OpenAI prepares its confidential $730 billion IPO — while Anthropic surpasses it in revenue

OpenAI prepares its confidential $730 billion IPO — while Anthropic overtakes it in revenue

🔎 The biggest transparency test of the AI era

OpenAI filed a confidential S-1 with the SEC in June 2026. The goal: an IPO as early as September, led by Goldman Sachs and Morgan Stanley, with a private valuation of $730 billion according to the New York Times (May 2026), and estimates going up to $1 trillion according to some sources.

Except that between the confidential filing and the publication of the S-1, the ground has shifted under OpenAI's feet. Anthropic passed the $47 billion annualized revenue run-rate mark in May 2026, surpassing OpenAI by roughly 35% according to Mashable. The IPO that was supposed to be a triumph is looking more and more like an open-book exam — with a grader named Anthropic that has already turned in its paper.

Why now? Because OpenAI can no longer remain private indefinitely. Employees want liquidity. Serious investors (SoftBank, Microsoft, Nvidia) want an exit. And above all, the maximum valuation window could close if Anthropic continues to take enterprise market share.

The essentials

• OpenAI filed a confidential S-1 with the SEC in June 2026, working with Goldman Sachs and Morgan Stanley for a stock market debut targeted in September 2026.
• The private valuation fluctuates between 730 billion (NYT, May 2026) and 850 billion to 1,000 billion depending on the sources, which would make it the largest tech IPO in history.
• Anthropic reached a $47 billion revenue run-rate in May 2026, compared to an estimated 24-25 billion for OpenAI at the same time — a complete reversal compared to the end of 2024.
• The publicly released S-1 will reveal OpenAI's actual unit economics for the first time: training cost, inference cost per request, churn rate, and net profitability.
• The Apple lawsuit (400+ poached employees) and the talent drain to Anthropic complicate OpenAI's narrative of dominance.

Recommended tools

IPO Figures: 730 Billion or 1,000 Billion?

OpenAI's pre-IPO valuation depends on who you listen to. The New York Times (May 2026) places the private valuation at $730 billion after a funding round this year — excluding the IPO premium. TechTimes and 247WallSt talk about $850 billion. em360tech cites sources mentioning up to $1,000 billion.

The reality: no one knows the exact figure before the S-1 is published. The confidential filing allows OpenAI to submit its financials to the SEC for review without making them public. This is a standard process for IPOs of this magnitude — Meta did it in 2012, Alibaba in 2014.

What is certain is that 730 billion already represents an absolute record. For comparison, Alibaba's IPO in 2014 raised $25 billion at a valuation of $231 billion. Meta's IPO in 2012 reached $104 billion. We are in a completely different order of magnitude.

The question is not "Is OpenAI worth 730 billion?" but rather "Can OpenAI justify 730 billion with the margins it displays?" And this is where the Anthropic problem becomes central.

Anthropic surpasses OpenAI: the numbers that change everything

In April 2026, Anthropic dépasse OpenAI en revenus avec 30 milliards de dollars de run rate, according to India Today and Trending Topics. Two months later, in May 2026, Anthropic's run rate reaches 47 billion dollars — about 35% more than OpenAI, according to Mashable and Moneycontrol.

The trajectory is staggering. At the end of 2024, Anthropic was running at about 1 billion dollars annualized. In February 2026, the company reached 14 billion. In April, 30 billion. In May, 47 billion. An exponential growth curve that very few companies in history have displayed.

Anthropic not only surpassed OpenAI — it did so with an enterprise product that is expensive but sells better. Claude Opus 4.7 displays a 92% honesty rate according to Mashable, a figure that particularly resonates in regulated environments (finance, healthcare, legal).

What makes the situation uncomfortable for OpenAI: the IPO will force the publication of audited numbers. No longer possible to hide the reality behind the veil of being private. Investors will be able to compare OpenAI and Anthropic's margins line by line. And Anthropic, even though it is not publicly traded, benefits from increasing transparency through its funding rounds and leaked metrics.

Anthropic vise maintenant 900 milliards de dollars avec un round de 30 milliards — a figure that, if it materializes, would place the company almost on par with OpenAI's IPO valuation. The irony is complete: the rival that OpenAI was supposed to crush in the stock market could itself be preparing for an introduction that would cast a shadow over Sam Altman's.

Claude Code : the growth engine OpenAI doesn't have

A figure revealed by TechnologyChecker deserves special attention: Claude Code, Anthropic's agentic coding tool, grew from 1 to 2.5 billion dollars in annualized revenue between late 2025 and February 2026.

