The Information's TITV Nvidia’s New Rubin Chips & Self-Driving Tech, Amazon’s Tough Sell for AI, Energy Boom | Jan 6, 2025
51 snips
Jan 6, 2026 
Joining the discussion are Steve Jang, founder of Kindred Ventures with expertise in hardware and robotics, Rocket Drew, an AI and robotics reporter, Catherine Perloff, an Amazon reporter, and Anne Davis Vaughn, an AI infrastructure columnist. They delve into NVIDIA's Rubin chips and the need for data center revamps, the promising capabilities of the Alpameo-1 self-driving model, and internal skepticism about Amazon's Nova models. Additionally, they explore how energy industries are adapting amid the AI boom, leveraging opportunities in natural gas and nuclear sectors.
AI Snips
Chapters
Transcript
Episode notes
Rubin Requires Full Rack Redesign
- NVIDIA's Rubin is sold as nodes and racks, not drop-in GPUs, requiring data center and rack redesigns.
- It promises ~10x inference and 3–4x training efficiency but forces costly replacement cycles for operators.
Inference Will Overtake Training
- Inference workloads are expected to outnumber training workloads soon, shifting chip design priorities to tokens-per-second efficiency.
- Rubin and future inference accelerators aim to optimize inference watts and throughput for agentic workloads.
NVIDIA King-Makes New Markets
- NVIDIA is "king-making" sectors by providing hardware, datasets, and software that steer industry adoption toward its stack.
- Open-sourcing models and tools serves chip sales and ecosystem lock-in simultaneously.