Energy is part of the computer.
The node starts at the utility boundary. Intake, protection, UPS and storage are drawn as one power train, from the grid into the rack.
VIYAT AI Data Centres
Sovereign by design.
Intelligence, sovereign by design.
Regional AI data centres for India, engineered as one system of power, cooling, compute and control.
- Opening tranche
- 300 kVA Planned
- Node scale
- 1 MW class Planned
- Cooling architecture
- Direct-to-chip Design intent
- Placement
- India Architecture
Platform
Five layers, one node.
Power, the hall, accelerated compute, policy and evidence are drawn to operate as one regional system.
A hall that can open small.
Modular blocks let a 300 kVA tranche open, and later capacity repeat on the same site as demand, financing and commitments mature.
Compute that can change generation.
High-density accelerated infrastructure for regional inference and private AI, laid out so later GPU generations can arrive without redrawing the building.
Policy lives in the control plane.
Placement, jurisdiction and customer policy are designed into the orchestration layer. This is architecture and roadmap, ahead of production evidence.
Sovereignty you can check.
Locality, execution records and recovery are designed so a workload can show where it ran, and fail over without leaving its policy.
- Power
- Coolant
- Control
Engineering
From the grid to the chip.
AI infrastructure is a thermal, electrical and software system. The reference architecture centres on modularity, direct-to-chip liquid cooling, and telemetry you can read.
Power runs left to right. The loop returns right to left, into telemetry.
- Power path
- Coolant path
- Control and evidence
Grid interface
Utility intake for the node. Metering and the site boundary sit here, before power enters the hall.
Switchgear
Isolation and fault protection between the grid and the equipment. The hall is fed only through this boundary.
UPS and storage
Ride-through and battery storage in the power train. Shown as architecture, not as a selected product.
Rack power
Distribution from the train into each rack, so a tranche can be energised without redesigning the yard.
Accelerated compute
The servers that run regional inference and private AI. The room is planned so GPU generations can change.
Liquid loop
Direct-to-chip coolant from the cold plate to the manifold. This is the design direction until procurement closes.
Heat rejection
Heat leaves the loop outdoors. The water target depends on this choice, the site and the load.
Telemetry
Energy, cooling, compute and evidence designed to be read as one operating picture.
Efficiency
Design targets for power and water.
Every watt and litre is treated as an engineering constraint. The figures on this plate are targets. They are not certified operating results.
Design targets are subject to final site, EPC and OEM validation, operating envelope and load profile. VIYAT AI does not present these targets as certified operating performance.
Design target
≤1.25
Year-1 power usage effectiveness
Design target
1.18–1.20
Mature PUE ambition
Design target
≤0.20
Litres per kilowatt-hour, water usage
First reference node
Nellore and Kodavalur.
The first Andhra Pradesh deployment is planned for the Nellore and Kodavalur region. It is drawn as a permanent 1 MW-class node, opening at 300 kVA and scaling in stages.
- Regional AI inference and private AI
- Sovereign and regulated workloads
- Business continuity and cyber-recovery
- Staged capacity on the same site
Illustrative share of the planned node. 300 kVA is the opening tranche, not a live load. The site is not yet an operating campus.
Sectors
Close to the work.
A regional node matters when the workload has a place: a district, a hospital, a plant, a public service.
Workload examples for the platform. Not current customer deployments.
Agriculture
Local intelligence for the physical economy.
Crop intelligence, remote sensing, sensors, supply-chain analytics and advice in the local language.
Healthcare
Private AI for clinical data.
Governed analytics, medical workloads, institutional control of data, and recovery.
Manufacturing
Industrial AI that keeps control on site.
Vision, predictive maintenance, digital twins and processing from the line back to the node.
Public infrastructure
Capacity that respects jurisdiction.
Regional digital services, governed workloads, resilience and public-sector data locality.
Sovereignty
Four questions for every workload.
The roadmap is built so a customer can ask where a job may run, where it ran, what evidence exists, and how recovery stays inside policy.
-
Where may this workload run?
Jurisdiction, customer policy and placement rules are designed into the control plane before a job is scheduled.
-
Where did it actually run?
The same layer is meant to record the node that executed the work, not only the node that was allowed to.
-
Can the execution boundary be shown?
Locality and audit records are part of the roadmap, so trust is something a customer can inspect.
-
Can it fail over without leaving policy?
Recovery is designed to move a workload only to a node that still satisfies the same jurisdiction and customer rules.
Build with VIYAT AI
Infrastructure for India’s intelligence economy.
Enterprises, institutions, infrastructure partners, OEMs, lenders and investors can talk with VIYAT AI about the Andhra Pradesh reference node and the distributed roadmap.
Email hello@viyatai.com