The race to build AI infrastructure is usually told as a story about GPUs. However, it's just as much a story about the fiber between them—and the economics of that optical connectivity now sits on the critical path.
To help quantify what’s at stake, ACG Research and Lumen analyzed the total cost of ownership (TCO) behind different optical connectivity models for AI-era data center interconnect. The research shows there’s no one-size-fits-all answer: the right optical model depends on where capacity, distance, operational control and time to market intersect.
The connectivity bottleneck no one put on the slide
Every AI buildout conversation starts with compute: how many accelerators are required, how much power is needed and how much cooling is necessary. Yet there's a quieter constraint that increasingly determines whether all that compute delivers: the optical layer that connects it.
Modern AI clusters can draw 50–100 kilowatt hours (kWh) per rack. That density often forces operators to spread compute across multiple sites in search of sufficient power. As a result, what appears to be a single AI factory is, in reality, a network of facilities that must function as one tightly coupled system. Training datasets, model checkpoints and distributed inference pipelines generate sustained east-west traffic between data centers that dwarfs the north-south patterns of traditional enterprise workloads.
The result is a fundamental shift in what the network is. Data center interconnect (DCI) was once a supporting function. In the AI era, it has become a primary scaling constraint and a key driver of total cost of ownership—the connective fabric between distributed compute resources. When the photonics layer can't keep pace, expensive GPUs sit idle waiting on data.
This mirrors a broader Cloud 2.0 shift: enterprises are rethinking cloud architectures not only for performance, but also for cost efficiency, control and scalability as AI and data-intensive workloads reshape infrastructure economics.
Why optical connectivity is non-negotiable for AI
When AI compute spans multiple facilities, connectivity becomes part of the compute architecture, not just the path between sites. AI workloads impose requirements that only optical transport can satisfy at scale:
- Bandwidth that keeps climbing. AI workloads are rapidly moving from 100G to 400G, 800G and soon 1.6T interfaces, driving multiterabit demand between sites.
- Low, predictable latency. Even small increases in latency can extend job completion times and reduce GPU utilization efficiency, directly undermining the productivity of the most expensive asset in the building.
- Deterministic performance. AI fabrics require near-zero jitter and consistent routing—not best-effort paths that shift under load.
- Linear scalability. Capacity must scale from hundreds of Gbps to multiple Tbps without disruptive architectural changes.
Optical transport—with no contending Layer 2 or Layer 3 traffic in the path—is what consistently meets these demands. The strategic question is no longer whether to use optical, but which optical model best fits a given route, scale and operating capability. The answer to this new strategic question differs based on the requirements for bandwidth, distance, operational control and time to market.
What differentiates Lumen?
Most providers offer a single optical model and expect customers to fit their AI roadmap around it. Lumen takes a different approach, offering three different models that provide the full spectrum of control, scalability and operational burden, so the architecture can match the workload—not the other way around.
Lumen® Wavelength Solutions deliver speed and simplicity through high-capacity optical connectivity over shared fiber, with each customer assigned dedicated, logically isolated wavelengths. Lumen owns and refreshes all the optical equipment, enabling fast deployment and making the service ideal for organizations that need rapid time to market and minimal operational overhead. Performance is comparable to dark fiber, with no Layer 2 or Layer 3 contention on the wave. Additionally, with Lumen® Wavelength RapidRoutes℠, customers get an award-winning 20-day SLA on qualifying site pairs.
Managed Optical Fiber Network (MOFN) uses dedicated fiber infrastructure like dark fiber—with Lumen installing and managing the optical layer—to provide control without the operational weight. Customers get the same architectural control and scalability as dark fiber through a hybrid CapEx/OpEx model—without the need to build and staff a dedicated network operations center (NOC) team. It's especially well-suited for enterprises and neocloud providers with massive multiterabit requirements.
Lumen® Dark fiber includes dedicated fiber infrastructure where hyperscale customers can design and operate their optical network, including a 24/7 optical NOC.
The TCO comparison across Wavelengths, MOFN and dark fiber
Here's where it gets interesting—and where the choice of solution has real money attached to it. ACG Research and Lumen partnered to build total-cost-of-ownership models across two representative AI DCI scenarios: a long-haul ring (Chicago–San Francisco–Dallas at approximately 5,995 route miles) and a medium-haul regional link (San Jose–Los Angeles at around 422 miles), modeled from 400G all the way to 60T.
A few key findings from the ACG research include:
- Wavelengths deliver the lowest TCO across most real-world scenarios—below approximately 20T on long haul and 5T on regional routes1—while also providing the greatest flexibility and the fastest deployment.
- MOFN wins at scale, becoming the most cost-effective option above the breakeven point, with 30% savings versus dark fiber on medium-haul and between 9% and 13% savings on long haul.1
- Dark fiber carries the highest TCO under typical assumptions, including an optical NOC cost of approximately $200K per month1, and pays back only at sustained hyperscale utilization.
The following figure shows the crossover point for long-haul use cases over a five-year period.
Read the full analysis
The complete ACG Lumen Research white paper, Choosing the Right Optical Transport for the AI Era, walks through the full TCO models, the 5- and 15-year cost curves, the decision framework and provides strategic guidance for selecting between Wavelengths, MOFN and dark fiber.
Your AI roadmap is only as scalable as the optical network behind it. Get the ACG and Lumen white paper to see how Wavelengths, MOFN and dark fiber compare on TCO—and which model can keep distributed AI infrastructure moving without overspending.
¹ACG Research, Choosing the right optical transport for the AI era, 2026.
This content is provided for informational purposes only and may require additional research and substantiation by the end user. In addition, the information is provided "as is" without any warranty or condition of any kind, either express or implied. Use of this information is at the end user's own risk. Lumen does not warrant that the information will meet the end user's requirements or that the implementation or usage of this information will result in the desired outcome of the end user. All third-party company and product or service names referenced in this article are for identification purposes only and do not imply endorsement or affiliation with Lumen. This document represents Lumen products and offerings as of the date of issue. Services not available everywhere. Lumen may change or cancel products and services or substitute similar products and services at its sole discretion without notice. © 2026 Lumen Technologies. All Rights Reserved.