APAC Data Center Pipeline Hits 26.5GW, Stressing Metro Fiber

Asia-Pacific's data center development pipeline reached a record 26.5GW as of H1 2026, according to Cushman & Wakefield, and the physical limits of metro hubs are now dictating where that capacity lands. Land and utility constraints are pushing new builds toward metropolitan edges and, in cases such as the Singapore–Johor–Riau Growth Triangle (SIJORI), across national borders altogether.

The constraint that follows is physics: the connectivity between distributed infrastructure environments. Enterprise hybrid architectures already spread applications across colocation facilities, private clouds and hyperscale IaaS/SaaS platforms, and data-intensive workloads — active-active disaster recovery, synchronous storage replication, distributed large language model training — generate heavy east-west traffic between those sites.

How fast is data center traffic growing?

Ciena's 2025 survey of 1,300 data center decision-makers across 13 countries found respondents predicting at least a sixfold increase in data center interconnect bandwidth within five years. Some 81 percent expected AI training itself to run across distributed facilities rather than a single site. The survey's author argues these projections will likely prove conservative and that demands on metro interconnection will continue to rise sharply.

East-west traffic is only half the equation. Much of it, along with traditional north-south traffic — users reaching websites, SaaS platforms, retail systems — has historically been latency-tolerant.

Why does AI inference change the picture?

AI inferencing workloads are different. McKinsey & Company expects inference AI models to overtake training by 2030, representing more than half of all AI compute and 30 to 40 percent of total data center demand. Many inference workloads will require metro deployments because they entail north-south traffic serving users or applications in real time. Latency tolerance is tightening to milliseconds, and increasingly microseconds.

The commercial stakes differ too. A slow training run costs money through inefficient use of IT resources. An inference service is more directly coupled to revenue: fraud checks, trading signals, customer interactions and cybersecurity all depend on response speed.

Where does fiber resilience break down?

Growth in both traffic directions raises capacity, latency and availability demands, making network design as important as compute, GPU and storage location. The uncomfortable reality, the analysis notes, is that contractual diversity does not always equal physical diversity.

Most organizations point to multiple telcos, two or more connectivity services across vendors, and SD-WAN overlay as evidence of resilience. On paper it holds. In practice, metropolitan networks in many APAC cities are built by a small number of carriers who frequently share the same conduits, duct routes and building entry points. Two contracts can, and often do, ride one pit-and-pipe pathway. Carrier capacity resale makes the true route almost impossible to see from outside. The result: diversity on paper, a single point of failure in the ground.

What does this change for enterprise network buyers?

Consideration and adoption of dark fiber solutions are rising as a direct response. Dark fiber enables IT leaders to verify physical routes, ducts and building entry points — potentially essential for business resilience SLAs and, for industry-regulated organizations, potentially subject to board-level review and sign-off. It can also deliver lower latency and unencumbered throughput that shared, medium-lit connectivity may struggle to match.

Organizations can insource optical connectivity or adopt a Managed Optical Fiber Network (MOFN) arrangement; either way, dark fiber gives them greater control over the capacity, latency and physical route diversity on which distributed, latency-sensitive workloads depend.

The shift is already underway. As inference deployments multiply toward McKinsey's 2030 threshold and the 26.5GW pipeline converts into distributed, edge-of-metro facilities, metro connectivity moves from supporting component to competitive advantage.