TL;DR

  • Driven by AI requirements, global hyperscale data center construction is booming. U.S. construction spending reached an annualized $40 billion in June 2025, with major inventory surges across top North American markets.
  • To support massive bandwidth and low latency, cabling is shifting from traditional 12-fiber trunks to ultra-high-density fabrics carrying up to 6,912 fibers, compressed into smaller spaces using emerging very-small-form-factor (VSFF) connectors.
  • Operators are moving away from field termination to factory-built, 100% pre-tested assemblies.
  • Modular, high-density interconnection strategies lower total cost of ownership (TCO) and maximize deployment speed, which has become a vital competitive requirement in the AI era.

# # #

Paulo Campos, President, R&M USA Inc.

The largest Hyperscale deployments are concentrated in the United States and China. Synergy Research Group reported that hyperscale operators ran 1,136 large data centers at the end of 2024, twice as many as five years earlier; the USA accounted for more than half of global critical IT load, with Europe and China each accounting for about a third of the balance.

Bank of America Institute and U.S. Census data shows that U.S. data-center construction spending reached an annualized $40 billion in June 2025, 30% higher year on year after a 50% surge in 2024. CBRE says inventory across the four largest North American markets — Northern Virginia, Atlanta, Dallas-Fort Worth and Chicago — rose 33% year on year in Q1 2026.

As a result of this growth and expansion, Hyperscale data centers are entering a brand-new cabling phase. AI and machine-learning clusters are demanding vast east-west bandwidth, low latency and repeatable build quality. What should DC operators bear in mind in order to optimize cabling for the next wave of AI driven requirements? For one thing, we’re seeing changes to the physical layer. In the past, many operators planned around 12-fiber trunks. Today’s AI fabrics may specify 6,912 fibers: 432 connectors per side, each carrying 16 fibers. Current hyperscale and 5G fiber counts can already reach 6,912, while SN-MT and MMC connectors are emerging to compress density into less space. (both connectors offer a 16-fiber connector with 2.7 times the density of 16-fiber MPO, and are a very-small-form-factor connectors that can be used in high-fiber-count interconnects, transceivers, and onboard optics).

The procurement model is changing too. Instead of buying bulk cable and leaving contractors to terminate on site, operators increasingly want assemblies manufactured to specification. If these are 100% tested in the factory, labeled, and delivered ready to install speed, quality control and supply-chain planning benefit: orders can be split across qualified facilities or partners, then combined without compromising documentation or consistency.

Pretermination can also significantly help reduce dependence on scarce specialist labor. Preterminated cabling can cut installation time by up to 80% compared with field termination. That matters during rollout, but also during changes and maintenance, because plug-and-play links are easier to identify, replace and restore.

Microtubes and blown fiber add another layer of flexibility and are rapidly being taken up by data centers. Microducts allow operators to install pathways first, then blow in fiber as architectures change. Dataintelo valued the air-blown fiber systems market at $3.8 billion in 2025 and forecasts $6.9 billion by 2034, driven by FTTH, 5G, data-center proliferation and modernization.

The best interconnection strategy is therefore modular, high-density and avoids splicing. Ideally, it  should combine preterminated trunks, VSFF multi-fiber connectors, microduct options, automated documentation, flexible manufacturing capacity and rigorous factory testing. High-density infrastructure can reduce TCO by saving space and operating cost, while modular cabling can be reconfigured or expanded without complete overhauls. That’s vital to hyperscale Data Center operators – in the AI-era, faster deployment is not a convenience; it is a competitive requirement.