NVIDIA Vera Rubin

NVIDIA Vera Rubin is the next-generation GPU architecture following Blackwell, announced by Jensen Huang at GTC 2026 and planned for 2026/2027 deployment. Named after astronomer Vera Rubin who provided key evidence for dark matter, the architecture promises another generational leap in AI inference and training performance. Key specifications revealed at GTC 2026: the 'Vera' ARM CPU as successor to the Grace processor with higher memory bandwidth and enhanced AI extensions, and the 'Rubin' GPU die as the primary compute engine. Together they form the Vera Rubin Superchip — analogous to Grace Blackwell. NVIDIA continues its annual roadmap cadence: Hopper (2022) → Blackwell (2024) → Blackwell Ultra (2025) → Vera Rubin (2026/2027). For the AI industry, Vera Rubin signals continuation of NVIDIA's hardware roadmap trend: every 1–2 years, inference performance per dollar doubles to triples. This drives LLM API prices falling 50–80% annually. Organizations with expensive inference workloads can expect dramatically lower costs once Vera Rubin-based cloud capacity is available. In the competitive landscape, NVIDIA competes with AMD's MI400, Google's Ironwood TPU (also announced GTC 2026), Intel Gaudi 4, and ASIC vendors like Groq, Cerebras, and Amazon Trainium 3.