OSFP-XD Pluggable Connectors are Finding a Home

By Robert Hult | June 24, 2025

OSFP-XD pluggables are targeting AI applications in data centers, high-performance computing (HPC), and select telecom networks.

Connectors designed to transfer information in and out of a server, switch, or router have long been recognized as a potential data bottleneck. Processing chips often spend half their time waiting to receive information necessary to function, resulting in significant loss of system efficiency and wasted energy. At a time when every picojoule of energy is being scrutinized, increasing system efficiency has become critical. For over 20 years, common input/output (IO) connectors such as USB and Ethernet have been constantly upgraded to satisfy performance and density demands. Years ago, we reached the point that traditional I/O connectors using a single lane were unable to keep up with the data transfer rates of high-speed chips. Techniques like high-speed serial transmission delivered some relief. Aggregating multiple lanes, such as four lanes of 100 Gb, is now commonly used to create a 400 Gb channel.

Another factor is the inherent limitations of copper conductors as data rates continue to increase. Although estimates of the practical limits of data rate and reach have been a constantly moving target, copper channels are continuing to become shorter as data rates surge. A general rule of thumb predicts reach is halved as data rates double.

A third factor involves I/O panel density. Advanced servers and switches may require I/O capacity measured in terabytes, a major challenge when many systems are packed in a 1RU form factor with I/O panel dimensions measuring only 1.7” by 18”. System designers labor to increase effective signal transfer per square millimeter while leaving panel space for required cooling air flow.

One of the solutions to this problem is the conversion from electrical to optical data transmission. Optical fiber has immense bandwidth and lower insertion loss, which allows more efficient high-speed transfer at much longer distances.

One of the first efforts to integrate the advantages of optical data transmission in computing equipment resulted in the development of the pluggable Gigabit Interface Converter (GBIC) by the Small Form Factor Committee (SFF) in 1995.

The PCB electrical I/O connector and guidance cage are typically mounted on the I/O panel of the enclosure and accepts mating optical transceivers or copper cable assemblies. The hot-swappable module and socket enabled system designers and integrators to select high-speed copper or fiber optic cable to address specific configuration requirements at the point of installation and make changes in the field as requirements evolve.

The advantages of this system have spawned a series of hot-pluggable small form factor connectors that offer greater channel speed and I/O signal density. Managed by a Multi-Source Agreement (MSA), the SFP specification has been upgraded over the years from one Gb/s to current QSFP and OSFP pluggable transceivers that are rated at up to 1.6 Tb/s aggregating 16 lanes at 100 Gb/s each.

For example, a QSFP-DD (Quad Small Form Factor Pluggable-Double Density) PCB connector and cage assembly consists of a 76-position right-angle surface mount connector that is located at the rear of a stainless-steel cage assembly that guides the incoming module providing EMI isolation and options for dissipating heat.

Competitive pluggable I/O modules offer variations in reach, bandwidth, and power consumption while conforming to IEEE 802.3 standards to assure plug compatibility. Stacked and ganged cages, as well as integrated cooling mechanisms including heat sinks and liquid-cooled cold plates are also available.

System engineers continue to design in the many iterations of pluggable transceivers and mating connectors depending on the requirements of the specific application. Both QSFP-DD and OSFP have become leading contenders for upgrades and new design with both offering 1.6 Tb/s (8×200 Gb, PAM4) pluggable copper cable and optical transceivers.

Relentless pressure to increase I/O bandwidth and panel density resulted in the announcement of the OSFP-XD (eXtended Density) pluggable interface by the OSFP MSA in 2021.

This new connector and module expanded the number of bi-directional electrical lanes from eight to 16 by creating two rows of 60 contacts each. Operating at 100 Gb/s per lane provides 1.6 Tb/s density with the potential of 3.2 Tb/s capability at a future 200 Gb/s per lane rate.

The electrical PCB mounted connectors feature 120 surface mount contacts on a 0.6 mm pitch. The PCB launch design has been optimized for signal integrity. The cage and connector assembly are designed to accept passive and active copper cable connectors as well as optical transceivers. A variety of heat sink options are also available.

The PCB electrical connector and cage assembly is designed to manage up to 40 watts of power dissipation and is capable of supporting 32 ports in a 1RU rackmount enclosure.

OSFP-XD transceiver modules incorporate a somewhat longer and higher profile than OSFP or QSFP-DD connectors.

OSFP-XD optical transceivers that draw less than 25 watts are already available on the market.

1.6 Tb optical transceiver modules are available from several suppliers with transmission reach up to 2 km.

Amphenol has been instrumental in spearheading the development and introduction of the OSFP-XD connector and currently offers OSFP-XD PCB connectors, cages, and external cable assemblies.

Passive copper cable assemblies from Amphenol Communications Solutions are constructed with 32 pair wire using flexible 32 AWG conductors to support a range of bandwidths, including 400G (16x25G), 800G (16x50G), 1.6T (16x100G), and 3.2T (16x200G) per cable. PCIe cables range to 3 meters in length, while Ethernet cables are available up to 2 meters.

Like everything in life, nothing is perfect. OSFP-XD connectors are still in the introductory stage, with limited experience among design and manufacturing engineers. The OSFP-XD profile is not backward compatible with OSFP or QSFP which may restrict adoption to new system design. The level of technical support may be more limited and the cost of OSFP-XD modules may be higher due to the current relatively smaller supplier base. Similar to QSFP and OSFP, thermal management requires careful design, especially in stacked and ganged applications.

OSFP-XD pluggables are targeting AI applications in data centers, high-performance computing (HPC), and select telecom networks. Customer specific applications that require high-speed links to peripheral devices using PCIe Gen 5 are particularly attractive. OSFP-XD links can be a valuable tool in the ongoing process of system disaggregation.

The recent exponential growth of computer clusters dedicated to artificial intelligence has greatly increased potential applications for both short reach passive copper cable and long-distance optical interconnects. Suppliers of OSFP-XD connectors compete on mechanical robustness, signal integrity, thermal management options, compliance with the MSA specification and availability of extensive global manufacturing resources.

The addition of OSFP-XD to the list of high-speed pluggable interconnects will allow designers of next generation switches, routers, and servers to more closely match mechanical and electrical requirements to the most efficient solution. Increased interest in linear optics may result in low power, low latency, and lower cost versions of OSFP-XD in the future. Alternative packaging technologies including co-packaged optics (CPO) will likely co-exist with pluggables to address the specific needs of emerging applications.

Given the incredible rate of advances in AI technology, 1.6 Tb technology available today will likely require an upgrade to 3.2 Tb. It is unclear at this time if 3.2 Tb pluggables are technically and economically feasible, especially in relation to thermal management of greater than 60 watts, but research is already underway to make that determination.

There is little doubt that data rates that seemed impossible just a few years ago have become standard. Few engineers are brave enough to forecast what transmission rates will be required within the next five years, but if system performance demands increase anywhere near the rate of advances in AI, 4.4 Tb or even 6.4 Tb may be next in line.

Read Bob Hult’s review of OFC 2025 and Visit Bob Hult’s Connector Supplier archive for more high-speed coverage, his Tech Trends series, and show reports.

Like this article? Check out our other Data Centers and High-Speed articles, our Datacom Market Page, and our 2024  and 2025 Article Archive. 

Subscribe to our weekly e-newsletters, follow us on LinkedIn, Twitter, and Facebook, and check out our eBook archives for more applicable, expert-informed connectivity content.

Robert Hult
Get the Latest News