🧵Thread: NVIDIA GTC 2026 Keynote – A Song to End the Presentation 1/ Jensen Huang's GTC keynote this year didn't end with a PowerPoint presentation; instead, AI sang a rap to summarize the entire presentation. The lyrics contain 5 key points from the speech. One Thread explains them clearly 👇 2/ Inference is the Protagonist, Not Training "Once upon an AI time, training was the paradigm. Sure it taught the models how, but inference runs the whole world now." In the past, AI discussions focused on training parameters and consuming computing power. Now, Vera (Rubin architecture) directly offers a "35x cost difference," bringing down inference costs, which is the trigger for the true widespread adoption of AI. 3/ What Does 40 Million Computing Power Mean? "We multiplied compute by 40 million." This isn't a metaphor. From the earliest GPUs to the Blackwell generation, NVIDIA claims a 40 million-fold increase in computing power per unit of AI inference. Blackwell is jokingly referred to as "the inference king" in the lyrics—not the fastest training card, but the cheapest inference card. This shift in positioning is noteworthy. 4/ AI Factory: From "Buying Cards" to "Building a Factory" Huang Renxun's most significant product this time isn't the chip, but the entire AI Factory solution—DSX for data center switching and Dynamo for inference scheduling, aiming to directly monetize computing power and shorten the cycle from "buying hardware" to "running business." 5/ The Agent Era Has Truly Arrived, But with Guardrails OpenClaw is the biggest ecosystem move this time: an open-source multi-agent framework, working in conjunction with NeMo Guardrails for security filtering. The lyrics specifically mention "open source," a clear signal from NVIDIA to developers—no ecosystem lock-in, welcome to build on it. 6/ [Physical AI: From Simulation to the Real World] Alpamayo is NVIDIA's own robot foundation model. "GPT moment for the bots"—this is what Huang Renxun is saying; the ChatGPT moment for robots is only just beginning. The Isaac Lab + Cosmos world model aims to allow robots to practice sufficiently in a virtual environment before deploying them in the real world, using computing power to fill data gaps. 7/ This last lyric is worth remembering: "When data's missing, there's no dispute, we just generate more with compute." Insufficient data? Generate it with computing power. This is NVIDIA's answer to the data bottleneck in the entire AI industry, and it's the underlying logic behind their Cosmos world model. Five layers of a cake (energy → chip → infrastructure → model → application), Huang Renxun summed it all up in one song. NVIDIA GTC 2026 Keynote Full Video