That's massive for a single product in a niche category. Claude Code doesn't target the general public but enterprise developers — exactly the segment where contracts are the most lucrative and churn is the lowest.

OpenAI has GPT-5.3 Codex, which scores 80 on the agentic leaderboard (compared to 94.3 for Claude Opus 4.7 Adaptive). But the raw score doesn't tell the whole story. What matters for enterprises is production reliability, the hallucination rate on critical code, and the ability to maintain context over codebases of 100,000+ lines.

Anthropic built its reputation exactly on these criteria. The result: a product that sells itself in CTO offices, without massive marketing. OpenAI probably spends 10 times more on user acquisition for a lower enterprise ARPU.

The Apple lawsuit: 400 employees and a crumbling narrative

One element that OpenAI's S-1 will need to address head-on: the Apple lawsuit for unfair hiring, involving more than 400 employees.

This is not a simple HR dispute. It's an indication that talent retention at OpenAI is under pressure. When 400 people leave (or are targeted for departure) to a single competitor, it signals a structural problem — whether in compensation, culture, or vision.

Anthropic attracts talent for several reasons. First, the promise of a clean IPO (without the governance chaos that OpenAI experienced in 2023-2024). Next, a "safety-first" culture that resonates with researchers who joined AI out of ethical conviction, not just financial FOMO. Finally, equity packages that could be worth more if Anthropic atteint 47 milliards de dollars de revenue run-rate et dépasse OpenAI — which is already the case.

The S-1 will have to list the risks related to retention. Institutional investors know how to read between the lines: a lawsuit of this magnitude is a red flag on human capital.

The unit economics of frontier AI: what the S-1 will reveal

This is the heart of the matter. Nobody outside of inner circles knows OpenAI's true margins. The S-1 will change that.

What we already know

Anthropic is running at a 47 billion run rate with a primarily enterprise offering. The large margins in enterprise come from multi-year contracts, low acquisition costs (the product sells itself), and premium pricing.

OpenAI has 200+ million monthly users on ChatGPT, but the majority are on the free tier. The freemium model works for distribution, but the inference costs on free queries are massive — especially with models like GPT-5.5 which is at the top of the agentic (98.2) and generalist (91) leaderboards.

What the S-1 will show

The training cost of GPT-5.5 and GPT-5.4 Pro. Nvidia compute spend. The average cost per query. The free → paid conversion ratio. The monthly churn rate. The gross margin by segment (consumer vs enterprise vs API).

These figures will define whether 730 billion is reasonable or delusional. A 60% gross margin with 50% annual growth? Plausible at this valuation. A 30% margin with high churn and exploding training costs? Investors will punish it.

Anthropic's advantage in this comparison is indirect but real: even without a public S-1, leaked revenue reports provide a baseline. If Anthropic is doing 47 billion with a smaller team and less marketing spend, investors will wonder why OpenAI, with 5 times more media noise, generates less revenue.

Model comparisons: who is really leading?

The June 2025 LLM rankings show a nuanced picture. OpenAI dominates the pure agentic segment with GPT-5.5 at 98.2, followed by Gemini 3 Pro Deep Think (95.4) and Claude Opus 4.7 Adaptive (94.3).

In the generalist category, it's tighter: Gemini 3.1 Pro leads at 92, followed by GPT-5.5 and GPT-5.4 Pro at 91, then Claude Opus 4.7 at 90. xAI's Grok 4.1 also comes in at 90.

But here is the problem for OpenAI: leading on a benchmark does not automatically translate into enterprise revenue. Claude Opus 4.6 scores 87 in generalist and 84.7 in agentic — lower than GPT-5.5. Yet, Anthropic generates more revenue. Why?

The answer lies in the product, not the benchmark. Enterprises don't choose a model because it scores 4 points higher on a synthetic test. They choose it because the model doesn't hallucinate on regulatory reports, because the API is stable, because enterprise support responds in under an hour.

Anthropic understood this before OpenAI. And it shows in the numbers.

The IPO timeline: is September 2026 realistic?

TechTimes reports that OpenAI is targeting a market debut as early as September 2026. The standard process for a confidential S-1: filing → SEC review (4-8 weeks) → public S-1 → roadshow (2 weeks) → pricing → first day of trading.

The timeline is tight but feasible. Goldman Sachs and Morgan Stanley are the best banks for this type of deal. The confidential nature of the initial filing allows OpenAI to clean up its financials in private, negotiate with the SEC on disclosures, and prepare the narrative.

