NVIDIA at GTC: agentic AI goes production
NVIDIA used GTC 2026 to push ‘agentic AI’ as an enterprise operating layer—releasing toolkits like NemoClaw and touting full‑stack AI factory deployments that pair Vera Rubin infrastructure with production agent frameworks. The message: multi‑step autonomous workflows and integrated hardware/software stacks are moving from demos to real deployments in manufacturing and logistics. (bain.com; blogs.nvidia.com)
Jensen Huang told the GTC keynote that NVIDIA is tracking demand that could amount to roughly $1 trillion in Blackwell and Vera Rubin orders through 2027. (cnbc.com 1) (cnbc.com 2) NVIDIA said Vera Rubin ships with seven new chips now in full production and enumerated five rack-scale system types including NVL72 GPU racks, Vera CPU racks, Groq 3 LPX inference racks, BlueField-4 STX storage racks and Spectrum-6 SPX ethernet racks. (nvidianews.nvidia.com) NVIDIA’s Groq 3 LPX, presented as a rack-level LPU accelerator co‑designed for Vera Rubin, was claimed to deliver up to 35x higher inference throughput per megawatt and a 10x revenue opportunity for trillion‑parameter models versus prior platforms. (developer.nvidia.com) NemoClaw was released as an enterprise OpenClaw stack that installs in a single command, bundles NVIDIA’s Nemotron models with the new OpenShell runtime for sandboxing and policy guardrails, and targets always‑on local deployments on NVIDIA GeForce RTX, RTX PRO workstations, DGX Station and DGX Spark systems. (nvidianews.nvidia.com) HPE showcased Vera Rubin NVL72 readiness and unveiled the GX240 blade with up to 16 Vera CPUs per blade and an option to scale to 40 blades (640 Vera CPUs per rack), with HPE saying certain Vera Rubin systems will be available in late 2026. (sdxcentral.com) (businesswire.com) Microsoft announced it had powered on Vera Rubin NVL72 hardware in its labs and made a Foundry Agent Service generally available, positioning a hyperscaler path to Vera‑backed agent deployments. (winbuzzer.com) NVIDIA also published a Vera Rubin DSX AI Factory reference design and an Omniverse DSX digital‑twin blueprint to let enterprises simulate rack‑scale layouts and token-per-watt tradeoffs, while partners such as Wiwynn and Eaton demonstrated NVL72 liquid‑cooled integrations on the GTC show floor. (finviz.com) (wiwynn.com) Industry coverage noted that Dell, HPE and storage vendors are packaging full‑stack “AI factory” services around Vera Rubin to accelerate production deployments, and NVIDIA highlighted more than 80 MGX ecosystem partners building for the platform. (datacenterknowledge.com) (nvidianews.nvidia.com)
