Optical Internetworking Forum (OIF) responds to AI’s accelerated network demands

Lightwave: Let’s start with an update on OIF’s Common Electrical Interface (CEI) 448G framework.
Nathan Tracy: Traditionally, whenever the OIF comes to a new data rate for the industry, we find it has great value when we publish a white paper or a framework document at the start of that industry transition. It helps to state the challenges of this new data rate, then outline the opportunities. That's what we did with 448. All members worked together to produce a white paper.

Cathy Liu: So, we published the white paper, which is based on collaborative work with all the OIF members. This white paper highlights the needs, challenges, and potential solutions for next-generation interconnects capable of over 448 Gbps to support AI growth.

LW: OIF’s CEI, which was described as providing a foundation for the next decade of interconnect innovation, is an example of the organization’s dedication to interoperability. Could you provide us with details on how we can bridge the new divides in interconnect innovation?
Liu: It’s the next decade or shorter because everything is moving faster right now, especially for AI growth. We typically take at least 5 years, and up to 7-8 years, to develop the first generation of the solution. Right now, we're still working on our 224G projects, and we expect to finish them next year. However, there is customer demand for the next generation that requires double the existing bandwidth and speed, not even today, but yesterday. And this is why I say maybe not the next decade, maybe only the next two or three years. Who knows? However, the need is there, and we need to move faster. The members are working very hard within the short window to identify the challenges and the solutions in front of us. We also need to identify the homework we need to do next year. We need to kick off the real work right away, and the framework identified the direction and the project we need to focus on for next year.

Tracy: Certainly, the industry has not reached full agreement on deliverable solutions or technical implementations, but it has agreed that this is the most challenging data-rate jump we've ever faced. When we look at the conflicting demands from AI and ML network operators, they're overwhelmed by the challenge of managing power consumption and reach: how do they physically establish GPU-to-GPU and compute-to-compute connections? But at the same time, you know we're debating, or we will be debating, what that implementation will look like from an electrical interface, an optical interface, a modulation perspective, and a latency impact perspective. Everyone agrees we need this framework document to get the industry started. With this foundation of information, we can now begin submitting project start proposals to the OIF. At 100 gig and 200 gig, we developed specifications for several reach objectives to optimize power consumption for each. And that is exactly what we've now set the stage to do, with the usual debate and consensus-building. There’s still work to do, but you know we now have a solid foundation to start with.

LW: In February, OIF held its first quarter meeting in Palm Springs. In addition to the 448 frameworks, were there any other items OIF addressed during that meeting that we might see in the future?
Tracy: These next-generation electrical interfaces and how they interact with higher-level architectures were a key focus of our discussion, and we had a productive discussion on which project proposals to bring forward and what our targets will be for those projects. At the same time, the OIF is a multifaceted organization, so we held parallel sessions at the same Q1 meeting to discuss and debate energy-efficient interfaces. This is where we define lower-power consumption, electrical, and optical implementations, and we are developing a white paper to support them.

Of course, we have our coherent optical working group, where the 400 and 800ZR work was done and has been published. We're now focused on completing the 1600 ZR project, ZR+ and our coherent light work to determine when the industry will be ready to bring it into buildings and campus environments. The Common Management Interface Specification (CMIS) activity is always in the spotlight. CMIS is the common management interface we have defined and published, and we continue to generate additional clauses and related work. So, this is the management interface between pluggable media, whether it's copper or optical, and the host silicon. This is what enables AI to scale at this rate: we share a common management protocol between the pluggable media and the host, whether a compute host or a switching host.

LW: Along with the first quarter meeting, OIF will have a large-scale demonstration at the OFC trade show. Do you expect that to be a big attraction during the event?
Liu: For the upcoming OFC show, I want to highlight our OIF booths and the interoperability demo. In the past few years, the industry interop demo has attracted significant interest, not only at OFC but also at ECOC. And that is how we demonstrate interoperability capabilities across multiple vendors within the OIF membership and across the industry, not only for electrical but also for coherent optical and CMIS. At OFC 2026, more OIF members can participate in the interoperability demonstration.

LW: Coherent optics. While not a new topic, how is this concept evolving to meet the needs of the DCI segment?
Tracy: It's a deep and wide topic. For years, coherent optics was the only technical solution that enabled long-reach optical links. The only way it could operate is if you had the same company's transponder on both ends. It was all about the secret sauce embedded in the transponders that enabled the network operators to get the longest possible reach. With the introduction of ZR, this is where, instead of trying to make coherent optics the absolute longest possible reach, they could be; it was about making coherent optics good enough. So, the reach goals were reduced to approximately 100 kilometers, with a range of 80 to 120 kilometers. These became the building blocks for establishing data center interconnect (DCI) networks, enabling one campus to connect to another with sufficient bandwidth to meet the requirement to appear as a single campus. It essentially became a fabric connection between two buildings, potentially with that much separation.

As soon as 400G ZR hit the market, it took off, spurring the development of 800G ZR due to insatiable bandwidth demand. Work is now underway at 1600G ZR, with various derivative products emerging from that. There are ZR+, open optics, and open ROADM. And other groups as well, where they need the basic building block of the ZR solution, but then they, you know, put some of their own unique application or features on top of that. If you can get the industry to agree to work on just-good-enough coherent optics, which is basically what ZR was, then the next logical building block is OK, as we look at this insatiable thirst for bandwidth and how we enable the industry to bring value to even shorter lengths, perhaps within a campus or within a building?

And if we'd look at it, you know, at a simplistic level, we'd say it's going to be too expensive, it's going to have too much power consumption. There are always disruptive implementations. And so that's what we're exploring with the coherent light activity. It is to see what can be accomplished, and my guess is it probably won't happen overnight. It will require some smart people to work together. Still, I do think we will come to a day when we see coherent deployments across a campus, and even more intimately within a building, simply because of the fat pipe that coherent enables and the need for these AI training and inference applications, which require that fat pipe.

LW: While there is great emphasis on power and cooling, are you seeing more data center providers get involved with OIF?
Tracy: It is truly mind-boggling the rate of growth and just the compound annual growth of the data center market segment. As you review the OIF's membership, we now have over 160 companies. We're focused on solving the problems these AI/ML systems need, while also serving enterprise datacom and telecom applications, driven significantly by hyperscale and AI compute.

LW: As we close, could we get some quick final thoughts about what is to come in 2026?
Tracy: It's just a profoundly exciting time. OIF is operating at the center of the storm. We're involved in all the hot topics in our industry about how to minimize power consumption, how to increase bandwidth density, and how to enable next-generation longer links, higher density with those longer links, how to control those links, and how to control the complex networks that are being enabled with these technical solutions. This is our busy time of year, so we held our first-quarter meeting in February. Immediately after that, DesignCon, where OIF had several panel sessions. Right on the heels of that will be OFC, where, as Cathy mentioned, we'll have our interoperability demo at the OIF booth, along with three panel sessions on the latest technologies, progress, and challenges.

Liu: I expect everyone will be very busy, not only with OIF work but also with our daily work for our employers. At Broadcom, where I work, we have many projects, and development is ongoing in parallel. I think the same pressures affect all organizations, not just companies or OIF. Standard development organizations were facing the same challenges and pressures. I think that to achieve results and move faster, working together is the only solution.

To address the 448G challenge, channel characteristics and bandwidth limitations, we must work together to identify and deliver solutions for our key customers and to enable collaboration among standard development organizations, including OIF, the Ethernet Alliance, and the IEEE. UALink and UEC will also be important. So, I am excited about 2026.