NVIDIA: OpenAI, Future of Compute, and the American Dream | BG2 w/ Bill Gurley and Brad Gerstner

What It Covers

Jensen Huang discusses NVIDIA's $100B OpenAI Stargate partnership, three AI scaling laws driving exponential compute demand, competitive moats in accelerated computing, China market strategy, and why token generation economics justify $5 trillion annual AI infrastructure spending by decade's end.

Key Questions Answered

• •Three Scaling Laws Economics: AI now operates on pretraining, post-training reinforcement learning, and inference-time reasoning scaling laws simultaneously. This creates compounding exponential compute demand as models both practice skills through post-training and think before answering, fundamentally changing infrastructure requirements from one-shot inference to continuous token generation requiring persistent AI factories.
• •Revenue Per Watt Metric: NVIDIA revenue correlates directly to power consumption as performance per watt becomes the critical metric. Alibaba plans 10x data center power increase by 2030 while token generation doubles every few months, making energy efficiency the primary competitive differentiator. Customers prioritize maximum revenue from fixed gigawatt capacity over chip purchase price discounts.
• •Annual Release Cycle Moat: NVIDIA ships six to seven co-designed chips annually across GPUs, CPUs, networking, and NVLink, delivering 30x performance improvement from Hopper to Blackwell in one year. This extreme co-design at data center scale requires starting hundreds of billions in wafer capacity years ahead, creating insurmountable barriers for competitors attempting single ASIC approaches.
• •OpenAI Hyperscaler Trajectory: OpenAI transitions from outsourcing to Microsoft Azure toward self-building AI infrastructure like Meta and X, establishing direct NVIDIA partnerships. Huang projects OpenAI becomes the next multitrillion-dollar hyperscaler company, making pre-IPO investment opportunities at current scale exceptionally valuable given 800 million weekly active users generating exponentially more tokens through reasoning.
• •China Market Strategic Imperative: Restricting NVIDIA sales to China created monopoly profits funding Huawei's three-year plan to surpass NVIDIA while eliminating 95% market share. Chinese AI researchers choosing US opportunities dropped from 90% to 10-15% in three years. Competing in China's market with half the world's AI engineers strengthens rather than weakens American AI leadership position.