Scaling AI Factories with Co-Packaged Optics for Better Power Efficiency

TL;DR

Co-packaged optics (CPO) integrate the optical engine directly onto the switch package, reducing electrical loss and power draw.
NVIDIA’s Quantum-X InfiniBand Photonics and Spectrum-X Photonics switch families are designed for generative AI and large-scale LLM workloads, achieving high bandwidth with streamlined signal paths.
CPO delivers up to 3.5x power efficiency gains and up to 10x resiliency by reducing the number of optical components, while enabling rapid deployment and improved serviceability.
Bandwidth and port density reach industry-leading levels (up to 409.6 Tb/s and 512 ports at 800 Gb/s). Commercial availability is planned for early 2026 for Quantum-X InfiniBand and in H2 2026 for Spectrum-X Ethernet.

Context and background

As artificial intelligence redefines the computing landscape, the network has become the critical backbone shaping the data center of the future. Large language model training performance is determined not only by compute resources but by the agility, capacity, and intelligence of the underlying network. The industry is witnessing the evolution from traditional, CPU-centric infrastructures toward tightly-coupled, GPU-driven, network-defined AI factories. NVIDIA has built a comprehensive suite of networking solutions to handle the quick-burst, high-bandwidth, and low-latency demands of modern AI training and inferencing at scale. This includes Spectrum-X Ethernet, NVIDIA Quantum InfiniBand, and BlueField platforms. By orchestrating compute and communication together, the NVIDIA networking portfolio lays the foundation for scalable, efficient, and resilient AI data centers, where the network is the central nervous system empowering the future of AI innovation. NVIDIA blog In traditional enterprise data centers, Tier 1 switches are integrated within each server’s rack, allowing direct copper connections to servers and minimizing both power and component complexity. This architecture sufficed for CPU-centric workloads with modest networking demands. In contrast, modern AI factories pioneered by NVIDIA feature ultra-dense compute racks and thousands of GPUs that are architected to work together on a single job. These require max bandwidth and minimum latency across the entire data center, which lead to new topologies where the Tier 1 switch is relocated to the end of the row. This configuration dramatically increases the distance between servers and switches, making optical networking essential. As a result, power consumption and the number of optical components rise significantly, with optics now required for both NIC-to-switch and switch-to-switch connections. This evolution reflects the substantial shift in topology and technology needed to meet the high-bandwidth, low-latency requirements of large-scale AI workloads. NVIDIA blog Traditional network switches that utilize pluggable transceivers rely on multiple electrical interfaces. In these architectures, the data signal must traverse long electrical paths from the switch ASIC to the PCB, connectors and finally into the external transceiver before being converted to an optical signal. This segmented journey incurs substantial electrical loss, up to 22 dB for 200 gigabit-per-second channels, as illustrated in Figure 2. This amplifies the need for complex digital signal processing and multiple active components. The result is a higher power draw (often 30W per interface), increased heat output, and a proliferation of potential failure points. NVIDIA blog In contrast, switches with co-packaged optics (CPO) integrate the electro-optical conversion directly onto the switch package. Fiber connects directly with the optical engine that sits beside the ASIC, reducing electrical loss to only ~4 dB and slashing power use to as low as 9W. By streamlining the signal path and eliminating unnecessary interfaces, this design dramatically improves signal integrity, reliability, and energy efficiency. This is precisely what’s required for high-density, high-performance AI factories. NVIDIA blog NVIDIA has designed CPO-based systems to meet unprecedented AI factory demands. By integrating optical engines directly onto the switch ASIC, the new NVIDIA Quantum-X Photonics and Spectrum-X Photonics (shown in Figure 4) will replace legacy pluggable transceivers. The new offerings streamline the signal path for enhanced performance, efficiency, and reliability. These innovations not only set new records in bandwidth and port density but also fundamentally alter the economics and physical design of AI data centers. With the introduction of NVIDIA Quantum-X InfiniBand Photonics, NVIDIA propels InfiniBand switch technology to new heights. This platform features: NVIDIA Quantum-X leverages integrated silicon photonics to achieve unmatched bandwidth, ultra-low latency, and operational resilience. The co-packaged optical design reduces power consumption, improves reliability, enables rapid deployment, and supports the massive interconnect requirements of agentic AI workloads. Expanding the CPO revolution into Ethernet, NVIDIA Spectrum-X Photonics switches are specifically designed for generative AI and large-scale LLM training and inference tasks. The new Spectrum-X Photonics offerings include two liquid-cooled chasses based on the Spectrum-6 ASIC: Both platforms are powered by NVIDIA silicon photonics, drastically reducing the number of discrete components and electrical interfaces. The result is a 3.5x leap in power efficiency compared to previous architectures, and a 10x improvement in resiliency by reducing the number of overall optical components that may fail. Technicians benefit from improved serviceability, while AI operators see 1.3x faster time-to-turn-on and enhanced time-to-first-token. NVIDIA blog NVIDIA’s co-packaged optics are enabled by a robust ecosystem of partners. This cross-industry collaboration ensures not only technical performance but also manufacturing scalability and reliability needed for large-scale global AI infrastructure deployments. The advantages of co-packaged optics are clear: The switch systems achieve industry-leading bandwidth (up to 409.6 Tb/s and 512 ports at 800 Gb/s), all supported by efficient liquid cooling to handle dense, high-wattage environments. Figure 5 shows NVIDIA Quantum-X Photonics Q3450, and two variants of Spectrum-X Photonics—single-ASIC SN6810 and quad-ASIC SN6800 with integrated fiber shuffle. Together, these products underpin a transformation in network architecture, meeting the insatiable bandwidth and ultra-low latency requirements posed by AI workloads. NVIDIA blog As hyperscale data centers demand ever-faster deployment and bulletproof reliability, CPO moves from innovation to necessity. NVIDIA Quantum-X and Spectrum-X Photonics switches signal a shift to networks purpose-built for the relentless demands of AI at scale. By eliminating bottlenecks of traditional electrical and pluggable architectures, these co-packaged optics systems deliver the performance, power efficiency, and reliability required by modern AI factories. With commercial availability for NVIDIA Quantum-X InfiniBand switches set for early 2026 and Spectrum-X Ethernet switches in the second half of 2026, NVIDIA is setting the standard for optimized networking in the age of agentic AI. Stay tuned for the second part of this blog, where we take a look under the hood of these groundbreaking platforms. We’ll dive into the architecture and operation of the silicon photonics engines powering NVIDIA Quantum-X Photonics and Spectrum-X Photonics, shedding light on the core innovations and engineering breakthroughs that make next-generation optical connectivity possible. To learn more about NVIDIA Photonics, visit this page.

