On June 8, 2026, OpenAI confidentially submitted its S-1 to the SEC — one week after Anthropic filed its own . The financial press will frame this as a historic milestone: the moment artificial intelligence legitimised itself as an industry worthy of public markets.
They will be wrong.
The S-1 filing isn’t a milestone. It’s a tell. A loud, unambiguous signal that the most capitalised startup in history has exhausted its options in private markets and needs public liquidity before the arithmetic catches up with the story.
Why Now?
Two companies that collectively consume 89% of all AI startup revenue, per The Information , are racing to public markets within weeks of each other. They’re not racing because business is booming. They’re racing because existing investors have stopped writing cheques large enough to cover the burn.
OpenAI spent more than $50 billion on compute in 2026 alone, according to Bloomberg . Its Q1 2026 non-GAAP operating margin was negative 122% — for every dollar of revenue, it lost an additional $1.22. Its March 2026 funding round closed at $122 billion — the largest private raise in history — and it will still prove insufficient. The company projects it will burn $852 billion through the end of 2030.
These are not the numbers of a company achieving escape velocity. These are the numbers of a company that has exhausted every private dollar available and now needs a new class of buyer who has not yet read the prospectus.
The Subscriptions That Aren’t There
The most damning data point hides in plain sight. According to The Information , OpenAI projects that its $20-per-month ChatGPT Plus subscriptions will decline from 44 million in 2025 to just 9 million in 2026. An 80% collapse in the company’s flagship consumer product.
The internal projections attempt to paper over this with ChatGPT Go — an ad-supported tier at $5 to $8 per month. They project Go subscribers will surge from 3 million to 112 million — a 3,600% increase year-over-year. That would be the single largest user acquisition campaign in history, aimed at the lowest-value customers, who will each cost the company more to serve than they pay.
The company is projecting a massive migration from profitable subscribers to a loss-leading alternative — and calling it growth. The S-1 will formalise this as a risk factor, and institutional investors will have to decide whether a $5-per-month ad-supported product can fill a $50-billion-per-year compute hole.
The Enterprise Contagion
Token-based billing — adopted by both OpenAI and Anthropic in early 2026 — has triggered a wave of cost-control measures across the corporate world. Uber now caps AI spend at $1,500 per user per month after burning through its entire annual token budget in a single quarter. T-Mobile limits spend at $2,000 per user per month . Brex restricts engineers to $500 per week and non-engineers to a token $5 per week. These are not companies expressing confidence in AI’s return on investment. They are installing emergency brakes.
A KPMG survey cited by the Wall Street Journal found that only 26% of companies have a comprehensive view of their AI costs. Half have partial visibility. The rest only know what they spent after the bill arrives. This is not sustainable revenue. It is revenue built on opacity and the fear of being left behind.
The S-1 will be forced to acknowledge concentration risk in its enterprise segment — and the reality that many of these customers are already pulling back.
The Compute Trap
OpenAI has secured more than $770 billion in compute commitments across Microsoft, Amazon, CoreWeave, Cerebras, and Oracle, per The Information . These are not optional. They are contractual obligations to pay for data-centre capacity regardless of whether demand materialises.
Here is the contradiction the S-1 will expose: OpenAI’s cost structure is fixed and growing, while its revenue base is shifting from high-value subscribers to low-value ad-supported users whose enterprise counterparts are actively capping their spending.
To meet its commitments, OpenAI needs to roughly double its revenue every year through 2029. That means going from roughly $30 billion in projected 2026 revenue to something approaching $184 billion by 2029. This requires AI demand to grow at a pace no technology in history has achieved — at a moment when the earliest enterprise adopters are already hitting the brakes.
Anthropic faces the same arithmetic with different numbers: $330 billion in compute commitments requiring $174 billion in annual revenue by 2029. Between them, these two companies need the combined revenue of six Fortune 500 companies materialising within three years — from a product category whose primary consumer subscription is collapsing.
What the S-1 Will Actually Reveal
The S-1 will not say “we are running out of money.” It will be more careful. It will disclose the performance obligations to compute providers. It will break out consumer subscription revenue by tier and reveal the mix shift. It will show customer concentration in the enterprise segment and the churn rates that token billing has triggered.
It will also reveal executive compensation structures, the terms of the for-profit conversion, and the governance arrangements that secured board seats for Microsoft, SoftBank, and others. These are not trivial disclosures — but they are not the story.
The story is what happens when you remove the ARR smoke and mirrors. OpenAI has reported its financials through carefully timed leaks of annualised recurring revenue — a metric that takes a single month’s revenue and multiplies by twelve, ignoring seasonality and churn. The S-1 will replace this with actual GAAP financials. The contrast will be brutal.
This Is Not Legitimisation
When WeWork filed its S-1 in 2019, the narrative was that flexible office space had arrived. When the books opened, the market discovered a real estate company spending $1.90 for every $1.00 of revenue. OpenAI is a compute company spending $2.22 for every $1.00 of revenue — and its primary product line is shrinking.
The IPO is not the moment when AI becomes real. It is the moment when the music stops and the market has to decide who is left holding the bag.
An IPO for a company burning $852 billion through 2030 is not a validation of the AI industry. It is a transfer of risk from private investors who have seen enough to public investors who have not yet read the prospectus.
The Last Buyer
The IPO window for AI companies is open for another 12 to 18 months before the books force a repricing. OpenAI and Anthropic are rushing through it because they know what everyone in private markets already suspects: there are not enough buyers at current prices.
The S-1 filing is the biggest tell yet — not for what it says, but for when it happened. It happened because the private capital that sustained a $50-billion-per-year compute burn rate has started asking harder questions. It happened because ChatGPT’s growth stalled and enterprise customers are installing spend controls. It happened because the alternative was running out of money in private markets, and the only way to keep the machine running is to find a new class of buyer who has not done the arithmetic yet.
The S-1 will make that arithmetic public. And when it does, the bag holders will not be the VCs who already unloaded their secondary shares. They will be the public-market retail investors who bought the CNBC headline without reading the prospectus.
Further Reading
- OpenAI, “Confidential submission of draft S-1 to the SEC” — OpenAI’s official S-1 announcement, June 8, 2026.
- Anthropic, “Anthropic confidentially submits draft S-1 to the SEC” — Anthropic’s official announcement, filed June 1, 2026.
- OpenAI, “OpenAI raises $122 billion to accelerate the next phase of AI” — Details of the March 2026 funding round, the largest private raise in history.
- Bloomberg, “OpenAI to Spend $50 Billion on Computing in 2026” — The compute cost figure that drives the whole arithmetic.
- The Information, “OpenAI and Anthropic Make Up 89% of AI Startup Revenues” — Market concentration data showing how little revenue exists beyond the two IPO-bound labs.
- Yahoo Finance / Investing.com, “Uber caps monthly employee AI spending at $1,500 per tool” — Enterprise cost-control in action at one of the earliest AI adopters.
- Wall Street Journal, “The Metric CFOs Struggle to Track: AI Usage” — KPMG survey data on AI cost visibility gaps.
No comments yet