AI Radar

Noteworthy AI conversations and writing, summarized from publisher notes.

Sam Altman on Sora, Energy, and Building an AI Empire

· The a16z Show

In a wide-ranging conversation with a16z's Ben Horowitz, OpenAI CEO Sam Altman framed the company as a vertically integrated stack spanning research, consumer products, and infrastructure, all aimed at building a 'personal AI' subscription that 'gets to know you.' He acknowledged having once opposed vertical integration but said the scale of what OpenAI is now building—described as 'the biggest data center in the history of humankind'—requires doing things in-house. He touted major compute partnerships with AMD, NVIDIA, and Oracle, calling the current moment 'time to go make a very aggressive infrastructure bet' driven by confidence in the research roadmap. Altman devoted significant attention to Sora, arguing that world models will be more important to AGI than people realize and that shipping the product helps 'society co-evolve' with capabilities like realistic video generation. He flagged Sora's social networking features and the surprise that 'more people want to create content' than the traditional 1% creator assumption, which is forcing OpenAI to rethink monetization. Looking ahead, he predicted that within two years models will be 'doing bigger chunks of science,' citing GPT-5's emerging ability to make novel math and physics discoveries as the equivalent of the Turing test moment for him personally. On policy and strategy, Altman argued against broad AI regulation, favoring scrutiny only of 'truly superhuman' models, and warned that US over-regulation while China advances would be 'extremely dangerous.' He expressed concern about universities adopting Chinese open-source models and praised the reception of GPT-OSS. On copyright, he predicted courts will treat training as fair use but develop new frameworks for generated content, noting some rights holders may actually want their characters used more. He also discussed the deep linkage between AI and energy, calling the prior Western rejection of nuclear 'incredibly dumb' and predicting solar-plus-storage and nuclear will dominate long-term energy supply. Altman closed with reflections on ChatGPT's 800 million weekly active users, the 'capability overhang' most of the world doesn't see, and his belief that AGI's arrival will be more continuous and less disruptive than feared. He said research consistently gets GPU priority over product features, addressed ChatGPT's 'obsequiousness' as a solvable personalization problem, and warned that breaking user trust with ads could be catastrophic. On future trillion-dollar companies, he admitted deep humility and urged founders to 'be deeply in the trenches' rather than pattern-match off OpenAI itself.

AI Radar wrote this summary from the publisher’s notes with AI, not from a transcript. How it works.