What’s new

Co-packaged optics (CPO) integrate the electro-optical conversion directly onto the switch package, enabling fiber to connect directly with the optical engine beside the ASIC. This design reduces electrical loss to ~4 dB and lowers power consumption to as low as 9W per interface, versus traditional paths that can incur up to 22 dB of loss and ~30W per interface. This significantly improves signal integrity, reliability, and energy efficiency. NVIDIA blog
NVIDIA’s CPO-based systems include Quantum-X InfiniBand Photonics and Spectrum-X Photonics, designed for high-bandwidth AI workloads and dense interconnects. Quantum-X Photonics leverages integrated silicon photonics to achieve ultra-low latency, high bandwidth, and resilience, while Spectrum-X Photonics targets Ethernet with liquid-cooled chasses based on the Spectrum-6 ASIC. NVIDIA blog
The combined offering reduces the number of discrete optical components and electrical interfaces, delivering a 3.5x leap in power efficiency and a 10x improvement in resiliency. The approach also supports faster deployment and improved serviceability, with a reported 1.3x faster time-to-turn-on and better time-to-first-token for AI workloads. NVIDIA blog
Availability timelines: commercial availability for NVIDIA Quantum-X InfiniBand switches is planned for early 2026, and Spectrum-X Ethernet switches are expected in the second half of 2026. NVIDIA blog

