Nvidia CEO Jensen Huang took the stage in Las Vegas yesterday to deliver a keynote that fundamentally redefined the company’s identity. Moving away from its heritage as a consumer graphics card manufacturer—notably announcing no new gaming GPUs for the first time in half a decade—Nvidia instead positioned itself as the architect of the “Physical AI” era. The presentation focused on a massive expansion into data center infrastructure, autonomous reasoning, and industrial robotics.
The Dawn of the Rubin Era
The centerpiece of the event was the official unveiling of the Vera Rubin platform, the successor to the record-breaking Blackwell architecture. Named after the pioneering American astronomer, the Rubin platform is built on an “extreme co-design” philosophy that integrates six distinct chips into a singular AI supercomputer. At its heart lies the Vera CPU, featuring 88 custom “Olympus” ARM cores, paired with the Rubin GPU, which delivers a staggering fivefold increase in AI inference performance over its predecessor. This hardware leap is designed specifically to handle the “explosive” demand for tokens required by next-generation reasoning models.
Beyond raw silicon, Nvidia introduced the Inference Context Memory Storage Platform, a new tier of AI-native storage powered by BlueField-4. This system acts as a high-speed “long-term memory” for AI agents, allowing server racks to reuse key-value cache data efficiently. By optimizing how data moves between the network and the processor, Nvidia claims the Rubin platform can reduce the cost of inference by up to 10x while requiring 4x fewer GPUs to train complex models.
From Reactive to Reasoning: The Alpamayo Breakthrough
In the realm of autonomous vehicles, Huang introduced Alpamayo R1, a family of open-source reasoning models that represent a shift from reactive driving to proactive logic. Unlike traditional self-driving stacks that rely on rigid rules, Alpamayo utilizes “Chain of Causation” reasoning to navigate complex “long-tail” scenarios—the rare and unpredictable road events that often baffle current systems. To foster trust in this technology, Nvidia is open-sourcing not just the model weights, but also the datasets used to train them. This technology is already hitting the pavement, as the 2026 Mercedes-Benz CLA was showcased as the first passenger vehicle to integrate the full Alpamayo-powered Nvidia DRIVE stack.
Physical AI and the Industrial Metaverse
The keynote also signaled a “ChatGPT moment” for robotics. Through the Cosmos and GR00T frameworks, Nvidia demonstrated how AI is gaining a physical form. A highlight of the show featured autonomous droids interacting with Huang on stage, having been trained entirely within the Omniverse—a physically accurate digital twin environment. This “Physical AI” isn’t just for robots; it is being applied to entire manufacturing plants. In an expanded partnership with Siemens, Nvidia showed how factories are now being designed as giant, integrated robots, where every trajectory and production line is simulated and optimized before physical construction begins.
A New Focus for the Desktop
While GeForce gamers didn’t see a new flagship card, Nvidia did not ignore the desktop entirely. The company introduced the DGX Spark, a deskside supercomputer designed to bring data-center-level power to individual developers and creators. This system allows for the local execution of AI agents and massive 200-billion-parameter models without relying on the cloud. By moving “Agentic AI” to the local desk, Nvidia aims to turn the PC into a responsive, physical collaborator capable of reasoning through complex tasks in real-time.
