“Even when the chips are free, it's not cheap enough. If you can't keep up with the state of the technology and the pace that we're running, even free isn't cheap enough.”
Jensen
0:18Jensen Huang argues that AI has crossed three distinct inflection points — generative, reasoning, and now agentic — with each transition multiplying compute demand by roughly 100x, putting us on a path to one-billion-times more inference than today. He makes the counterintuitive case that Nvidia's $50B data centers are actually cheaper than competitors' $30B alternatives because token throughput is 10x higher, and that a $500K engineer spending only $5K on tokens is as wasteful as a chip designer using pencil and paper. Beyond the hardware economics, Huang frames Open Claude as the blueprint for a new computing paradigm — a personal AI operating system with memory, scheduling, and tool use — and warns that America's greatest AI risk isn't misuse, but repeating the mistakes of solar, rare earth minerals, and telecom by ceding the entire tech stack to China. Any executive making infrastructure, talent, or AI adoption decisions right now needs to hear his specific frameworks for thinking about token spend per employee, inference factory ROI, and the geopolitics of the global AI race.
Discussion on NVIDIA's VERRubin architecture for diverse agentic workloads and how it expands NVIDIA's total addressable market (TAM) with components like Bluefield, Grock, and CPUs.
“We just really evolved from a GPU company to an AI factory company.”
This reframing of Nvidia's identity signals a massive strategic pivot — from selling chips to selling the entire infrastructure layer of the AI economy.
“We used to be a one rack company — we now added four more racks. Nvidia's TAM increased from whatever it was to probably something like 33%, 50% higher.”
Casually acknowledging a 33–50% expansion of total addressable market in a single product cycle is a staggering claim that reframes the chip wars entirely.
“When you're running an agent, you're accessing working memory, you're accessing long-term memory, you're using tools, you're really beating up on storage really hard. You have agents working with other agents.”
This vivid description of agentic AI workloads reveals why the entire data center stack — not just GPUs — needs to be reinvented, with major implications for infrastructure investment.
“Agents are the computer of the AI revolution — the operating system of modern industry.”
Framing AI agents as the new operating system is a bold, provocative analogy that implies whoever controls the agent layer controls the next computing platform — a thesis with enormous strategic consequences.
“25% of your data center space should be allocated to this Grok LPU-GPGPU combo — we should add Grok to about 25% of the Rubin systems in the data center.”
Publicly recommending that a quarter of next-generation data center capacity go to a competitor's chip is a striking, counterintuitive move that challenges the assumption Nvidia plays a zero-sum game.