OpenAI's ads team reportedly reached $1B annual revenue within about a year
According to V. J. Raji, OpenAI's advertising team of about 50 people went from zero code to a $1 billion annual revenue run rate in roughly seven to eight months.
OpenAI's advertising team, which Vijay Raji describes as about 50 people, went from no code to a $1 billion annual revenue run rate in roughly seven to eight months, according to Raji's interview with Nate B. Jones. Raji, OpenAI's CTO of Applications, said the team had been working for about nine or ten months in total and reached that figure within that time. The numbers come from a conversation published as a video and a paid Substack interview. They are Raji's own account rather than published financials.
The more technical part of Raji's account concerns how the ad system is built. He described a pipeline with several stages: an ad indexer, a query rewriter, candidate generation, an auction, and a ranker. Results are then sent back to the client for display. The team did not build a separate machine-learning model for each stage, he said. Instead it used LLMs across the pipeline, and a stage-by-stage approach would have taken much longer.
Raji also made a claim about improvement over time. Because the stages are built on LLMs, the ad system gets better when the underlying models improve, which he described as the system getting smarter "for free." He presented this as a design principle: a system built to benefit from better models has been built correctly. He also said the team is small for its output. He offered the roughly 50 people figure as a guess, and framed it as evidence that the right tools and models make small teams capable.
What changed
OpenAI's advertising team reportedly reached a $1 billion annual revenue run rate within about seven to eight months.
Type: Business
Why it matters
Raji's account suggests a small team using general-purpose models across a multi-stage pipeline can move faster than one building separate models for each stage. Companies planning similar systems could reconsider how much custom ML they need.
What you can do
Map your existing multi-stage ML pipelines to see which stages a general-purpose model could handle before training separate models.
Original source
We have no link to the original source for this story yet.
Where we saw it
Nate B Jones
AI wrote this summary from the sources listed, and nobody reviewed it individually before publication. Check the original source before you rely on it. How it works