Driven by the rapid rise of AI and its demand for higher speeds and even higher densities, co-packaged optics (CPO) – where optical components are directly integrated onto the same package or substrate as the main electronic chip – have become a much-needed tool in the data center arsenal.

Inside today’s facilities, the links between optics and integrated circuits within connectivity equipment are operating at a higher pace, making it increasingly challenging to establish these links effectively via traditional approaches.

Once considered a far-off, futuristic technology, CPO involves packaging the optics significantly closer to the electronics, shortening the electrical path between the chip and optical engines, creating a connection that is not only more efficient and more reliable, but at the same time consumes less power.

At a time where the escalating demands of AI continue to push data center processing and networking capacities to their limits, the ability to bolster efficiency and reduce energy consumption has never been more mission critical.

Against this backdrop, in a recent DCD>Broadcast episode as part of Corning’s Telco Innovation Day 2025, Benoit Fleury, director of CPO business development at Corning Optical Communications explores how the company is redefining AI connectivity via CPO, and the hurdles to securing mainstream adoption.

From plate-to-plate to chip-to-chip

Today, optical links are used to interconnect data center racks. But crucially, the optical signal terminates right at the faceplate – a component that provides a clean connection point for networking cables – typically via what is referred to as a pluggable transceiver.

“Pluggable transceivers convert the optical signal to an electrical signal and vice versa – that signal is deployed from plate to plate,” explains Fleury. “With CPO, the signal can go beyond the plate, and it becomes a chip-to-chip connection. In other words, it delivers a truly end-to-end connection from one chip in a rack to another chip elsewhere in the same rack, or in one completely separate.”

A key driving factor behind this shift from plate-to-plate connections to chip-to-chip is power consumption. Power hungry racks are responsible for speeding up the rate of each optical link and increasing capacity inside connectivity equipment. As a result, establishing a connection between the optical transceivers on the faceplate and the target is no longer the most efficient solution.

“One of the key differences between CPO and pluggable transceivers is a much lower power consumption, which is crucial because all of the individual elements inside a rack really add up,” adds Fleury.

Higher densities are also driving this shift: compared to pluggable transceivers, more optical connectors can fit on any given surface area of a faceplate, helping to meet connectivity demands more efficiently and within a smaller footprint.

Taking a holistic view

In terms of design, adopting this approach means it’s more important than ever to ensure that the end-to-end link is strong, robust, and fully optimized in terms of performance. Fleury explains why that can only be achieved by addressing the system as a whole:

“Now that there’s an optical network inside each rack, plus an existing one in between, the network is completely integrated from what's inside the box and what's outside, in an end-to-end fashion. That means it's vital to look at these setups, or links, as a complete system in terms of how they're designed and deployed.”

Rather than considering individual components – the connectors in the front blade, how many fibers are going to be needed inside the box, or how the fiber is going to be routed – taking a holistic view of connectivity systems helps to avoid potential design pitfalls, such as bending the fibers too tightly to fit the space.

“Understanding the requirements around performance, density, reliability, and ease of assembly, – reviewing those trade offs and optimizing all of them in tandem – can only be done by viewing it as a complete, integrated system,” adds Fleury.

The ability to scale up and increase the efficiency of not just individual components but entire systems within the data center, will determine whether networks can grow alongside AI or fall behind. And that’s what a completely end-to-end, chip-to-chip approach delivers: optimization of the complete connectivity ecosystem.

It takes a village

CPO isn’t an entirely new concept, but to deliver broad, mainstream adoption and replace well-established pluggable transceivers, key players across the supply chain must align. That includes everyone from hyperscale operators, down to the chipset manufacturers and providers.

“Each layer of the ecosystem must be prepared to deploy this technology that’s been developed over several years,” says Fleury. “And there's a lot of challenges both in terms of the technology itself and being able to implement that type of architecture at scale.”

Individual optical links operate at approximately 200 gigawatts per second, and each one needs to be lined up accurately onto a single optical engine to meet its corresponding channels. That process must be carried out with expert precision to deliver the robust, reliable, high-performance links that today’s operations demand.

That said, the challenges for achieving high-volume production and deployment are twofold: manufacturing the components themselves, and installing them at scale without sacrificing on precision or performance – a non-negotiable given the complexity involved with alignment and tolerances in particular – an integral piece of the puzzle when it comes to minimizing signal degradation and ensuring efficient light transmission.

Fleury believes that the key to overcoming these hurdles lies in taking a collaborative approach: “There's a lot of partnerships that need to happen between many different players within the data center ecosystem. Certainly one of the key drivers of collaboration is going to be the data center chipset providers. These systems take a long time to develop, and so it's important to work with original equipment manufacturers (OEMs) very early on in the development cycle.”

Working in conjunction with key players across the supply chain also brings valuable insight into technological developments, and how these systems are going to be deployed before it happens. Fleury explains how Corning is thinking strategically for the long-term:

“There are, in fact, multiple development streams going on for the next wave of CPO products, and there will be many more. Our partnerships mean we have visibility over what that roadmap is going to look like and what needs to be done in terms of the technological capabilities required to enable these next generation systems.”

Full speed ahead

Looking to get in front of the challenges that lie ahead, earlier this year Corning demonstrated a proof of concept for its glass bridge technology – an innovation that is helping catalyse the next generation of connectivity. Fleury explains more about the expansion of Corning’s Glassworks AI solution:

“We're looking at a glass-based solution that will achieve much, much higher levels of density and integration. That’s achieved via what we call the glass bridge, which is essentially a little piece of glass with wave guides inscribed on the glass that can be created by means of lithography. The glass substrate and wave guides make the connection between traditional fibers and the silicon. We can achieve this on a very high scale, and it is very cost effective.”

Corning has already been applying the technology in volume – integrating different technologies previously used in different applications, in new and unique ways.

You don’t need to be able to see into the future to know that power demands, densities, and capacities are going to continue to rise, and with that trend, today’s challenges will quickly become overshadowed by those that emerge in just a few short years as a result.

“What CPO is really doing is enabling very high transmission speeds in a way that is optimal in terms of power consumption. I think that’s one of the primary drivers of compact edge optics,” concludes Fleury.

CPO is also meeting the demand for more density within increasingly reduced footprints, and crucially, by maintaining reliability and prioritizing high-performance across the whole system – the solution drives cost-efficiency.

In the face of multiple hurdles to producing and deploying CPO at scale, for Fleury and for Corning, the path ahead is clear: it starts with a holistic view of the complete connectivity system, underpinned by strong partnerships across the entire data center ecosystem to ensure end-to-end optimization.

>> Curious about integrating CPO into your data center? Click here for a personalized consultation.