The rapid rise and evolution of AI has created a real opportunity – if not a necessity – across the data center industry to rethink facility infrastructure and build in a way that is both more efficient and more sustainable.
Power and cooling are often in the spotlight, and rightfully so, but the physical infrastructure required to connect increasing volumes of compute power is growing just as rapidly.
To help identify where efficiency and sustainability gains can be made at the connectivity layer, in its latest whitepaper, Corning examined two AI data hall configurations: one using traditional single core fiber and the other using its innovative multicore fiber solution.
The study used a cradle-to-gate, third party-reviewed lifecycle assessment to evaluate the passive optical infrastructure – including optical fiber, cables, housings, trays, ducts, and packaging within the data hall.
“We saw how multicore’s higher pathway density improves material efficiency, deployment speeds, and carbon footprints,” explains Carly Gaj, sustainability manager at Corning. “The results showed that multicore fiber enables up to a 60 percent reduction in embodied carbon per GPU for the passive optical components within the scope of our study.”
Against this backdrop, in a recent DCD>Broadcast episode, Gaj shares further insights into Corning’s study to reveal how high-density pre-terminated passive optical infrastructure can act as a valuable tool in helping to address some of the key constraints shaping today’s data center landscape.
Pressure on multiple fronts
Amid a complex web of AI-driven challenges the central question becomes: where is the greatest pressure coming from? Gaj shares her perspective:
“One that rises to the forefront is density. As GPU clusters grow, the physical space within the data center is becoming constrained, and that space gets exhausted before optical performance limits are reached.”
At the same time, the industry is experiencing a skilled labor shortage. And with AI demand continuing to accelerate, operators are simultaneously facing aggressive schedules and demanding deadlines, driving up both cost and risk. And Gaj says there’s an additional sustainability factor to consider as well.
“Operators are facing rising pressure to decarbonize and meet aggressive greenhouse gas emissions targets. This pressure is extending into the supply chain as well,” says Gaj. “At Corning, we wanted to leverage our decades of expertise in optical fiber and focus on an area that is often overlooked – specifically, the sustainability of passive optical infrastructure.”
Given the interconnected nature of these hurdles, it’s no longer feasible to address each one in isolation. Today’s strategies must increasingly address multiple challenges across the ecosystem in one neatly packaged, quick-to-deploy solution.
The power of four
As AI continues to scale, cable trays are filling up, duct capacity is running out, and racks are becoming congested. This is where Corning’s multicore fiber solution comes in. The solution packs four data pathways, or cores, within the same 125-micron cladding used in traditional single core fiber solutions, making it a fast and straightforward replacement for existing setups. Multicore fiber not only addresses spatial challenges by enabling more fiber inside the same physical footprint, but it also contributes to sustainability initiatives in the process.
“In practice, this translates into fewer cables and fewer housings, less space taken up in trays and ducts, and a more compact physical footprint overall,” explains Gaj. “It’s higher density, which enables improved efficiency, and that directly translates into a reduction in the carbon footprint as well. The total amount of material required is reduced while still delivering the same level of connectivity as traditional single core fiber.”
Gaj is also keen to dispel misconceptions around multicore solutions – including the idea that they require a complete overhaul of existing network designs, or are only suited to new builds:
“In reality, our multicore fiber was designed to be a drop-in replacement for today’s single core fiber,” she explains. “It helps relieve physical pathway constraints while preserving compatibility with existing cable designs and deployment practices.”
Another misconception is that higher density automatically increases operational complexity. “What we found in our study is really the opposite,” adds Gaj.
With fewer cables and connections, deployment and testing can be completed faster – helping to alleviate concerns around labor shortages and pressure on project timelines all at once.
Break it down
Scaling sustainably is a complex industry-wide responsibility that cannot be undertaken in isolation. It requires end-to-end collaboration across the entire data center ecosystem. Equally, it demands clear direction, and Corning is aiming to lead by example.
Providing some clarity to the complexity, the company has broken the sustainability challenge down into two useful categories: a footprint and a handprint. Gaj explains:
“Our footprint is the impact our operations have, and our handprint is the impact we enable through our products. This study on multicore fiber is a great example of that handprint.”
