Rubin Platform Boosts AI Power 5x, Triples Rack Power to 110kW
NVIDIA has revealed its Vera Rubin (VR) platform is in full production, setting a new standard for artificial intelligence performance that necessitates a fundamental overhaul of data center infrastructure. Announced at CES in Las Vegas, the VR200 NVL72 model delivers between 3.5 and 5 times the AI computing capability of its Blackwell GB300 predecessor. This generational leap in performance comes with a correspondingly massive increase in energy requirements.
The platform's power demands represent a step-change for data center operators. The VR200 NVL72 requires 110kW power modules in a 3x3U configuration, more than tripling the 33kW modules commonly used for the Blackwell GB300. This architecture, known as Oberon, is built on a 54-volt DC power supply and makes an uninterruptible power system (UPS) a mandatory component, signaling a significant new wave of required capital investment.
NVIDIA Mandates 100% Liquid Cooling for New Rubin GPUs
To manage the intense heat generated by the new platform's power density, NVIDIA has mandated 100% liquid cooling for all Rubin systems. This is a significant increase from the Blackwell platform, which required an 80% liquid cooling ratio. The VR200 NVL72 incorporates an upgraded thermal system featuring a micro-channel cooling plate design, explicitly engineered to handle the increased thermal load.
This shift is expected to trigger a significant spending cycle across the data center supply chain. Companies specializing in advanced liquid cooling technologies and high-capacity UPS solutions are positioned to see a substantial increase in demand. The move also suggests that older data centers not designed for high-density liquid cooling and massive power draws may face accelerated obsolescence. Looking ahead, NVIDIA's roadmap indicates a future shift to an 800-volt high-voltage DC power supply, with the Rubin Ultra product launch in 2027 expected to further accelerate this transition.
