AI Factory demo at GTC
Nebius and DataRobot demoed a validated 'AI Factory' stack at NVIDIA GTC designed to sustain token throughput and low‑latency production agents while keeping governance in place—an industrial approach to productionizing inference at scale. The demo highlights patterns for keeping agent throughput high without sacrificing control. (x.com)
At NVIDIA GTC on March 17, 2026, Nebius and DataRobot presented a validated “AI Factory” stack with DataRobot’s Agent Workforce Platform certified to run on Nebius AI Cloud in a co‑engineered configuration with NVIDIA. (nebius.com) Nebius’s Token Factory advertises sub‑second inference, autoscaling throughput, 99.9% uptime and the ability to sustain workloads described as “hundreds of millions of tokens per minute.” (nebius.com) The joint validation targets agent workloads specifically, with DataRobot’s Agent Workforce Platform integrated into NVIDIA’s Enterprise AI Factory validated design to provide lifecycle management, operational controls and governance for production agents. (datarobot.com) NVIDIA is deepening the relationship with Nebius through a strategic investment and partnership intended to scale a full‑stack AI cloud and to fold NVIDIA blueprints for physical and inference‑optimized factory architectures into Nebius’s platform. (nvidianews.nvidia.com) Nebius’s product materials list support for multiple open and OEM models — DeepSeek, GPT‑OSS, Llama, NVIDIA Nemotron and Qwen — and explicitly offer options to host customer‑owned models for tighter access control and compliance. (nebius.com) Separately, Nebius secured local approval on March 3, 2026 for a planned gigawatt‑scale AI factory in Independence, Missouri, with filings citing up to 1.2 gigawatts of capacity and an approximate 400‑acre campus footprint. (nebius.com)
