AI Decoded · 3 min read
AI Decoded · The Business of AI3 min · Nov 2026

OpenAI's $10 Billion Quarter — and Its Bigger Loss

OpenAI reportedly pulled in about $10 billion in a single quarter and still lost more than it earned. The number every AI builder should internalize isn't the revenue — it's the margin.

Written by northstar editorial·Updated 18 Nov 2026

The headline number is staggering in both directions. OpenAI reportedly booked about $10 billion in revenue in a single quarter — and reportedly lost even more than that running the business.

For anyone building a product on top of generative AI, the revenue isn't the interesting part. The margin is. Because the economics of the lab at the top of the stack eventually become the economics of everyone building below it.

What the numbers actually say

On November 10, 2026, Reuters — citing a Wall Street Journal review of investor documents — reported that OpenAI generated roughly $10 billion in third-quarter revenue, up about 50% year over year. In the same breath, it reported an operating loss of about $10.9 billion for the quarter.

Bloomberg framed the same results through margin: OpenAI's adjusted operating margin was roughly -109% in Q3, narrowing from about -150% in Q2. Read that carefully — the margin is improving in percentage terms while the absolute losses keep climbing, because the company is scaling faster than it's closing the gap. For context, OpenAI reportedly did about $5.7 billion in Q1 2026 revenue, so growth is real and fast.

One caveat worth keeping honest: OpenAI is private and doesn't file public results, so these figures come from leaked investor materials and reporting, and some prior-quarter numbers conflict between sources. The direction is well-corroborated; the decimals are estimates.

Newsletter

Like this decoded read? Get the next one in your inbox.

One sharp take on the latest in AI for PMs, founders and operators. Every few days. Free.

Free forever. Unsubscribe anytime. No spam.

Why a $10B business still bleeds

This is not a company that forgot to charge for its product. It's a company whose cost of goods — compute, inference, training runs, and scarce talent — scales with usage in a way traditional software's never did.

Classic SaaS has near-zero marginal cost: the hundredth-thousand user on a web app costs almost nothing more to serve. Frontier AI inverts that. Every query runs an expensive model on expensive hardware, so more usage means more cost, not just more revenue. Add multi-billion-dollar training runs and a talent market where individual researchers command eight-figure packages, and you get a business that looks like a software company on the top line and a capital-intensive industrial company on the bottom line.

The lesson that flows downhill

If you build on these models, this is your story too, one layer removed.

  • Token prices probably won't fall forever. Inference costs have dropped fast, but a supplier losing two dollars for every dollar earned has limited room to subsidize you indefinitely. Model your margin against the possibility that your input costs stay flat or rise.
  • Flat-rate pricing on metered cost is a trap. When your own cost of goods scales with usage, a flat subscription turns your heaviest users into your biggest losses — exactly the dynamic playing out at the top of the stack. Build usage-aware pricing before your power users teach you this lesson. (More in how to price an AI feature.)
  • "Growth will fix the margin" is a bet, not a plan. The lab is betting scale and efficiency eventually flip the sign. Maybe. But if a company with $10B a quarter and the best models in the world hasn't gotten there yet, a startup assuming it will grow into profitability on thin AI margins is making a far riskier version of the same bet.

The takeaway

OpenAI's quarter is the clearest public picture yet of generative AI's core tension: extraordinary demand, brutal unit economics. The revenue proves people will pay for this. The losses prove that serving it at the frontier is still a money-losing proposition, for now.

For builders, the signal isn't doom — it's discipline. Treat inference as a real cost of goods, price so usage can't sink you, and don't assume the subsidies at the top of the stack last forever. The companies that survive the AI era won't be the ones with the most usage. They'll be the ones who made the unit economics work before the music stopped.

Sources

Original reporting on the Q3 figures was by Reuters (citing a Wall Street Journal review of investor documents) and Bloomberg; OpenAI is private and did not comment. Accessible coverage and analysis:


Want more sharp takes on the business of AI? Browse the AI Decoded archive.

Frequently asked

(4)
How much did OpenAI make in Q3 2026?
According to Reuters, citing a Wall Street Journal review of investor documents, OpenAI generated roughly $10 billion in third-quarter 2026 revenue — about 50% higher than a year earlier. OpenAI is private and does not file public results, so these figures come from investor materials and reporting, not audited filings.
If revenue is so high, why is OpenAI losing money?
Because the cost of serving frontier AI — compute, inference, training, and talent — still outruns revenue. Reuters put the quarterly operating loss at about $10.9 billion, and Bloomberg reported an adjusted operating margin of roughly -109%, meaning OpenAI spent more than two dollars for every dollar it booked. The losses are narrowing in percentage terms but growing in absolute dollars as it scales.
What does this mean for startups building on OpenAI's models?
It's a reminder that the economics flow downhill. If the lab serving the model loses money at scale, the price you pay for tokens is unlikely to keep falling forever, and flat-rate pricing on top of per-request cost is dangerous. Build usage-aware pricing and don't assume today's API prices are the floor.
Are these numbers confirmed?
They're reported by reputable outlets (Reuters, Bloomberg, the Wall Street Journal) based on investor documents, not official filings, because OpenAI is private. Treat the direction as reliable and the exact figures as best-available estimates; some prior-quarter numbers conflict across sources.