Singapore-based AI Research Institute
Service:
AI Infrastructure & Model Services
Industry:
Education
Size:
Multi-department
A Singapore institution of higher learning created a shared GPU-enabled cloud environment for professors, researchers, students, and commercial research projects while reusing existing heterogeneous hardware.
Introduction
GPU-enabled research projects had historically been provisioned independently, each with its own approval process, budget, hardware, vendor environment, and operating model.
As research demand increased, this project-by-project approach created duplicated investment, inconsistent user experiences, limited visibility into available resources, and long lead times for accessing AI compute.
Challenge
The institution needed to address several problems:
GPU resource provisioning could take 3–6 months
Each research project procured and managed infrastructure separately
Existing hardware could not easily be shared between projects
No standard management or security environment existed across deployments
Identity and access needed to follow university Active Directory policies
Network access, audit, VPN, and 2FA requirements had to remain consistent
Different vendor environments created fragmented user experiences
Solution
A shared enterprise cloud environment was implemented remotely within six weeks using heterogeneous hardware reused from previous projects.
Departments and projects could be created within a common environment and assigned to professors, researchers, and students.
Users were able to request GPU-enabled compute resources across multiple Nvidia GPU generations using passthrough or virtualized modes, while existing university identity, network, and security policies remained in place.
Result
The institution moved from isolated project infrastructure toward a reusable AI and research-computing foundation.
Reported outcomes included:
Deployment completed in six weeks
80% reported hardware cost savings
Approximately 50% lower spending through shared procurement and reduced duplication
Existing heterogeneous hardware reused
Multiple projects and departments supported through one environment
Existing AD, 2FA, VPN, audit, and network policies retained
GPU resources available across multiple Nvidia accelerator models
24×7 operation, including during maintenance and upgrades
