Science & Tech · Big Tech & Internet · 1 hr ago
Cisco and Nvidia plan 12.8 Tbps optical module for AI data centers
Cisco and Nvidia are developing a compact optical module called USFP for AI data centers.
These facilities connect thousands of processors, and the growing volume of data they exchange puts pressure on network links and cabling.
USFP is designed to carry up to 12.8 terabits per second, far above the roughly 1.6 Tbps capacity cited for current top-end modules.
That could let operators connect more systems through existing switch infrastructure and reduce the number of cables or extra network layers needed.
The module would use a two-step connector and support up to 100 watts, but its denser design also brings engineering and reliability challenges.
Cisco and Nvidia are working with connector makers on an open standard for compatible modules.
No commercial launch date has been announced, and the announced performance and efficiency gains remain design targets.
Cisco and Nvidia are developing a pluggable optical module called USFP for AI data center networks.
The companies say the module could deliver up to 12.8 Tbps of bandwidth.
The design aims to increase bandwidth density while retaining the replaceable format of existing optical transceivers.
The companies plan to develop a common hardware standard with connector manufacturers TE Connectivity and Amphenol.
A commercial launch date has not been announced.
- Who
- Cisco and Nvidia, with connector manufacturers TE Connectivity and Amphenol contributing to development.
- What
- They are developing USFP, an optical module designed to deliver up to 12.8 Tbps.
- When
- Announced in October 2026; a commercial launch date has not been announced.
- Where
- At the Open Compute Project global summit, according to The Register. The sources do not state where the module will be manufactured.
- Why
- To provide higher bandwidth density for networks connecting AI computing systems.
This story does not have two clearly opposing sides.
Nvidia first teased its Vera Rubin Ultra platform, according to The Register.
Cisco and Nvidia announced plans for USFP; The Register reported the announcement at the Open Compute Project global summit.
- Maximum bandwidth
- Up to 12.8 Tbps
- Lane speed
- 400 Gbps per lane
- Initial lane count
- 16 full-duplex lanes planned; future versions could expand to 32 or more
- Power
- Designed to dissipate up to 100 watts
- Connector manufacturers
- TE Connectivity and Amphenol
3 sources
What people are saying
Building the world’s largest AI factories starts with rethinking the smallest of connections
Posts on X
NVIDIA and @Cisco introduce pluggable optics to scale AI factories. USFP is the first module to support 400Gb/s per lane, delivering 4x the bandwidth density of today's optics. AI builders can connect more compute to the same switch — no extra network tiers, less power per bit