The timing risk: if markets are volatile between July and September (US midterm elections, geopolitical tensions, tech corrections), the IPO could be pushed back. A delay to Q4 2026 or even Q1 2027 is not out of the question — and each month of delay is another month for Anthropic to widen the revenue gap.

What the IPO changes for the AI industry

OpenAI's IPO is not an isolated event. It's an inflection point for the entire industry.

First, it creates a market price for frontier AI. All other players — Anthropic, xAI, DeepSeek, Google DeepMind — will be implicitly valued relative to OpenAI's multiple. If the market gives OpenAI 20x revenue, Anthropic at a $47 billion run rate is theoretically worth $940 billion. The $30 billion round that Anthropic is targeting makes sense in this context.

Second, it forces a transparency that will benefit the entire ecosystem. Series A investors in AI startups will finally be able to model their projections on real data, not estimates.

Third, it accelerates subsequent IPOs. If OpenAI succeeds in its listing, Anthropic will likely be public before the end of 2027. Musk's xAI will follow. The AI market is shifting from a "venture" phase to a "public markets" phase — with all the discipline that entails.

Risks the S-1 will have to list

An S-1 is not a marketing document. It is a legal document that must exhaustively list risks. Here is what will almost certainly appear:

Dependence on Nvidia. OpenAI spends billions on GPU compute. If Nvidia raises its prices or if the competition (AMD, Google TPU) fails to catch up, margins will suffer.

Competition from Anthropic. The S-1 will have to explicitly mention that a rival generates more revenue with fewer users. Failing to do so would be a material omission — and the SEC does not tolerate that.

Apple lawsuit. 400+ employees, impact on retention, potential damages.

Regulation. The EU AI Act, US executive orders, the risks of exponential legislation. A regulatory change could force OpenAI to revise its models, its APIs, or its pricing.

Revenue concentration. If 60% of enterprise revenue comes from 20 clients, that is a major concentration risk.

Every risk point is an opportunity for short-sellers and bearish analysts to build a negative narrative. OpenAI will need to have prepared answers for the roadshow.

❌ Common mistakes

Mistake 1: Confusing valuation and revenue

A 730 billion valuation does not mean that OpenAI generates 730 billion. With an estimated run rate of 24-25 billion (Trending Topics, April 2026), the multiple is around 30x. This is premium, but not irrational for a company with this kind of growth. The mistake is comparing valuation to revenue as if they were the same thing.

Mistake 2: Ignoring the difference between run rate and actual revenue

A 47 billion run rate for Anthropic does not mean that Anthropic has collected 47 billion. It is an annualized projection of the current month. The actual trailing twelve months revenue is necessarily lower. But the direction of the curve is what matters — and it is pointing upwards for Anthropic, laterally for OpenAI.

Mistake 3: Thinking that the IPO solves OpenAI's problems

The IPO brings liquidity and visibility. It does not solve the underlying issue: Anthropic monetizes enterprise better. The IPO could even increase the pressure by making quarterly comparisons inevitable.

❓ Frequently Asked Questions

What is the exact date of OpenAI's IPO?

OpenAI has not announced a firm date. The confidential filing took place in June 2026. TechTimes is targeting September 2026, but a delay to Q4 2026 or Q1 2027 remains possible depending on market conditions.

Will Anthropic also go public?

Nothing is official, but the $30 billion round Anthropic is targeting suggests a pre-IPO. An offering in 2027 is likely, especially since OpenAI's IPO will create a valuation benchmark.

Is OpenAI still the AI leader?

In terms of brand awareness and users, yes. ChatGPT remains the most well-known AI product in the world. In terms of revenue and enterprise monetization, Anthropic took the lead in the first half of 2026. The "leader" therefore depends on the metric chosen.

Which AI model to choose in 2026?

For enterprise code, Claude Opus 4.7 Adaptive is the most reliable choice according to honesty metrics and Claude Code revenues. For complex agentic tasks, GPT-5.5 dominates the benchmarks. The best choice depends on the use case, not the raw ranking.

✅ Conclusion

OpenAI's $730 billion IPO will be the most scrutinized financial event of the AI decade — but not for the reasons Sam Altman would have hoped. The S-1 will not tell the story of an undisputed leader. It will tell the story of a company that owns the biggest consumer AI brand but is being overtaken in revenue by a rival more disciplined on enterprise. September 2026 will not be a coronation. It will be an exam.