OpenAI's Revenue Grew 50% to $10 Billion — and Its Loss Still Widened

OpenAI generated about $10 billion in third-quarter revenue, up roughly 50%, while its operating loss widened to $10.9 billion from $10.1 billion, according to a Wall Street Journal review of investor documents reported by Reuters on November 10, 2026. The company did not immediately respond to Reuters' request for comment.

The numbers put OpenAI's commercial scale and its cost burden side by side. Sales are growing fast, but they haven't caught up with what it costs to run the business.

What do the third-quarter numbers actually show?

Three figures frame the quarter. Revenue reached approximately $10 billion, an increase of roughly 50%. The operating loss widened in dollar terms to $10.9 billion from $10.1 billion. And the adjusted operating margin improved to negative 109% from negative 150% in the prior quarter.

That last pair of facts is the part worth unpacking. A margin of negative 109% means OpenAI lost about $1.09 for every dollar of revenue it brought in — better than the $1.50 it lost per dollar the quarter before. But because revenue grew so much in absolute terms, even a better ratio produced a larger dollar loss.

So the picture is genuinely mixed. Efficiency is improving, and the hole is still getting deeper. Both things are true at once.

The takeaway in plain words: OpenAI is losing money more slowly per dollar earned, but it's still losing more money overall each quarter.

Why is compute the whole story here?

OpenAI has made compute the center of its account of the business, and the third-quarter numbers only make sense against that framing.

In a January 18, 2026 article, CFO Sarah Friar said annualized revenue exceeded $20 billion in 2025, alongside approximately 1.9 gigawatts of available compute. Friar described revenue as directly tracking available compute, and said additional capacity in earlier periods would have accelerated customer adoption and monetization.

In other words, OpenAI's leadership treats compute supply as the binding constraint on growth, not demand. The argument is that if the chips and data centers had been there sooner, the revenue would have followed sooner.

The business model spans consumer and workplace subscriptions, usage-based APIs, and a free tier supported by advertising and commerce. The loop the company describes runs in a circle: compute supports research and model capability, stronger models improve products and adoption, and adoption generates revenue to finance further compute.

There's a timing problem baked into that loop, and Friar acknowledged it directly. Compute commitments must be made years in advance, with capital deployed in tranches against demand signals, and infrastructure supplied through partnerships rather than ownership. Friar said capacity sometimes leads usage and usage sometimes leads capacity.

That asymmetry cuts both ways. The partnership-based approach gives OpenAI flexibility across providers and hardware. But it also makes the timing of investment consequential, because demand and capacity do not necessarily expand together.

The takeaway: OpenAI is betting that if it builds the compute, the revenue will come — and that bet requires spending billions before the revenue exists.

How big is the enterprise business?

OpenAI's April 8, 2026 enterprise update gave the clearest picture of where revenue beyond individual subscriptions comes from. Enterprise customers generated more than 40% of revenue and were on track to reach parity with consumer revenue by the end of 2026.

On the consumer side, ChatGPT had 900 million weekly users, according to the same update.

The enterprise customer list names both new logos and expanding accounts. New enterprise customers included Goldman Sachs, Phillips and State Farm. Usage was growing at existing customers including Cursor, DoorDash, Thermo Fisher and LY Corporation.

Those disclosures describe revenue opportunities across two fronts at once: mass consumer adoption and organizational deployments. That dual-track structure matters for the loss question, because enterprise contracts and API usage tend to be stickier and more predictable than consumer subscriptions alone.

The takeaway: nearly half of OpenAI's revenue already comes from businesses, not individual subscribers, and that share is heading toward half.

Where does the money for all this come from?

OpenAI's March 31, 2026 funding announcement said it had closed a round with $122 billion in committed capital at an $852 billion post-money valuation. That valuation places OpenAI among the most valuable private companies ever assessed, and the round's size reflects the capital intensity of the compute strategy Friar described.

In that announcement, OpenAI presented durable compute access as the basis for research advances, better products, wider access and lower delivery costs at scale.

The infrastructure portfolio at the time spanned cloud providers Microsoft, Oracle, AWS, CoreWeave and Google Cloud, plus chip and data-center partners. Nvidia remained the foundation of the training fleet and the majority of OpenAI's inference stack.

The breadth of that provider list is the partnership model in practice. Rather than owning its data centers outright, OpenAI spreads its compute commitments across five major cloud platforms and a network of chip and data-center partners — with Nvidia hardware underneath most of it.

The takeaway: $122 billion in committed capital and a five-provider cloud portfolio are what make the "build compute first" strategy financially possible.

Can the loss narrow from here?

The reported quarter leaves OpenAI with a clear scoreboard. Rapid sales growth, a substantial enterprise business and committed capital provide the basis for expansion. The widening loss shows that revenue has not yet caught up with operating costs.

The margin improvement from negative 150% to negative 109% is the number investors will watch most closely. It suggests the cost of serving each dollar of revenue is falling as scale grows. The dollar loss, moving from $10.1 billion to $10.9 billion, shows how much ground still needs covering.

The central issue for future disclosures is whether adoption can translate into a smaller operating loss while OpenAI maintains the compute capacity behind its products. If Friar is right that revenue tracks available compute, the roughly 1.9 gigawatts of capacity already in place — and whatever tranches of the $122 billion get deployed next — will set the ceiling on how fast that translation happens.