Nvidia GTC: data‑center push
At Nvidia GTC Jensen Huang unveiled a push into AI inference and new data‑center initiatives, showcased robotics demos and announced partnerships including Groq’s LPU integration — signaling a tighter focus on inference economics (The Motley Fool coverage, Mar 23; CNET GTC roundup, Mar 23) ( ). Jacobs’ data‑center digital twin even featured in the keynote, highlighting simulation + ops convergence (The Motley Fool coverage, Mar 23) (fool.com).
NVIDIA says the Vera Rubin platform will include Groq 3 LPU inference-accelerator racks alongside Vera Rubin NVL72 GPU racks and BlueField-4 STX storage racks, positioning Groq silicon as a dedicated inference tier in the company’s rack lineup. (nvidianews.nvidia.com)) Nvidia told investors and developers at the GTC keynote that cumulative purchase orders for its Blackwell and Vera Rubin architectures could reach $1 trillion through 2027, a figure CEO Jensen Huang cited during his March 16, 2026 address. (cnbc.com)) The Vera Rubin NVL72 rack is described by Nvidia as a 72‑Rubin GPU / 36‑Vera CPU liquid‑cooled system that the company says can deliver roughly 3.6 exaFLOPS of NVFP4 inference per rack and scale to multi‑rack PODs for up to about 60 exaFLOPS. (nvidianews.nvidia.com)) Nvidia presented rack‑level efficiency claims at GTC, saying Rubin configurations can yield up to 5×–10× inference performance improvements and as much as a 10× reduction in cost‑per‑token versus prior Blackwell deployments at comparable scale. (electronicsweekly.com)) Engineering partners showed how Rubin and rack hardware connect to operations tooling: Jacobs announced a Data Center Digital Twin on March 16, 2026 to plan, simulate and optimize gigawatt‑scale AI data centers, and Jacobs’ twin was demonstrated in Nvidia’s Omniverse segments during the keynote. (tmcnet.com)) Nvidia’s physical‑AI showcase included on‑stage robot sequences rendered in Omniverse with the new Newton physics engine and live expo demos from partners such as Disney’s autonomous Olaf and Techman’s TM Xplore I, illustrating real‑world robotics training and simulation pipelines. (youtube.com))
