NVIDIA Adopts Hybrid Copper-Optics Strategy for AI Factories, NVLink 6 Connects 72 GPUs with Multi-Layer Resiliency

September 20, 2026

Elliot’s Musings USA

Overview

NVIDIA is implementing a hybrid interconnect strategy for its AI factories, utilizing both extensive copper cabling for NVLink to connect 72 GPUs and developing integrated optical engines closer to switching chips. This dual approach tackles the challenge of delivering data to expensive processors while minimizing power, space, and cost. The interconnect market is simultaneously undergoing three major transitions: electrical-to-optical conversion, packaging shifts in optical links, and network topology changes, leading to a redistribution of revenue within the optical supply chain.

In Depth

Key Findings

NVIDIA has strategically embraced a hybrid approach, integrating both copper and optical interconnects to power its AI factories. This strategy is exemplified by NVLink 6, which uses numerous copper cables to link 72 GPUs within a single scale-up domain, while simultaneously advancing switches that embed optical engines closer to the switching chip. This dual-pronged development addresses the critical challenge of efficiently supplying data to high-cost processors while rigorously minimizing power consumption, physical footprint, and overall cost.

Technical / Clinical Details

The core of NVIDIA’s approach for the Vera Rubin platform is the NVL72 core rack-scale compute engine, designed to connect 72 Rubin GPUs into a cohesive scale-up domain. NVLink 6 ensures this connectivity primarily through extensive copper cabling, which remains effective for short-reach, high-bandwidth communication within the rack. Crucially, NVLink 6 is engineered for multi-layer resiliency, achieving a natively lossless fabric through a combination of Forward Error Correction (FEC), Physical Layer Retry (PLR), and Universal Physical Layer (UPHY) recovery. This system corrects signal errors at the silicon level with near-zero latency and employs credit-based flow control to eliminate packet drops, ensuring uncompromising reliability for demanding AI workloads. Concurrently, NVIDIA is investing in next-generation optical solutions, including Co-Packaged Optics (CPO), which involve integrating optical transceivers directly onto the same package as the switching ASIC. This will become essential for longer reaches and higher port densities, pushing beyond the limits of copper.

Background & Context

The burgeoning demands of large-scale AI models have created unprecedented pressure on data center interconnects. Traditional copper cables are reaching their physical limits in terms of bandwidth, reach, and power consumption when connecting hundreds or thousands of GPUs. The interconnect market is currently navigating three concurrent transitions: a shift from electrical links to optical links for increased speed and reach; a move towards more integrated packaging of electronics and photonics within optical modules (e.g., CPO); and a fundamental change in how networks connect machines, evolving towards flatter, more disaggregated, and optimized topologies. These shifts are profoundly altering the optical supply chain, necessitating new design paradigms and leading to a significant redistribution of revenue as different components and integration methods gain prominence.

Strategic Significance & Outlook

NVIDIA’s hybrid strategy is a pragmatic response to the immediate and future demands of AI infrastructure, balancing proven copper technology with cutting-edge optics. While copper will continue to serve crucial short-distance roles, the long-term trend for AI factories points towards increasing reliance on advanced optical interconnects, particularly CPO, to overcome physical barriers. This strategic direction has far-reaching implications for the semiconductor and optical industries, stimulating innovation in high-speed optical components, advanced packaging, and network architectures. The continuous drive for improved power efficiency and increased bandwidth is critical for scaling AI compute capabilities, unlocking the development and deployment of even more complex AI models, and accelerating industrial innovation across various sectors globally. This evolution will solidify the foundation for a truly data-driven future.

Source: https://alphaseeker84.substack.com/p/follow-the-connection-copper-and

Get our weekly technology intelligence — free

Receive an infographic that lets you judge at a glance whether each field’s analysis report is worth reading.

Subscribe Free — Weekly Tech Intelligence

By subscribing, you’ll receive Troy-Technical’s weekly technology intelligence newsletter.