Corning is approaching sustainability from a practical and analytical standpoint to provide operators with an actionable roadmap.
“We’re a company that’s been around for 175 years, and deeply rooted in material science expertise and innovation,” adds Gaj. “So, we take any claims made about our products very seriously and want to make sure we have the data to support them.”
Using a third-party-reviewed lifecycle assessment, Corning’s recent study was built around a real data hall architecture with a simple goal: to create a qualitative and objective assessment that operators can use as they scale AI infrastructure responsibly.
Testing, testing
To identify these valuable outcomes, the design and setup of the study were critical. Rather than simply examining individual components, Corning modeled the entire data hall architecture. Gaj explains the company’s essential thinking:
“When setting up an experiment, it’s important to manipulate only one variable. We were very intentional in being specific about the data center design and the system boundary, making sure we were only looking at the difference between single core and multicore fiber while focusing on the passive optical infrastructure.”
The system boundary intentionally focused only on the passive optical layer: optical fiber, cables, housings, ducts, trays, and packaging – the areas most directly affected by higher-density fiber solutions.
The study compared two generative AI data halls: one using traditional single core fiber and the other using Corning’s four core multicore fiber. Both scenarios supported the same compute footprint – specifically 18,432 GPUs – with the comparison focused solely on passive optical infrastructure.
“When comparing the two data halls, the single-core hall generated about 4,500 metric tons of carbon dioxide equivalent, while the multicore hall generated about 1,800 metric tons,” explains Gaj. “To compare the results fairly, we normalized them on a per-GPU basis, which translated into that 60 percent reduction.”
Keeping the functional compute power consistent effectively removes performance from the debate – bandwidth, latency, and compute capability don’t become trade-offs.
This consistency allowed Corning to examine how higher density impacts material efficiency, carbon footprint, and deployment speed without introducing additional variables.
“We’re not saying multicore fiber solves every sustainability challenge within a data center, and we’re also not making broad statements about sustainability and AI in general,” states Gaj. “We were very intentional about being specific in what we studied and demonstrating quantitatively that the passive optical layer can materially reduce embodied carbon.”
More for less
In today’s high-stakes competitive data center landscape, beyond product innovation, speed to market continues to top priority lists. Corning’s study also highlights faster deployment speeds and operational efficiencies via its multicore approach. Gaj expands:
“Material and deployment efficiency go hand in hand. The modeling in the paper found about a 60 percent reduction in passive optical installation and testing when compared to traditional single core fiber, which translated into up to a nine-month reduction in cable installation and testing schedules in the network design we studied.”
Secondary material waste can also be reduced, particularly when it comes to packaging. While this element may not have the largest impact on the overall carbon footprint, at AI scale it becomes a highly visible waste stream.
“In our study, the reduction in cable also translated into a reduction in cable reels – from about 74 metric tons with single core fiber down to about 19 metric tons with multicore fiber. That’s roughly a 74 percent reduction,” continues Gaj.
This not only reduces waste but also minimizes staging space requirements inside the data center, limiting handling needs to help address labor constraints in the process.
Putting paper into practice
These findings and potential carbon savings only become meaningful when translated into practical, real-world deployments. Corning believes the benefits are ready and waiting:
“The carbon savings are very real, measurable, and actionable,” says Gaj. “They were derived from physical material reductions and primary data, not assumptions based on modeled operational scenarios. For operators, this means the reduction of up to 60 percent embodied carbon per GPU is something they can realize today.”
By making the right design choices at the connectivity layer, operators don’t have to wait for future efficiency gains or major infrastructure changes.
And while power and cooling may continue to dominate sustainability discussions, passive optical infrastructure represents a meaningful lever to pull for unlocking not only sustainability gains, but faster deployment speed, space efficiency, and reduced labor demand with a singular pull.
To hear more about lower carbon connectivity for high density data centers, watch the full DCD>Broadcast episode with Carly Gaj, here.
Read the whitepaper, Multicore fiber for AI-scale data halls: A lower-carbon path to density, speed, and scalability, here.
More from Corning