Why it matters (impact for developers/enterprises)

For developers building AI models and pipelines, the network becomes a critical performance lever. By reducing electrical loss and power per port, AI workloads can access higher bandwidth with lower latency across large GPU clusters. NVIDIA blog
Enterprises running hyperscale data centers gain energy efficiency, improved reliability, and lower total cost of ownership due to fewer optical components and simplified signal paths. This helps scale training and inference for agentic AI workloads while maintaining operational resilience. NVIDIA blog
The ability to deploy high-density, high-wattage networking with liquid cooling enables data centers to sustain aggressive growth in AI capacity without compromising temperatures or serviceability. NVIDIA blog

Technical details or Implementation

Key performance and topology shifts:
Traditional Tier 1 switches within server racks vs. end-of-row placement in AI factories, creating longer electrical paths and higher component counts. This contributed to higher power consumption and more potential failure points. NVIDIA blog
Co-packaged optics (CPO) eliminate many interfaces by placing optical engines directly on the switch package, reducing electrical loss from up to 22 dB to about 4 dB, and cutting power per interface from ~30W to as low as 9W. This streamlines the signal path and improves reliability and energy efficiency. NVIDIA blog
Products and architectures:
NVIDIA Quantum-X InfiniBand Photonics adapts silicon photonics to InfiniBand, delivering unmatched bandwidth and ultra-low latency for AI interconnects. NVIDIA blog
NVIDIA Spectrum-X Photonics targets Ethernet, with two liquid-cooled chassis based on the Spectrum-6 ASIC and powered by NVIDIA silicon photonics. This reduces components and interfaces and delivers a major efficiency and resiliency boost. NVIDIA blog
Bandwidth and density highlights:
The co-packaged optics approach enables high bandwidth and port density, with capabilities up to 409.6 Tb/s and 512 ports at 800 Gb/s. The design supports dense, high-wattage environments with liquid cooling. NVIDIA blog
Physical and architectural implications:
The integration of optical engines on the switch ASIC reduces the number of discrete optical components, improving serviceability and reliability, and reshaping the economics and physical design of AI data centers. NVIDIA blog
Availability and next steps:
NVIDIA indicates commercial availability for Quantum-X InfiniBand switches in early 2026 and Spectrum-X Ethernet switches in the second half of 2026. NVIDIA blog

Key takeaways

Co-packaged optics dramatically reduce signal loss and power per port, enabling denser, faster AI data-center networks. NVIDIA blog
The Quantum-X InfiniBand and Spectrum-X Photonics platforms push bandwidth, latency, and resilience to new levels for large-scale AI workloads. NVIDIA blog
Availability is on the near horizon, with Quantum-X InfiniBand planned for early 2026 and Spectrum-X Ethernet in H2 2026. NVIDIA blog

FAQ

Q: What are co-packaged optics (CPO)? A: CPO integrates the electro-optical conversion directly onto the switch package, enabling fiber to connect directly with the optical engine beside the ASIC. NVIDIA blog
Q: How does CPO affect power consumption? A: Traditional architectures can incur up to 22 dB loss and ~30W per interface; CPO reduces loss to ~4 dB and power to as low as 9W per interface. NVIDIA blog
Q: What products are involved in this move? A: NVIDIA Quantum-X InfiniBand Photonics and Spectrum-X Photonics switches, including configurations based on Spectrum-6 ASIC, with silicon photonics. NVIDIA blog
Q: When will these products be available? A: Quantum-X InfiniBand switches are planned for early 2026; Spectrum-X Ethernet switches are expected in the second half of 2026. NVIDIA blog
Q: Why should enterprises care about this architecture? A: It enables ultra-dense, high-bandwidth, low-latency networks with improved energy efficiency and reliability, meeting the demands of agentic AI workloads. NVIDIA blog