NVIDIA and Cisco have announced a new networking standard for AI data centres. It is designed to link more computing systems without expanding the existing network.
The companies unveiled the Ultra-Dense Small Form Factor Pluggable (USFP) module at the Open Compute Project (OCP) Global Summit in Santa Clara, California. USFP supports electrical signals of up to 400 gigabits per second (Gb/s) per lane.
The companies say it offers four times the bandwidth density of today’s 200 Gb/s-per-lane optical modules. It also uses about 1.3 times less power for the same work. NVIDIA and Cisco are developing the standard with TE Connectivity and Amphenol to encourage wider adoption.
Why Data Centres Need It
AI models are growing and demand for AI inference is rising. Data centres must connect more computing systems while managing power, cost and network complexity. Traditional expansion can call for more switches and optical links. That adds latency and uses more energy.
USFP aims to carry more data in the same physical space. The first planned modules will have 16 full-duplex lanes. They will be similar in size to the widely used Octal Small Form Factor Pluggable (OSFP) module. Total capacity can reach 12.8 terabits per second (Tb/s). Operators could then connect more computing resources without adding network layers or rebuilding existing facilities.
Design and Cooling
The module supports up to 100 watts of power dissipation. It needs no separate cooling circuit inside the transceiver. Its connector design aims to cut contact wear and improve signal reliability.
Cisco said USFP is designed to deliver a new level of bandwidth density. It keeps the serviceability, deployment flexibility and operational simplicity of pluggable optics. Bill Gartner, Senior VP and GM of Optical Systems and Optics at Cisco, said this lets AI infrastructure scale without extra operational complexity.
Gilad Shainer, Senior VP of Networking at NVIDIA, said the largest AI factories start with rethinking the smallest connections. He said the companies are building an ecosystem to link more computing systems more efficiently for the next generation of AI factories.
An Open, Multi-Supplier Path
NVIDIA and Cisco are working with TE Connectivity and Amphenol on the module’s cages, connectors and optical components. They plan to advance USFP through a community-led, multi-source project. That would let suppliers build compatible hardware.
Wider Open-Standards Push
The announcement is part of NVIDIA’s broader push for open standards in AI infrastructure. At the summit, NVIDIA also contributed designs for its Vera Rubin rack and compute trays through its open MGX architecture.
NVIDIA announced software components for managing AI infrastructure. It also worked with industry partners on standards for data centre design and power grid resilience.
Meta is adopting NVIDIA’s Spectrum-6 Ethernet switches for its AI infrastructure. NVIDIA and Google announced support for Google’s Falcon transport protocol on future NVIDIA networking products.
