GPU capacity in Southeast Asia, built on power and data centre sites we develop ourselves. One plan connects contracted megawatts to your running cluster.
We plan the power, data centre and compute as one system. That gives you a clear line of sight from a reservation today to capacity targeted for 2027.
Our six signed sites are backed by more than 1 GW of contracted power. We start with power we can secure, then plan the capacity we can offer.
We are developing sites for dense AI clusters and liquid cooling from day one. The power and thermal design do not depend on retrofitting a general-purpose hall.
Our first sites are in Southeast Asia. Teams can plan training and inference capacity in-region, with data location considered at the infrastructure layer.
Calptic is being designed for sustained training workloads on dedicated clusters. It will not suit every job, and we think you should know that before you reserve.
Single-tenant nodes provide direct hardware access. There is no hypervisor between the job and GPU, and no other tenant shares the fabric.
We are designing checkpoint storage, restart workflows and node replacement so a long training run can resume without starting over.
We design around InfiniBand bandwidth and predictable node performance. An aggregate GPU count says little about distributed training.
We are building for teams with committed roadmaps. If you need one GPU for twenty minutes, there are better places to get it.


Planning a sustained training workload in Southeast Asia? Tell us the scale, timing and location you need. We will tell you what we can plan for.
Not ready to reserve? Talk to our team.