The era of Silicon Photonics and Co-Packaged Optics (CPO) is upon us as AI firm Lambda unboxes one of the first NVIDIA Quantum-X InfiniBand platform.
NVIDIA Makes Head Start With Silicon Photonics Co-Packaged Optics, Delivering the First Solution To Lambda AI
Networking has become a major part of AI factories, delivering interconnectivity at unimaginable speeds. These solutions make sure that data moves between massive GPU clusters at the speed of light and also help in reducing switch power, reducing failure points, and bringing higher token throughput efficiency to the table.
NVIDIA has now started delivering its brand new Quantum-X Infiniband solutions to AI firms such as Lambda AI, which have unboxed and housed the platform within their AI ecosystem. The Quantum-X AI networking solution makes full use of silicon photonics co-packaged optics, adding to the AI compute capabilities with 800G capabilities in GB300 NVL72-scale racks. Lambda states that the back-end fabric now accounts for 86% of their networking power in a three-layer cluster.
The company is among the first to receive NVIDIA's latest Co-Packaged Optics solution powered by the Q3450-LD switch. With these switches, the power consumed by the switching layer is reduced drastically, giving GPUs more headroom within AI factories. A standard switch solution consumes roughly 7.0kW of power while NVIDIA's Silicon Photonics solution consumes 3.95kW, offering 3.05kW of savings on GB300 "Blackwell Ultra" platforms.
| GB300 NVL72 cluster size | CPO switches | Network power freed | Power-equivalent extra GPUs |
|---|---|---|---|
| 576 GPUs | 12 | 37 kW | +26 GPUs |
| 4,608 GPUs | 100 | 305 kW | +217 GPUs |
| 10,368 GPUs | 216 | 658 kW | +470 GPUs |
| 41,472 GPUs | 1,440 | 4392 kW | +3137 GPUs |
CPO also reduces failure points. Lambda states that a 128,000-GPU data center uses 655,000 discrete transceiver modules across its switching fabric. Each one of these modules is a potential failure point. With CPO, you get a massive reduction in optical components in the fabric, leading to fewer failures.
So what does the NVIDIA Quantum-X Infiniband "Q3450-LD" look like? The engineering samples sent to Lambda AI are composed of 18 removable light source modules that feed 144 MPO ports. Instead of the traditional OSFP cages, the NVIDIA Quantum-X solution is made out of fiber-array connections that feed directly into the silicon photonics engine.
On the rear end of the unit, NVIDIA houses a 48V DC for power with DGX-compliant busbar connectors. Cooling to the CPO is provided through four UDQ4 liquid cooling connections with dual internal loops, & for those who have already deployed GB300 NVL72 racks, there are a lot of familiar design choices.
| Spec | Detail |
|---|---|
| Form factor | 4U |
| ASIC | NVIDIA Quantum-X800 |
| Ports | 144 x 800G InfiniBand |
| Optical connectivity | 144 MPO connectors |
| Switching capacity | 115.2 Tb/s non-blocking |
| Power input | 48V DC busbar |
| Cooling | Liquid, dual loop |
| Light source | 18 removable external modules (one per eight ports) |
With Agentic AI pushing AI data centers to offer more token throughput and efficient compute capability, the need for elastic and resilient data movement has become essential, and that's where CPO (Silicon Photonics) comes in, allowing more compute in the same data center footprint. NVIDIA is leading this race, and others are trying to catch up with the AI giant.
About the author: A Software Engineer by training and a PC enthusiast by passion, Hassan Mujtaba serves as Wccftech's Senior Editor for hardware section. With years of experience in the industry, he specializes in deep-dive technical analysis of next-generation CPU and GPU architectures, motherboards, and cooling solutions. His work involves not only breaking news on upcoming technologies but also extensive hands-on reviews and benchmarking.
Follow Wccftech on Google to get more of our news coverage in your feeds.
