OpenAI Passes $10 Billion in Quarterly Revenue as Losses Narrow
OpenAI's quarterly revenue crossed $10 billion for the first time in the third quarter of 2026, according to Bloomberg, citing anonymous sources. The same report puts the company's annual revenue run rate, a figure extrapolated from recent sales, at $50 billion.
The earlier figures are lower. OpenAI recorded $6.7 billion in sales in the second quarter of 2026. Revenue grew about 18% between the first and second quarters, from a base of $5.7 billion in quarterly revenue. The third-quarter figure therefore sits well above the second-quarter total, and the step is larger than the sequential growth seen earlier in the year.
Profitability moved in the same direction but remains deeply negative. OpenAI's adjusted operating margin, which excludes costs the company chooses to leave out of the calculation, narrowed from -150% in the second quarter to -109% in the third. A margin of -109% means operating losses, on an adjusted basis, still exceeded revenue for the quarter. The cost base, which for a frontier model provider is dominated by training and inference compute (the processing that runs a trained model to answer requests), grew alongside its top line.
OpenAI says it has more than one billion users.
Anthropic's numbers were reported in the same coverage. Anthropic reported Q3 2026 revenue of $17 billion. Its annual revenue run rate rose by $10 billion within a month, to $81 billion as of the end of October 2026. Set against OpenAI's $50 billion run rate, the two companies are now measured on the same scale.
The broader context here is that both companies now report revenue in quarterly billions, not annual hundreds of millions. For enterprises that buy model access by the token (a unit of text the model reads or writes) or by the seat, the vendor market has a large paying base and is no longer a pre-revenue experiment. For infrastructure teams, demand for inference capacity is being priced into real contracts.
In my view, the margin line deserves more attention than the headline. Revenue growth at this pace is the easier half of the equation to report. Whether operating losses narrow as fast as revenue rises will determine how much pricing pressure, capacity planning, and model-release cadence the sector can sustain without outside capital. A margin that moved 41 points in a single quarter is a sign of real operating leverage, but one quarter does not establish a trend, and the figures are adjusted, so readers should check what those adjustments remove.
The practical reading for working engineers is simpler. A product with more than a billion claimed users and a $50 billion run rate is a platform that other software will be built on, and its unit economics will shape what gets cheaper, what gets rate-limited, and what gets deprecated. I would watch the next quarterly figures for the margin trajectory and for whether paying-user growth keeps pace with growth in compute spend. If it does, the result is a broader, better-funded base for the tools developers are already using.
