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

Summary

• Jensen Huang frames the $100B OpenAI/Stargate partnership as helping OpenAI become a self-build hyperscaler — “I think that OpenAI is likely going to be the next multi-trillion dollar hyperscale company” — with 10 gigawatts implying roughly $400B of potential Nvidia revenue, all additive to the Azure, OCI (~5–7 GW buildout with OCI, OpenAI, and SoftBank; the projects are contracted) and CoreWeave buildouts. Brad’s only regret on the equity: “we were so poor, we didn’t invest enough… I should have given them all my money.”
• Wall Street consensus has Nvidia flatlining to 8% growth 2027–2030; Jensen calls the opportunity “much larger than the consensus.” His math: AI augments the ~$50T of world GDP tied to human intelligence; if $10T of token revenue at 50% gross margins needs $5T of annual AI-factory capex, versus a ~$400B market today, the TAM is a 4–5x from here. Glut probability is “extremely low” until all general-purpose computing converts to accelerated computing — “that’ll take a few years.”
• On the bear case that $100B of 2026 AI revenue must reach $1T by 2030, Jensen’s answer is “we’re already there” — hyperscaler revenue (TikTok, YouTube Shorts, Meta’s feeds, search) is already AI-driven and only recently moved to GPUs. Brad’s framing is that the trillion is near-certain before counting anything incremental.
• The moat is compounding, not eroding: annual cadence, six-seven co-designed chips a year, and “extreme co-design” delivered 30x Hopper→Blackwell in one year when Moore’s law gives nothing. The kicker: even if ASIC competitors priced their chips at zero, power-limited customers would still buy Nvidia — giving up 30x tokens-per-watt on scarce gigawatts is an opportunity cost no discount covers.
• He rebuts circular-revenue/round-tripping charges directly: OpenAI’s $400B buildout is funded by offtake, equity and debt that “smart investors and smart lenders” will price. Brad says “the revenue side of it has nothing to do with the investment side.” And Jensen plants a flag: “I think Nvidia will likely be the first 10 trillion dollar company” — because Nvidia is an AI infrastructure company, not a chip company.
• On China, Brad argues export bans handed Huawei monopoly profits, and Jensen is blunt: China is “nanoseconds behind us” — not two or three years. Guidance already includes no China, but “anybody who thinks the Chinese market is not important has their head deep in the sand,” and the China-hawk label is “a badge of shame.”
• The $100K H1B fee is “a great start… I just hope it’s not the end” — but the talent KPI is flashing red: a lab leader estimates top Chinese AI researchers wanting to come to the US fell from 90% to 10–15% in three years, which Jensen calls “the early indicators of a future problem.”

Deep dive

Not yet available upstream; scheduled sync will retry.