Rafay Platform with GPU PaaS and Token Factory

Rafay Systems is an infrastructure orchestration company headquartered in Sunnyvale, California, co-founded by CEO Haseeb Budhani. It helps neoclouds, sovereign AI clouds, telecom operators and enterprises turn raw GPU and CPU capacity into governed, self-service AI cloud services that can be sold or shared internally.

The Rafay Platform delivers bare-metal GPUs, virtual machines, Kubernetes clusters and SLURM as a service with hard and soft multi-tenancy, quotas and policy controls. Its AI Factory layer gives users a self-service AI platform, while Token Factory exposes models through OpenAI-compatible endpoints with token metering, pricing and access control for chargeback or monetization. Observability, Kubernetes fleet management across EKS, AKS, GKE and on-premises, and cost optimization complete the stack. Customers include Telus, Firmus Technologies, Cassava Technologies, Yotta, Neysa, Sharon AI and MoneyGram.

• GPU PaaS: bare-metal GPU, VM, Kubernetes and SLURM as a service for tenants.
• Multi-tenancy: hard and soft isolation across customers, teams and projects with quotas.
• AI Factory: self-service AI platform for training, fine-tuning and serving.
• Token Factory: OpenAI-compatible model endpoints with token metering, pricing and access control.
• Observability: multi-tenant monitoring with Atlas AI root-cause analysis.
• Kubernetes fleet management: lifecycle and policy across EKS, AKS, GKE, OKE and on-premises.

Vietnamese telecom and data center operators can turn new GPU clusters into a multi-tenant GPU cloud with billing. Large enterprises can give internal teams self-service GPU and Kubernetes access with chargeback. Sovereign AI projects can sell token-metered access to Vietnamese language models hosted in country.

GPU owners can launch a sellable AI cloud with self-service portals, multi-tenancy and billing data without building the software layer themselves. Token metering lets operators sell pre- and post-paid token plans for models. One control plane standardizes Kubernetes and GPU operations across public, private and on-premises environments. Rafay cites Minima producing 52.5% more tokens with 1.53x less GPU time in production inference.

Telecom operators and data center firms across Asia are investing in GPU clusters and sovereign AI clouds, but most lack the software to sell that capacity as a service. Token-based pricing for AI models is emerging as the business model for these AI factories, including in Southeast Asia where Rafay reports deployments.

Headquartered in Sunnyvale, California. NVIDIA Cloud Validated and Inception member, Cisco Solutions Plus partner, CNCF Silver Member and Kubernetes Certified Service Provider, with AICPA SOC compliance. Partners include NVIDIA, Cisco, Dell Technologies and Lenovo. Reports AI factory deployments across six continents, including Southeast Asia.