AI Radar

Noteworthy AI conversations and writing, summarized from publisher notes.

NVIDIA's Jensen Huang on OpenAI Deal, Inference Scaling, and the Global AI Race

· BG2Pod with Brad Gerstner and Bill Gurley · BG2Pod

Jensen Huang sits down with Brad Gerstner and Bill Gurley to argue that the AI buildout is still in its earliest innings and that Wall Street is systematically underestimating NVIDIA. He frames the recent announcement of a $100B NVIDIA investment in OpenAI and a preferred-partner deal covering roughly 10 gigawatts of new capacity—potentially ~$400B in revenue—as additive to existing Azure, Oracle, and CoreWeave buildouts, and predicts OpenAI will become the next multi-trillion-dollar hyperscaler. Huang doubles down on the claim he made a year ago that inference compute will scale roughly a billion-fold, citing three compounding scaling laws: pre-training, post-training (reinforcement learning), and now thinking/inference, all reinforced by agentic, multimodal systems. On competition, Huang argues NVIDIA's moat is widening rather than narrowing. The company's annual release cadence (Hopper to Blackwell's ~30x jump, then Vera Rubin, Ultra, and Feynman) is enabled by extreme co-design across six to seven chips per system, plus networking, switching, and software, making it nearly impossible for ASIC entrants to keep up. Even if a competitor gave chips away for free, Huang says, the power-and-shell bottleneck and the 30x performance gap make NVIDIA's total cost of intelligence better. He also walks through sovereign AI as the next major wave, criticizes U.S. export policy toward China as a strategic own-goal that handed Huawei a monopoly, and urges the administration to preserve the immigration pipeline that fuels the American AI brand. The conversation closes with a discussion of social contract issues—Invest America accounts for newborns, re-industrialization, and robotics/digital twins—culminating in Huang's prediction that NVIDIA itself will be the first $10 trillion company.

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