Cyberjaya
Cyberjaya, Malaysia
Primary campus inside the Multimedia Super Corridor. Tier III design intent, with the production fleet and the token factory in operation.
IT loadMulti-MW
PowerGrid
CoolingAir
AI infrastructure · Southeast Asia · expanding network
Compute that stays
Compute that stays
in the region.
AGATE GREEN runs GPU capacity across Southeast Asia — for teams that need serious performance without exporting their data, their assets or their pipeline.
The estate · multiple sites · one operator
Built as modules.
Built as modules.
Run as one estate.
Power and cooling ship inside the module, tested in a factory before it leaves. Capacity arrives in blocks that match your ramp — not a landlord's financing cycle.
The platform · two tiers · measured, not quoted
Two tiers.
Two tiers.
No oversell.
RTX PRO 6000 Blackwell for inference and visual computing. MI300X for work that has to be spread across devices. We benchmark your model on the tier we recommend, before you sign anything.
Power in · grid + gas Data out · to users
Multi
Site network
Southeast Asia · expanding
Next-Gen
AI infrastructure
Supporting various GPU platforms
Off-grid
Power
Microgrid connection · grid + gas
100%
Green computing position
Documented per customer on request
Our network
Multiple sites, one operating team.
Reserved, planned and on-request sites across Southeast Asia, with the network still expanding. Every site is modular, and every site is run by the same engineers.
Klang
Klang, Malaysia
Grid plus an on-site gas microgrid, planned at 20–40 MW by phase. Firm capacity commissioned in blocks instead of waiting on transmission reinforcement.
IT load20–40 MW
PowerGrid + Gas
CoolingLiquid · chip/rack
Melaka
Melaka, Malaysia
The largest site in the plan: 80–100 MW built for training, on the same grid plus gas model as Klang, laid out as repeated halls around one power spine.
IT load80–100 MW
PowerGrid + Gas
CoolingLiquid · chip/rack
Bangkok
Bangkok, Thailand
Thailand capacity on request, serving mainland Southeast Asia's enterprise AI market from a modular hall built to the Cyberjaya pattern.
IT loadPer request
PowerGrid
CoolingAir
Central Java
Central Java, Indonesia
Frontier training site at 1.3 MW, built for bandwidth-bound workloads. Air-cooled, carrier-neutral, sized to the interconnection available today.
IT load1.3 MW
PowerGrid
CoolingAir
West Java
West Java, Indonesia
Second Indonesian site at 10 MW. Diversity, disaster recovery and overflow for customers who cannot accept a single-site dependency.
IT load10 MW
PowerGrid
CoolingLiquid (direct-to-chip)
How we deliver
From module to model.
One team carries a workload from site selection to a metered endpoint. Nothing in that chain is subcontracted to a party we do not operate.
01
Site & power
Land, interconnection and a power strategy agreed before a single GPU is ordered — including the renewable position.
02
Modular build
The hall arrives as factory-assembled, factory-tested modules, which moves commissioning risk off the critical path.
03
GPU integration
Nodes, fabric and scheduler configured against the workload you actually run, not a vendor reference sheet.
04
Operate
24/7 operations, on-site hardware response and consumption reporting broken down per site and per customer.
Get started
Tell us what you're running.
Send the workload, the model class and the latency target. We come back with a reference configuration — and a measured benchmark on the target hardware before you commit.