NVIDIA GTC 2026: Huang keynote and the Rubin roadmap for the next cycle
In one sentence At GTC 2026 NVIDIA confirms its annual cadence: details on Rubin (Blackwell's successor), new rack-scale configurations, updated software stack for training and inference.
GTC has become the single most important moment of the year to read where global AI infrastructure is heading. NVIDIA uses Jensen Huang's keynote to set the roadmap for the next 12-18 months: which GPUs are coming, in which configurations, what changes in the software.
2026 edition: focus on the Rubin family, successor to Blackwell (the 2024-2025 star). Rubin promises more memory, more bandwidth, more energy efficiency per generated token — the three vectors where everything is decided, given that electricity cost is now a strategic factor for hyperscalers.
For AI data center operators: details emerge on the calendar (who buys what when) and on rack-scale configurations (NVL successor). For developers: CUDA, TensorRT-LLM, NIM updates, and likely new primitives for agentic inference.
The economic effect is huge: NVIDIA remains the de facto supplier of the world's AI computing, and every GTC move shifts everyone else's investment calendar.
Companies
NVIDIA
Tools
Rubin, NVL, CUDA
Tags
Sources