Overview
- CEO Jensen Huang told GTC attendees Nvidia now expects at least $1 trillion in orders for its current AI systems through 2027, up from a prior $500 billion view through 2026.
- Nvidia introduced the Groq 3 LPX inference chip integrated with its Vera Rubin platform, which Huang said can accelerate inference workloads by up to 35 times.
- Production of the Groq-based hardware is slated for the second half of 2026 with Samsung as the manufacturer, according to the presentation.
- The roadmap pairs Vera Rubin’s rollout with a confirmed next architecture called Feynman planned for 2028, signaling continued two‑year cadence.
- Shares closed about 1.6% higher after the keynote as analysts noted Nvidia’s dominance even as hyperscalers and specialized chip startups develop rivals that could pressure share from 2027.