OpenAI CEO Sam Altman outlined a broad ambition for the company: build artificial general intelligence (AGI), work out how to make it safe, and find ways to deliver its benefits. Reaching that goal, he said, depends on research as well as far more computing power—and on discovering what current AI systems still lack.
Building a business around AI tools
OpenAI was seeking additional financial backing from Microsoft and other investors to meet the high costs of developing more capable models. Altman described the partnership with Microsoft CEO Satya Nadella as “working really well” and said he expected “to raise a lot more over time.”
Microsoft had invested $10 billion in OpenAI earlier in the year through a “multiyear” agreement that valued the company at $29 billion, according to people familiar with the talks. Altman said the company’s revenue growth had been good that year, but it remained unprofitable because of training costs. He did not provide financial details.
At an event on November 6 attended by Nadella, OpenAI introduced tools for developers and companies, along with upgrades to GPT-4. The announcements included customizable versions of ChatGPT and a GPT Store, a marketplace for apps built with the technology. OpenAI’s stated plan was to share revenue with popular GPT creators.
Altman framed these offerings as routes to a broader product: AI intelligence delivered through different channels. The company was also building its enterprise business, including with chief operating officer Brad Lightcap, who previously worked at Dropbox and startup accelerator Y Combinator.
Agents that can take action
OpenAI’s plans extended beyond chat responses. Altman said the company was working on more autonomous agents that could carry out tasks such as executing code, making payments, sending emails, and filing claims.
He expected these agents to become more powerful and handle increasingly complex actions. That could create business value across different categories, though the interview did not specify when these capabilities would arrive or how they would be deployed.
The distinction matters: an AI system that can perform a sequence of actions has a different role from one that only generates a response. OpenAI’s announced custom GPTs pointed toward task-specific applications, while Altman described a longer-term effort to make agents capable of doing more on a user’s behalf.
More models require more data and computing
OpenAI was working on GPT-5, but Altman gave no release timeline. He said training it would require more data, potentially combining publicly available internet datasets with proprietary information from companies. OpenAI had recently asked organizations to provide large datasets that were not already easily accessible online, with a particular interest in long-form writing and conversations in any format.
Altman cautioned that it was difficult to predict exactly what GPT-5 would be able to do before training began. He said forecasting new capabilities was important for safety, while acknowledging that the model’s specific advances over GPT-4 were not yet clear.
Training also depends on specialized hardware. OpenAI used Nvidia’s H100 chips, and Altman described a “brutal crunch” caused by supply shortages. He said the company had received some chips and expected more, adding that “next year looks already like it’s going to be better.” He also pointed to Google, Microsoft, AMD, and Intel as companies preparing rival AI chips.
The unresolved question behind AGI
Altman said large language models were one important component in the effort to build AGI, but that other elements would be needed. OpenAI had concentrated primarily on language models, which Altman argued use language as a way to compress information. He said the company viewed that as a promising path for developing intelligence.
Yet he identified a deeper problem: how to make systems produce fundamental advances in understanding and generate new knowledge. He compared today’s models to someone learning by reading textbooks and practicing problems, then argued that simply absorbing existing material would not be enough to make the kind of leap associated with inventing calculus.
That gap helps explain why OpenAI’s ambition involves more than scaling a chatbot. The company is pursuing better models, practical agents, expanded computing resources, and research into what might enable new knowledge. Altman described that missing idea as one of the biggest questions to work on.