Nvidia has given journalists a tour of its Vera Rubin AI infrastructure platform, showcasing new hardware designed to power "AI Factories" that run agentic AI workloads. The event, held at a mini-datacentre in Sunnyvale, California, aimed to demonstrate the physical reality behind AI computing, with executives emphasising faster installation, greater efficiency and infrastructure built specifically to handle the sustained, multi-step reasoning that AI agents require.

The Vera Rubin platform comprises six chips, including the Vera CPU and Rubin GPU, and is packaged into five rack-mountable systems for data centres. Nvidia claims the platform delivers ten times more tokens per watt than its previous GB200 NVL72 system, based on initial CoreWeave results using DeepSeek-R1, while automated assembly cuts compute tray installation time from around 90 minutes to just one minute, a 90-fold improvement. The showcase came a day after Dutch activists targeted a Microsoft-linked data centre with chemical-filled balloons in protest at the environmental and political impact of such facilities, underscoring the tension between the industry's rapid expansion and public concern over its resource demands.