China has launched its first AI computing cluster built around Huawei Technologies' Ascend 950DT chips, with the stated goal of becoming an export hub for artificial intelligence computing power.
The concept is straightforward: instead of shipping chips across borders - which export controls increasingly restrict - sell tokens produced by those chips. Compute becomes a service export rather than a hardware export.
That reframing has regulatory logic. Curbs target advanced semiconductors and the equipment to make them; they are less clear about inference capacity delivered over a network, although Washington has begun closing that gap with rules on remote access through overseas data centres.
The engineering challenge is real. Multi-datacentre training and serving require interconnect, scheduling and fault tolerance that only a handful of companies have mastered, and network latency across borders limits which workloads can be served economically.
Still, the direction is clear: capacity located where the chips are, revenue booked where the customers are.
