OpenAI's confidential IPO filing

Strip away the AI mystique and OpenAI is asking Wall Street to price something the public markets have essentially never seen at this scale: a company with a roughly negative 122% non-GAAP operating margin, meaning it loses about $1.22 for every dollar of revenue it books ^[1]. That is not a young startup burning seed money — it sits on top of an estimated $20-25 billion in annualized 2025 revenue, yet is projected to lose roughly $14 billion in 2026, with cumulative losses possibly reaching $44 billion before any path to profitability around 2029 ^[1]. "Confidential filing" here simply means OpenAI handed the SEC a draft S-1 (the prospectus every company must file before selling shares) privately, so regulators can give feedback before the full financials become public ahead of the roadshow — it buys time, not solvency.

The deeper problem is the trajectory of the cost base. Gross margin reportedly fell from about 40% in 2024 to roughly 33% in 2025, while inference costs — the price of actually running the models to answer queries — are projected near $14.1 billion for 2026, up about 68% year over year ^[1]. OpenAI is burning roughly $25 billion in cash in 2026, and HSBC analysts estimate it may need more than $207 billion in additional capital by 2030 even under optimistic projections ^[1]. That is the real reason the IPO exists: private rounds, even $122 billion ones, cannot reliably refill a tank that empties this fast, and only public markets offer funding at the scale the compute bill demands ^[3]. The bet embedded in the filing is that GPU and inference costs fall faster than revenue growth slows — a race against the company's own electricity bill.

Most IPOs sell investors a claim on both the upside and the steering wheel. OpenAI's does not. The issuing entity is OpenAI Group PBC, a for-profit public benefit corporation restructured out of the original 2015 nonprofit, but control still flows through the OpenAI Foundation, the nonprofit that retains special voting and governance rights and can fire board members if a model is judged dangerous ^[4]. The Foundation holds a 26% stake worth roughly $130 billion; Microsoft holds about 27%; employees and other investors split the remaining ~47% ^[4]. So public shareholders would buy into the economics of a frontier-AI lab while the mission-bound nonprofit keeps the levers — a structure that is untested in public markets and is expected to draw scrutiny over board appointments and mission-versus-returns conflicts ^[4].

The governance questions extend to the top of the house. Sam Altman reportedly holds no confirmed equity stake — any grant would dilute existing holders — and regulators are said to be examining undisclosed financial ties between his personal investments and companies OpenAI considered backing ^[5]. For a tech-adjacent reader, the practical translation is blunt: in a normal IPO, if management destroys value, shareholders can eventually force change; here, a nonprofit board sitting above the for-profit can override returns in the name of safety, and the CEO driving the listing has no obvious equity skin in the game. That is a feature if you trust the mission and a material risk if you are pricing a stock.

OpenAI's filing is not a standalone event — it is the third frontier-AI S-1 in under two weeks, following Anthropic's confidential filing on June 1 at a $965 billion valuation and SpaceX's public filing on May 20 targeting roughly $1.8 trillion ^[2]. Bloomberg pegged the combined pipeline at about $3.6 trillion, comparable to the GDP of France, all chasing the same pool of investor appetite in a narrow window ^[2]. Wedbush's Dan Ives framed it as the moment "the floodgates for the IPO market are officially open" ^[6], and OpenAI itself seems to be moving with that urgency, announcing the filing rather than waiting because it expected a leak anyway ^[7].

Sequencing is the quiet strategic variable. The market read circulating in investor and community discussion is that whoever prices first sets the tone — and if Anthropic's debut were to disappoint, it could sour sentiment for everyone behind it, which gives OpenAI an incentive to get in early rather than wait. Perplexity's Aravind Srinivas, who is keeping his own company on a 2028 track, captured the contagion risk directly: "I certainly think there will be ripple effects if they don't go well" ^[3]. EMARKETER's Nate Elliott added that OpenAI "is filing to go public at a precarious moment," losing early consumer ground to Google and enterprise ground to Anthropic even as it needs the capital ^[3]. The competitive logic is therefore double-edged: filing early captures attention and demand first, but it also exposes OpenAI's unfinished profitability story to public scrutiny before rivals have to show theirs.

The bull-bear gap on this deal is unusually wide because the entire valuation rests on an assumption about the future, not the present. Bridgewater's Greg Jensen put the bear case in one line: the implied roughly 35x forward revenue multiple is "priced for a monopoly outcome that does not yet exist" ^[1]. Dan Niles, who says he prefers Anthropic, argues "I think OpenAI is stuck between the two" — boxed in by Google in consumer and Anthropic in enterprise — while still acknowledging the revenue ramp is historically fast ^[3]. Decision Tree Research's Gregory Allen reframed it as an execution question rather than a thesis question: "can they time the investments correctly...such that they don't run out of cash" ^[3]. The numbers do not settle the debate; they sharpen it, because a -122% operating margin can read as either a death spiral or the steepest investment curve in tech history depending on which growth assumption you plug in ^[1].

Community reception tracks that same split. On X, the conversation has been broadly event-driven and bullish on the spectacle — framed as potentially the largest tech offering in history — while a strong undercurrent fixates on OpenAI's Microsoft dependency and the nonprofit-to-PBC cap table, with one widely circulated thread highlighting that the S-1 names Microsoft itself as a business risk. Investing-focused Reddit communities lean far more skeptical, with a recurring framing of the IPO wave as a top signal and a wealth-transfer event, and one academic voice predicting a dot-com-style shakeout in which the underlying technology survives but most of the firms riding the hype do not. Cutting against the gloom, a practitioner claiming to lead AI at a consultancy offered a concrete counterweight — collapsing a 90-day insurance renewal to seven days and turning a "$17M process" into a "$6M process" — the kind of bottom-line proof point that, if it generalizes, is exactly what would justify the multiple the bears say is a fantasy.