The backbone ring connects data halls across a hyperscale campus. AI clusters, spine fabric, and storage all converge here — demanding maximum density and minimum duct occupancy. A single Celesta 6912F pull is the complete solution.
GPU compute buildings connecting to spine switches in AI hyperscale topology. A building housing 4,000+ H100 GPUs s may saturate a 3456F cable. The 6912F serves a full AI compute building in a single pull — leaving 62% of duct capacity free.
STL’s FEC cabinets and pre-stubbed enclosures provide organised, scalable MMR entry for every carrier and tenant — with enough dark fiber reserve in a single Celesta I/O cable for years of additions without further pulls.
STL’s I/O Celesta IBR routes from campus duct through the building entry to the MMR — eliminating the greyspace splice, recovering 0.5–1.5dB, and simplifying the build.
Connecting a primary campus to a DR site, colo, or remote PoP. The Celesta 6912F’s mass-fusion-splicing efficiency makes it the preferred medium for long-haul extension builds where schedule, budget, and fiber count are all constrained.
AI comes in many forms — including training AI generative AI, inference AI, and agentic AI — and each places distinct demands on data center infrastructure.
Generative AI creates new content by learning patterns from large datasets, generating intense GPU-to-GPU traffic that requires ultra-high-density, low-latency optical connectivity between facilities.
Inference AI applies trained models to make real-time predictions on new data, demanding consistent, reliable fiber paths across core and edge data center for always-available performance.
Agentic AI powers autonomous systems that perceive, decide, and act independently, requiring resilient, always-on interconnect for continuous high-frequency data exchange across distributed environments.
Data center interconnect has become a critical layer in AI infrastructure — every workload depends on moving very large volumes of data between halls/facilities with extremely low-latency, using multiple high fiber count cables. STL’s interconnect portfolio is engineered for exactly this type of application: maximum fiber density per duct pull, factory-verified splice-free performance, and modular infrastructure that scales seamlessly as AI architectures evolve — ensuring the physical network never becomes the bottleneck between ambition and deployment.
Your input is valuable to us. Fill out the form below and we’ll get in touch as soon as possible.
