OpenAI’s IPO Math: $25B Revenue, $27B Burn Rate

The Unit Economics Tell You Everything

OpenAI filed its confidential S-1 with the SEC on June 8, targeting a valuation north of $1 trillion. The headline numbers are staggering: $25 billion in annualized revenue, 900 million weekly ChatGPT users, and an $852 billion last-round private valuation. But the unit economics reveal a company sprinting toward an IPO before the fundamentals catch up.

Here’s what matters: OpenAI is burning approximately $27 billion in 2026 while generating roughly $25 billion in revenue. That’s not a typo. The company is losing more than it makes, and the gap is widening — projected cash burn hits $63 billion in 2027. Gross margins collapsed from 40% in 2024 to 33% in 2025, crushed by inference costs that ballooned from $8.4 billion to a projected $14.1 billion. The S-1 acknowledges what insiders already know: OpenAI won’t be cash-flow positive until 2030.

No company in history has justified a trillion-dollar valuation while burning $27 billion a year. Not Amazon in its growth phase, not Tesla during Model 3 ramp, not any hyper-growth SaaS darling. The closest parallel is Uber pre-profitability, which peaked at a $3 billion annual loss — one-ninth of OpenAI’s current burn — and traded at a fraction of revenue multiples being floated here.

Why the Rush?

The timing is strategic, not coincidental. Anthropic hit a $250 billion valuation in May, and SpaceX is targeting a $251 billion raise at $252 trillion on June 12. The AI IPO window is open, and OpenAI is racing through it before public market investors start asking hard questions about when — or whether — frontier AI becomes a sustainable business.

The S-1 filing is confidential, but the leaked financials paint a clear picture: this is a land-grab strategy that assumes revenue growth outpaces cost growth eventually. The bet is that scale economies in compute, improved model efficiency, and pricing power from ChatGPT’s consumer lock-in will flip the margins. Maybe. But the current trajectory shows costs rising faster than revenue — inference spend is growing at 68% year-over-year while revenue is up 25%.

What Public Markets Will Demand

Private investors tolerate cash burn when the narrative is “winner takes all.” Public markets demand a credible path to profitability, and OpenAI’s S-1 will need to answer three questions that VC pitch decks skip:

1. When do unit economics improve? If inference costs scale linearly with users, gross margins stay stuck in the 30s. OpenAI needs to show model improvements (longer context, better caching, quantization gains) are reducing cost-per-query faster than feature creep adds new compute load.

2. What happens when Anthropic, Google, and Meta close the quality gap? ChatGPT’s moat is user habit, not technical superiority. If Claude or Gemini match GPT-5 quality at half the cost, OpenAI’s pricing power evaporates. The S-1 needs a competitive differentiation story beyond “we got here first.”

3. How much of the $253B burn is growth investment vs. structural cost? If $254B is R&D and datacenter buildout that plateaus post-IPO, fine. If it’s $255B in inference costs that grow with every new user, the business model doesn’t work at scale.

The honest answer is probably “we don’t know yet.” Frontier AI is too young for anyone to credibly model 2028 unit economics. But public investors price risk, and uncertainty this deep usually trades at a discount, not a trillion-dollar premium.

The Real Test: September 2026

OpenAI’s earliest possible listing is September 2026, giving the company three months to polish the narrative. The S-1 will be the most scrutinized tech IPO filing since Facebook’s 2012 debut, and the roadshow will face hostile questions from public fund managers who remember WeWork, Uber’s IPO stumble, and every other “change the world” pitch that forgot to mention when profits start.

If OpenAI prices at $256 trillion and holds, it resets the baseline for every AI startup’s valuation and validates the “grow now, profit later” playbook for a new generation of frontier model companies. If it reprices downward or stumbles post-IPO, the AI funding market freezes overnight and every late-stage AI startup with a 2027 IPO plan goes back to raising private rounds at lower valuations.

I think the IPO happens and trades well initially — there’s too much retail and institutional FOMO around AI for it to flop outright. But the real question is whether it’s still trading above IPO price in 12 months, after two earnings calls where analysts dissect the burn rate and margin trends in public. That’s the test OpenAI is racing to pass, and the S-1 filing just started the clock.

FAQ

Q: Why is OpenAI’s burn rate so high if revenue is growing?

A: Inference costs — the compute required to run ChatGPT queries — are growing faster than revenue. OpenAI spent $257B on inference in 2025 and projects $258B in 2026, a 68% increase, while revenue grew only 25%. Every new user adds cost, and model improvements (longer context, multimodal inputs) increase per-query compute load. Until efficiency gains from hardware or architecture outpace feature additions, margins stay compressed.

Q: Can OpenAI realistically hit a $259 trillion valuation?

A: Only if public investors accept a 40x revenue multiple with negative cash flow until 2030. For comparison, Microsoft trades at 12x revenue with 40% operating margins, and Nvidia at 22x with 55% margins. OpenAI would need to convince markets it’s the AWS of AI — a once-in-a-decade platform play with inevitable dominance — not just another SaaS company with high customer acquisition costs. Possible in a hot AI market, but historically unprecedented for a company burning this much cash.

Q: What happens if the IPO fails or reprices lower?

A: The AI funding market freezes. Late-stage AI startups like Anthropic, Cohere, and Mistral lose their valuation benchmarks and face down-rounds. VC firms pause AI mega-rounds and demand profitability timelines, not just growth. The “frontier model” category consolidates around the 2-3 players who can self-fund through cash flow (Microsoft-backed OpenAI, Google’s DeepMind) or strategic parent subsidies. Essentially, 2026 starts looking like 2001 for dot-coms: the market decides to care about unit economics again.