Nvidia unveils Rubin AI chips at GTC, claims computing needs will surge 100x
Mar 19, 2025
Key Points
- Nvidia CEO Jensen Huang unveiled the Rubin chip line at GTC, claiming AI computing demand will surge 100x, driven primarily by reasoning models that require vastly more computation to process internal reasoning tokens before generating answers.
- Rubin chips deliver 3x the computing power of Nvidia's Ultra chips, with the Ultra version offering 14x Blackwell performance and expected to generate nearly $40 billion in revenue in its first year when shipping in late 2026.
- Nvidia announced partnerships with General Motors on autonomous driving and with Disney and Google DeepMind on Newton, a physics simulation platform, positioning robotics and AI agents as major growth vectors alongside chip sales.
Summary
Jensen Huang unveiled Nvidia's Rubin chip line at GTC, arguing that AI computing demand will surge 100x. The claim responds directly to investor skepticism sparked by DeepSeek, a Chinese startup that demonstrated efficient AI models requiring fewer Nvidia chips than previously assumed.
Huang's argument rests on reasoning models. These systems generate internal reasoning tokens before delivering answers, producing far more data than standard inference. Processing that data quickly requires higher chip density. Users will not tolerate waiting ten times longer for answers even if those answers rely on ten times more computation, so faster inference demands more chips.
The Rubin line delivers that density. The chips offer more than 3x the computing power of Nvidia's Ultra chips. The Ultra version of Rubin delivers 14x the performance of the current Blackwell generation. Rubin ships in the second half of 2026. Analysts project the line will generate nearly $40 billion in revenue in its first year and over $95 billion in its second year.
Huang also positioned AI agents as a major source of demand. These systems complete tasks like booking flights or making dinner reservations. An AI agent product focused on travel or hospitality bookings could reach significant revenue before facing competition from larger platforms.
Nvidia announced three partnerships beyond the chip line. General Motors will work with Nvidia on autonomous driving technology. Disney and Google DeepMind joined Nvidia on Newton, a physics simulation platform. Huang emphasized robotics, framing labor shortages as the driving force. Nvidia's experience building virtual worlds where objects behave realistically translates into physical AI capabilities for robots to perceive and reason about their environments.
Nvidia shares fell more than 3% after the presentation, slightly outpacing broader tech declines. The broader selloff reflects uncertainty around tariffs, competitive pressure from entrants like Intel, and the lingering effect of DeepSeek, despite Huang's confidence in the 100x demand thesis.