GET IN EARLY! I'm Investing In This Breakthrough AI Chip

Nvidia's Vera Rubin platform represents a fundamental departure from Blackwell, not merely a generational upgrade. AI workloads are shifting from short human-written prompts to autonomous agents like OpenClaw that call tools, browse websites, write code, and run for millions of tokens at a time — workloads that cost thousands of times more tokens than regular chat. Power-efficient, low-latency inference has become the new main cost driver for AI, which is why Rubin is engineered to produce as many useful tokens as possible per rack, per watt, and per dollar.

This architectural shift has huge implications for data center spending. While most analysts predict a slowdown, the host expects data center capital expenditure to accelerate because models are now continuously fine-tuned via reinforcement learning and agents demand orders of magnitude more compute. Rubin's design — seven new chips working in concert — is purpose-built to make these OpenClaw-style agents affordable to deploy at scale, setting the stage for a new wave of infrastructure investment that Wall Street isn't yet pricing in.

At GTC, the host saw an entire ecosystem of robots — for warehouses, hospitals, and retail — all being trained on the same Isaac and Cosmos world model stack. Agility's Digit humanoid is already running real shifts in GXO warehouses under a robotics-as-a-service model, handling logistics for brands like Nike, Amazon, and Apple. What most investors are missing is how fast this can ramp once even a handful of designs prove themselves in the field, because a capability learned in simulation for one warehouse can be tweaked and reused for the next 100 customers.

On the autonomous vehicle side, the host spent an hour in Nvidia's L2++ Mercedes driving through downtown San Francisco, navigating double-parked cars, construction zones, and erratic drivers with ease. Nvidia-powered robo-taxis using Drive Hyperion and Alpio are planned to roll out on Uber's network in cities like LA and San Francisco as soon as next year, expanding to 28 cities through 2028. Companies like BYD, Geely, Nissan, and Isuzu are developing their own level four vehicles for ride hailing and commercial fleets. When asked about the biggest near-term application for agentic systems, Jensen said autonomous vehicles — and noted that automotive is less than 1% of Nvidia's revenue today, just as CUDA once was.