At the 2026 OCP Global Summit, QCT unveils co-designed AI-infrastructure offerings powered by NVIDIA technologies, including the Vera Rubin POD, ARC-Pro platform, and AI factory digital twins built with NVIDIA DSX blueprint.

Supporting servers extend the ecosystem with CPU-focused units such as the liquid-cooled QuantaPlex T46X-1U and the air-cooled QuantaGrid D66Q-2U to boost throughput and responsiveness for AI workloads.

The NVIDIA Vera Rubin POD unifies five rack-scale systems into a single AI factory platform, featuring 72 NVIDIA Rubin GPUs and 36 NVIDIA Vera CPUs across 18 QuantaGrid D76V-1U trays, with up to 20.7 TB of HBM4 memory for large-scale training and inference.

Early customers include Meta and CoreWeave, with Amazon Web Services also lined up to add Vera as a compute option for agentic workloads.

Vera is positioned to challenge CPU incumbents amid rising demand for agentic workloads, with Arm entering the enterprise CPU race.

Google Falcon Transport support was announced for future ConnectX-10 SuperNICs and BlueField DPUs to improve network efficiency and interoperability with Google’s data center transport.

Google and NVIDIA jointly announced Falcon Transport support on future ConnectX-10 SuperNICs and BlueField DPUs to enable efficient, interoperable networking across Google data centers using DOCA software.

The company markets an edge-to-data-center distributed AI infrastructure as a path to faster deployment, lower latency, and improved token-per-watt efficiency for agentic and generative AI applications.

Open standards and open-source components were released to accelerate AI factory deployment, efficiency and scalability, including MGX 1.2 and DSX AI Factory components.

NVIDIA and Cisco introduced the Ultra-Dense Small Form Factor Pluggable (USFP) to scale 400 Gb/s networking with higher energy efficiency, with collaboration across the ecosystem for cages, connectors and modules.

USFP is a 400 Gb/s-per-lane pluggable optic form factor to boost bandwidth density and power efficiency, developed with industry collaboration for interoperability.

A Vera CPU rack is described as a discrete system adjacent to an NVL72, freeing cores for agentic workloads.