one staring at every analyst in the syndicate: the unit economics do not work yet, and there is no public-comparable lab that has solved them. Per Where’s Your Ed At’s premium analysis of OpenAI’s burn, OpenAI’s compute costs sit near 40% of revenue while Google runs infrastructure at roughly 15% of revenue using vertically integrated TPUs. The 25-point margin gap is structural — Google owns its silicon, OpenAI rents it — and absent a credible path to in-house compute, OpenAI’s operating leverage curve is bounded above by Microsoft’s wholesale pricing decisions. Per futuresearch.ai’s revenue forecast, even with optimistic 2027 and 2028 revenue ramps, the company does not project GAAP profitability before 2029. Public investors are being asked to underwrite a three-year window during which the company will burn between $30 and $50 billion annually. The first quarterly print that disappoints will reprice the stock in a way private markets never had to confront.

The second failure mode is ChatGPT itself, and it is more alarming than the IPO marketing will admit. Per the Where’s Your Ed At analysis, ChatGPT’s weekly active user count hit 920 million in February but averaged 905 million across the quarter, meaning January or March was lower — the first time the consumer product’s growth has not been monotonic since launch. Per Backlinko’s 2026 ChatGPT statistics, the consumer subscription tier converts at roughly 5–6% of free users, a ratio that has not moved in a year. The S-1 will need to either explain the stall or explain why it does not matter, and both explanations weaken the bull case. The bull case requires consumer flywheel + enterprise expansion + API growth; if the first leg has plateaued, the multiple has to come down.

The third failure mode is regulatory, and the timing is precarious. Per our coverage of the Trump administration’s AI executive order, the federal AI oversight posture has been intentionally light in 2026, but Per CNBC’s reporting on the Center for AI Standards and Innovation, OpenAI signed a pre-release model-review agreement with the U.S. Commerce Department’s evaluation arm, agreeing to give the government access to frontier models before public deployment. That commitment is enforceable and will become a recurring item in the S-1 risk-factor section. Add the EU AI Act, the four active state-level AI bills moving in California and New York, and the still-pending Musk-related litigation Per our coverage of the May 2026 jury verdict, and OpenAI faces a regulatory exposure profile that no software IPO has ever priced. A single adverse model-evaluation finding before listing day would collapse the deal book.

The fourth failure mode is the public-market discipline that Altman has spent four years calling incompatible with the mission. Per Let’s Data Science’s analysis of the trillion-dollar filing, OpenAI’s own CFO has warned internally that the discipline of quarterly reporting will constrain the company’s ability to ship multi-year research bets that may not have any near-term revenue impact. The structural tension between a research lab’s optimal time horizon and a public company’s optimal time horizon is real and not solvable by communication. Google solved it by becoming so large and so profitable that DeepMind’s spend is a rounding error; OpenAI does not have that luxury. The first quarter where investors realize that GPT-6 is delayed because GPT-5.5’s margin push is consuming research capacity is the quarter the multiple compresses. The PBC structure mitigates this somewhat, but not enough — public-market gravity is real, and OpenAI is about to feel it for the first time.

The fifth failure mode is the macro one. Per Investing.com’s IPO test analysis, the IPO calendar for the second half of 2026 includes SpaceX (with a reported $400 billion target), Anthropic ($900 billion+), Stripe, Databricks, and a half-dozen smaller AI infrastructure plays. Combined new equity supply could exceed $135 billion in the second half of the year alone. That is more new tech equity than the public markets have absorbed in any rolling twelve-month window in history. If the macro tape is even moderately weak — a Fed pause, a credit event, a sovereign repricing — the syndicate will price OpenAI’s deal into a market that is already saturated with frontier-AI paper. The deal can still get done at a $700 billion valuation. That would be the worst possible signal to send to the rest of the listing pipeline, and it is not a small probability.

What operators should be doing this week

The IPO is not just a financial-markets event. It is a forcing function for every operator who builds on, sells to, or competes with OpenAI. Per the Enterprise DNA analysis of what the filing means, the company will spend the next four months executing the most aggressive public-relations campaign in tech history — and the substance of that campaign will be a sustained argument that OpenAI’s pricing power, enterprise expansion, and infrastructure leverage justify a trillion-dollar mark. Operators who depend on OpenAI need to assume API pricing will reset upward, enterprise contracts will tighten on terms favorable to OpenAI, and the company’s tolerance for low-margin customer segments will collapse. The cheap-API-access window is closing, and the listing event is what closes it.

The second operator implication is competitive substitution. Per Anthropic’s $30 billion revenue announcement coverage, Anthropic is now a credible drop-in competitor on price, capability, and enterprise compliance posture. Per TechCrunch’s coverage of Google’s I/O 2026 Gemini updates, Google has dropped its top-tier AI Ultra subscription from $250 to $100 per month and pushed Gemini 3.5 Flash to general availability at $1.50/$9 per million tokens — pricing that undercuts every comparable OpenAI tier. The operator who has architected only on OpenAI now carries a single-vendor risk that the public-market spotlight makes more dangerous. Multi-cloud-style multi-model abstraction is no longer a 2027 problem; it is a Q3 2026 problem.

