June 3, 2026
NVIDIA GTC Taipei 2026 — Every Major Announcement in 8 Minutes
NVIDIA held GTC Taipei 2026 and Jensen Huang laid out the next era of AI computing. Here's every major announcement, with the actual keynote footage.
The headlines
- Agentic AI in production — GitHub commits nearly tripled (300M → 1.5B, 2023→2026); $3T in developer salaries now generating $9T in output
- Tokens = revenue — the new benchmark is tokens per watt, not raw performance; data centers as token factories
- The agent stack — Model · Orchestration · Tools · Runtime, the same pattern across data centers, laptops, vehicles, robots, satellites
- Vera Rubin — multi-rack supercomputer, no cables/hoses/fans; assembly 2 hours → 5 minutes
- Vera CPU — 88 Olympus cores, 3.6 TB/s fabric, 40% lower latency than x86
- DSX AI Factory — full-stack blueprint, single sites hitting 1 GW
- Nemotron 3 Ultra — hybrid SSM + MoE, 5× faster inference, fully open
- RTX Spark — NVIDIA's first Windows-on-Arm chip (Blackwell RTX + 20-core Grace + 128GB unified)
- Physical AI — Cosmos 3, Alpamayo 2, Isaac GR00T (31-DoF humanoid reference platform)
Full keynote (official) → https://www.youtube.com/live/wSp6AiNIrsY
The whole keynote in 8 minutes in the video above. Subscribe for more AI news breakdowns.