NVIDIA

Overview

NVIDIA is the dominant designer of GPUs and accelerated computing platforms for AI training and inference. Its data center GPU architectures — including Blackwell and Vera Rubin, now in full production — define the power, cooling, and rack-level infrastructure requirements that the rest of the data center ecosystem must accommodate. At Computex 2026 (GTC Taiwan), Jensen Huang declared "useful AI has arrived" and announced Vera Rubin as a five-rack-pod supercomputer explicitly built for agentic computing, alongside the Vera CPU (NVIDIA's first CPU, with 88 Olympus cores and 1.2 TB/s LPDDR5X memory), the RTX Spark PC platform co-designed with Arm and MediaTek, and the open Nemotron 3 Ultra model. NVIDIA's rapid release cadence (approximately 12 months between GPU generations) creates continuous pressure on facility infrastructure to evolve, particularly in liquid cooling, power delivery, and rack design.

Within the OCP community, NVIDIA's influence is pervasive across the Data Center Facilities track. The company co-leads the Power sub-project and CDU sub-project through Pardeep Shahi, who also co-presented the rack emulator design and participates in the liquid cooling TCS panel. NVIDIA's 45 degrees C TCS specification for Blackwell and Rubin GPUs is a key driver of cooling system design across the industry. The company's promotion of 800V DC rack architecture is shaping power distribution standardization efforts, and its cold plate channel width reductions are driving tighter filtration standards for liquid cooling loops.