In an all-hands meeting this month, OpenAI’s chief financial officer delivered a message the company rarely puts on the record: the growth has not slowed down. Sarah Friar told employees that annualized revenue for July had already surpassed the total revenue the company recorded in its entire second quarter, according to a person familiar with the meeting.
The disclosure, reported by CNBC, is the clearest signal yet of the pace at which OpenAI is converting the AI boom into sales. The company does not publish financial results, and its revenue figures circulate through leaks and investor disclosures. Friday’s statement, delivered internally rather than in a filing, was notable for its precision.
The comment also landed at a delicate moment. OpenAI is raising money at a valuation that has climbed into the hundreds of billions, and its investors are watching for evidence that growth justifies the numbers. The gap between OpenAI’s reported scale and its disclosed revenue has been a recurring subject of debate among analysts, and Friar’s remark was aimed squarely at that question.
The same week brought a second piece of personnel news that turned heads in the AI research community. Lilian Weng, who co-founded the startup Thinking Machines Lab after leaving OpenAI, has returned to the company, according to people familiar with the move. Weng had stepped back from Thinking Machines for health reasons, and her return to OpenAI comes just months after she left it.
Weng will work on recursive self-improvement, the practice of using AI models to help build the next generation of models, the people said. The assignment aligns with a direction OpenAI executives have signaled repeatedly: using its own frontier models to accelerate the development of their successors, an approach the company has described as a core part of its roadmap.
Weng is one of the most recognizable names in AI research. She spent years at OpenAI leading safety and alignment work, and her departure in 2024 was widely seen as a sign that the company’s priorities were shifting toward speed at the expense of safety. Her return, on a research track rather than a safety one, suggests she sees the company’s current direction as the place where the most consequential work will happen.
The return is a coup for OpenAI at a moment when talent has been flowing in both directions. OpenAI has lost a stream of senior researchers over the past two years, several of whom founded rival labs including Anthropic and Thinking Machines. Weng’s return suggests the gravitational pull of OpenAI’s scale and compute budgets still carries weight, even as competitors offer equity and independence.
For Thinking Machines, the departure is a loss. Weng was one of the lab’s most prominent figures, and her exit had already been preceded by her health-related step back. The lab has said it remains focused on its own research agenda, but losing a founding researcher twice in one year is a setback, people close to the company said.
The two developments together sketch the shape of the AI industry’s middle period. OpenAI’s revenue is growing fast enough that it can speak confidently about it internally, and its research bench is deep enough to reclaim talent it lost. Both are signs of a company consolidating its position at the top of the field.
The competitive backdrop is sharpening. Microsoft, OpenAI’s biggest backer, disclosed this week that it is becoming a more direct rival, and Anthropic continues to push into enterprise customers. OpenAI’s answer has been a combination of product speed, price cuts on its APIs and a research pipeline that it argues keeps it a step ahead.
Weng’s move has a backstory the AI world has followed closely. She led safety systems at OpenAI, then left in 2024 to help found Thinking Machines, a lab backed by prominent investors that aimed to build safer frontier models. Her departure from that lab, announced as a health-related step back, and her rapid return to OpenAI were the subject of intense speculation in research circles before the move was confirmed this week.
The research agenda she is joining is OpenAI’s most consequential. The company has said it intends to use its own models to generate training data and to help design the next generation of systems, an approach that compounds progress but also concentrates it inside one lab. Critics have warned that recursive self-improvement, pursued too fast, removes the human oversight that safety testing depends on. OpenAI’s leadership has argued that the alternative, slower development, carries its own risks.
None of this appears in OpenAI’s balance sheet, because OpenAI does not publish one. What Friar’s comment offers instead is a rare internal data point: the company’s revenue engine is still accelerating. Whether the pace holds, and whether the research team that rebuilt itself this week can keep the frontier moving, are the questions the market will ask when OpenAI’s next funding round comes around.
