OpenAI 'Lost' $20B in Revenue It Never Lost: The Accounting Lesson That Sank AI Stocks | inicIA
Direct answer
On October 8, 2026, the Financial Times reported that OpenAI told investors its annualized revenue was approaching $50 billion at the end of September —about $20 billion less than the ~$70 billion that had circulated among investors in prior weeks. The market reacted with a broad selloff: Nvidia fell nearly 3%, Oracle almost 6%, Arm more than 6% and CoreWeave nearly 8% in the session. But no $20 billion was lost: the gap is accounting. OpenAI and Anthropic recognize sales made through cloud partners differently (the ASC 606 principal-vs-agent framework): Anthropic counts the gross of its partners' sales (AWS, Google Cloud); OpenAI counts only its share. Investors tried to "normalize" OpenAI's figures to Anthropic's method and the adjustment produced a compounding ~$20 billion arithmetic error. The underlying business keeps growing: OpenAI reported 77% run-rate growth in the third quarter and its enterprise revenue more than doubled since July. It projects ~$35 billion in full-year 2026 revenue and expects to reach or exceed $70 billion annualized by year-end.
The numbers, with context
- End-2025: the run rate was around $20B.
- August 2026: it exceeded $40B.
- End-Sep-2026: ~$50B annualized (the FT figure).
- Projection: ~$35B in total 2026 revenue; $70B+ annualized by year-end.
- The capital context: OpenAI is reportedly discussing raising at least $30B at a ~$1.4 trillion pre-money valuation, and its IPO moved to early 2027 (TechCrunch). In March alone it raised $122B.
Why the Anthropic comparison was a trap
TechCrunch summarized it precisely: OpenAI and Anthropic calculate annualized revenue differently —Anthropic includes its cloud partners' sales; OpenAI doesn't. Comparing the two headlines unadjusted is apples to oranges. When investors adjusted OpenAI's figures to "match" Anthropic's method, they invented $20 billion that didn't exist. The stock selloff didn't measure the business's health; it measured the surprise of those betting with the wrong yardstick.
The lesson for your business (the one that actually touches you)
You don't buy lab stocks; you buy solutions from AI vendors. And vendors also choose their yardstick. Three questions before signing:
- How do they measure what they sell you? "99% accuracy," "40% savings," "ROI in 3 months": ask against which benchmark, with what data, and who audited it.
- Vendor growth or vendor profitability? OpenAI's numbers show real demand (77% in a quarter), but also a brutal spending race. A vendor that grows while burning capital can change prices, terms or priorities —negotiate exit clauses.
- What happens if your vendor wobbles? The October 8 selloff was a reminder that the whole ecosystem hangs on two or three labs. Diversify critical vendors and avoid architectures where a single model is irreplaceable.
Related reading
- Anthropic showed its numbers ahead of the $2T IPO: $16.8B in Q3 and adjusted-profitable
- AI ROI: the real 2026 numbers
- SMB AI spending soared and Anthropic took OpenAI's throne: your advantage as a buyer
Frequently asked questions
Did OpenAI lose $20 billion?
No. The Financial Times reported on Oct 8, 2026 that OpenAI told investors annualized revenue was ~$50B at the end of September, versus ~$70B estimates. The difference comes from different accounting methods between OpenAI and Anthropic for cloud-partner sales, not lost contracts.
Why did AI stocks fall?
Nvidia (~-3%), Oracle (~-6%), Arm (>-6%) and CoreWeave (~-8%) fell on October 8 because the figure surprised investors using the $70B estimate to value AI infrastructure spending.
Is OpenAI's business in trouble?
The data the company itself shared says otherwise: 77% run-rate growth in Q3 and enterprise revenue more than doubled since July. It projects ~$35B in 2026 and $70B+ annualized by year-end.
What should I demand from my AI vendor?
That they show how they measure their promises (benchmark, data, audit), exit clauses if they change prices or priorities, and an architecture that doesn't depend on a single irreplaceable model.
Take the next step
Vendor numbers are read with the fine print included. If you're going to invest in AI, do it with a partner that explains the yardstick it measures with. Talk to inicIA on WhatsApp: https://wa.me/17866120930
Chat on WhatsAppBy inicIA
