MoAI Platform offers the most flexible way for organizing and utilizing heterogeneous GPU resources.

  • Per-process fine-grained GPU allocation

  • GPU aggregation (N GPUs to 1 virtual GPU)

  • GPU partitioning (1 GPU to N virtual GPUs)

  • Seamless GPU scaling

  • Customizable scheduling policy

  • Integrated into Kubernetes

Our core value is virtualizing heterogeneous GPUs to enable efficient AI infrastructure.

Unified Access to Any Accelerator

Without MoAI Platform

Different GPUs require different hardware and software platforms, and users need to migrate between them.

With MoAI Platform

Users can utilize different types of GPUs through a unified software stack and switch between without migration.

Flexible Infrastructure Scaling

With MoAI Platform, AI data centers can flexibly adopt and combine diverse accelerators, avoiding vendor lock-in and achieving optimal performance across AI workloads without sacrificing ease of use.

What’s Inside?

Framework-Level Virtualization

Our virtualization technology is implemented at the software level (within PyTorch), independent of the underlying GPU hardware. The virtual device, called MoAI Accelerator, is exposed as a PyTorch device and can execute any PyTorch program. This approach minimizes overhead while providing more comprehensive functionality than hardware-level virtualization.