Hong Kong telecommunications provider HKT has announced plans to roll out a 3.2Tbps AI Data Centre Interconnect (DCI) Superhighway, a high-capacity network designed to connect multiple data centres with ultra-low latency and support the growing computing demands of artificial intelligence workloads. The company said the service will be the first AI-focused DCI network of its kind in Hong Kong and is aimed at enabling enterprises, AI infrastructure providers and research institutions to access distributed computing resources more efficiently.
The new infrastructure comes as organisations increasingly require access to computing power spread across multiple data centres rather than relying on a single facility. Large language models, AI training, inference workloads and scientific simulations are placing unprecedented demands on networking infrastructure, making high-speed, low-latency connectivity a critical component of AI deployment.
HKT said the AI DCI Superhighway delivers bandwidth of up to 3.2 terabits per second and is an upgrade from its earlier 800Gbps AI Superhighway. The enhanced platform is designed to support distributed GPU clusters and next-generation AI applications that require rapid movement of large volumes of data between geographically separated facilities.
According to the company, the network supports several advanced industry standards, including Remote Direct Memory Access (RDMA), RDMA over Converged Ethernet version 2 (RoCEv2), Multi-path Reliable Connection (MRC) and Ultra Ethernet Transport. These technologies are intended to reduce communication bottlenecks between AI clusters while improving throughput for training and inference workloads.
HKT also plans to deploy advanced hollow-core fibre technology on parts of the network. Compared with conventional optical fibre, hollow-core fibre reduces transmission latency by allowing light to travel through air rather than glass, enabling faster communication between data centres. The initial deployment will connect the Lok Ma Chau Loop innovation hub with Tseung Kwan O, one of Hong Kong's major data centre clusters.
The company said the platform is expected to benefit AI cloud providers, enterprises, universities and research laboratories by enabling them to pool computing resources across multiple locations. This approach allows institutions to scale AI training, perform complex scientific simulations and improve infrastructure utilisation without being constrained by the capacity of a single data centre.
The announcement aligns with Hong Kong's broader efforts to strengthen its AI infrastructure and digital economy. Governments and telecom operators worldwide are investing in AI-ready networking as demand grows for hyperscale computing environments capable of supporting increasingly complex AI models. Industry analysts note that networking has become as important as computing hardware in enabling next-generation AI services.
Telecom operators globally are expanding beyond traditional connectivity services by building AI infrastructure, including GPU interconnects, optical transport networks and data centre interconnection platforms. These investments are intended to support enterprises deploying generative AI while positioning network providers as key enablers of AI ecosystems.
With AI workloads continuing to grow in size and complexity, high-speed data centre connectivity is emerging as a strategic layer of digital infrastructure. HKT's latest deployment reflects the increasing focus on building networks capable of supporting distributed AI computing, enabling organisations to access scalable compute resources while reducing latency across interconnected data centre environments.
Related Articles