The third implication is hiring. Per Axios’s coverage of the Karpathy hire, elite AI talent is moving between labs at a velocity that did not exist a year ago, and the IPO’s lockup structure will affect when and how OpenAI’s tenured staff become liquid. Standard 180-day lockups will create a tradable equity event in March 2027 that will shake free a meaningful percentage of the company’s mid-senior researchers. Recruiters at Anthropic, Google DeepMind, xAI, and Meta Superintelligence Labs are already building 2027 hiring plans around that window. Operators who need senior AI talent should be putting term sheets out now, not in March, when the post-lockup talent market will be the most efficient — and competitive — in the history of the field.

The fourth implication is procurement. Per the Searchlab statistics roundup, enterprise revenue now constitutes more than 40% of OpenAI’s mix and is expected to reach parity with consumer by year-end. That growth is being driven by multi-year commitments, and the listed-equity status will make OpenAI more aggressive on contract length, more inflexible on price escalators, and more demanding on data-residency and audit terms. Procurement teams that have been operating under expiring 2024 and 2025 OpenAI master agreements should lock in extensions now, before the company’s sales motion shifts into post-IPO mode. The leverage tilts decisively in the seller’s favor the day the bell rings.

The fifth and most consequential implication is governance. The OpenAI Foundation retains super-voting control via the PBC structure, which means public shareholders will own economics without governance. Per the Davarion Group’s S-1 analysis, the dual-class structure mirrors what Google and Meta did at IPO but pushes further, because the Foundation can theoretically override commercial decisions if it deems them inconsistent with the safety mission. For any operator building a long-term dependency on OpenAI, the practical consequence is that the company can pivot product direction, pricing, or model availability at any time without shareholder ratification. That is the price of doing business with a mission-driven public company, and it is materially different from doing business with a normal one. Plan accordingly.

The operator checklist that comes out of this week:

• Audit every OpenAI dependency above $100K/year and benchmark Anthropic, Gemini, and Llama alternatives at production volume before the September listing window opens.
• Renegotiate or extend long-term OpenAI contracts before pricing power tilts seller-favorable post-IPO; lock multi-year terms with capped escalators where possible.
• Build retention bonuses and counter-offers into Q3 2026 hiring plans for any role that touches frontier AI; assume Anthropic, xAI, and Google DeepMind will pull harder in late summer.
• Track the Center for AI Standards and Innovation pre-release evaluation timeline as a leading indicator of frontier-model availability and deployment friction in the U.S.
• Stress-test internal AI infrastructure assumptions against a scenario where OpenAI’s API pricing rises 20–30% in the twelve months following IPO; the unit economics need it.

In other news

Andrej Karpathy joins Anthropic’s pre-training team — OpenAI co-founder and former Tesla AI director Andrej Karpathy announced May 19 that he is joining Anthropic to launch a new team focused on using Claude itself to accelerate pre-training research. Per Axios’s coverage of the move, it is the highest-profile lab-to-lab move of 2026 and a significant talent signal in the run-up to Anthropic’s own IPO.

Anthropic withholds Mythos model under ASL-4 protocol — Anthropic confirmed its frontier preview model Mythos crossed the ASL-4 safety threshold after autonomously discovering zero-day vulnerabilities in production software at unprecedented scale. Per Axios’s reporting on the decision, the model is shipping only via Project Glasswing to roughly a dozen approved defensive-security partners including AWS, Apple, Microsoft, and JPMorgan Chase.

Google AI Ultra drops to $100/month with Spark agent — Google cut its top-tier consumer AI subscription from $250 to $100 per month at I/O 2026 and bundled in 20TB storage, YouTube Premium, and beta access to Gemini Spark, a 24/7 cloud-resident agent. Per TechCrunch’s I/O 2026 coverage, the 60% price cut is the steepest move on capable AI pricing any major lab has made this year.

Anthropic signs $1.25B/month compute deal with SpaceX Colossus — Anthropic expanded its compute partnership with SpaceX, agreeing to spend roughly $1.25 billion per month through 2029 for capacity on the Colossus supercomputing fabric. Per the Axios analysis of the AI compute reshuffle, the deal locks in compute headroom that supports Anthropic’s own listing plans and intensifies the GPU supply squeeze for second-tier labs.

Qualified Health raises $125M for clinical AI workflows — Healthcare AI startup Qualified Health closed a $125 million growth round to embed generative AI directly into health-system clinical and operational workflows. The round signals continued enterprise appetite for vertical AI plays even as horizontal foundation-model valuations stretch into trillion-dollar territory.