NVIDIA's Vera Rubin Reveal
NVIDIA used GTC 2026 to pitch the Vera Rubin platform as the backbone for industrial-scale “AI factories” that orchestrate agentic AI with new CPUs, GPUs and networking — and analysts are sizing this as a structural shift for enterprise AI. The conference also framed a tokenized, end‑to‑end stack and projected massive hardware demand as businesses move from experimental models to operational AI at scale. (franksworld.com)
NVIDIA said seven new chips are now in full production and listed the five rack-scale systems that make up the Vera Rubin platform: NVL72 GPU racks, Vera CPU racks, NVIDIA Groq 3 LPX inference racks, BlueField‑4 STX storage racks and Spectrum‑6 SPX Ethernet racks. (nvidianews.nvidia.com) NVIDIA and coverage from DatacenterKnowledge detailed the NVL72 rack as integrating 72 Rubin GPUs and 36 Vera CPUs, and independent breakdowns show a full Vera Rubin POD can scale to roughly 40 racks, about 1,152 GPUs and on the order of 60 exaflops of inference capacity. (datacenterknowledge.com) The platform incorporates Groq technology after a roughly $20 billion licensing/asset agreement announced on December 24, 2025, with Groq 3 LPUs positioned specifically to accelerate low‑latency decode and token‑generation phases of inference. (cnbc.com) NVIDIA shipped Dynamo 1.0 to general availability at GTC 2026 as an open‑source, datacenter-scale inference OS for routing, KV cache management and disaggregated serving, and vendor benchmarks cited at the show reported up to 7× inference throughput improvements on Blackwell-class GPUs under Dynamo orchestration. (nvidianews.nvidia.com) The company also published the Vera Rubin DSX AI Factory reference design and an Omniverse DSX digital‑twin blueprint developed with partners including Cadence, Dassault Systèmes, Eaton, Jacobs, Schneider Electric, Siemens and Vertiv to standardize power, cooling, networking and token‑per‑watt optimization for large deployments. (datacenterdynamics.com) During the keynote Jensen Huang raised NVIDIA’s revenue outlook toward $1 trillion through 2027 and company briefings plus platform roadmaps forecast initial Vera Rubin partner and cloud deployments in H2 2026, with major cloud providers and OEMs lined up for early rollouts. (datacenterknowledge.com)
