Spectro Cloud, an NVIDIA DSX ecosystem partner, is pleased to join NVIDIA in today’s launch of NVIDIA Run:ai as an open-source platform. NVIDIA Run:ai provides AI workload and GPU orchestration within NVIDIA DSX OS, the open, modular software portfolio of the NVIDIA DSX AI Factory Platform. Through a tested PaletteAI stack profile, customers running PaletteAI can now access, deploy, and operate NVIDIA Run:ai as part of a lifecycle-managed AI infrastructure stack. Open source enables Spectro Cloud to integrate the NVIDIA Run:ai platform more directly into PaletteAI while reducing licensing friction. Together, NVIDIA Run:ai and PaletteAI pair intelligent GPU scheduling with lifecycle-managed infrastructure, helping customers put more of their GPU capacity to work without taking on greater operational complexity.
Turning idle silicon into throughput
GPUs are among the largest investments in AI infrastructure, yet utilization often remains stubbornly low. At current GPU prices, the cost of that idle capacity adds up quickly. NVIDIA Run:ai helps operators address this challenge through dynamic workload scheduling and GPU resource allocation. It pools GPUs across teams instead of statically assigning them to projects, and allocates capacity dynamically based on priority, quota, and workload need.
NVIDIA Run:ai also supports fractional GPU allocation, enabling multiple workloads to share a single GPU rather than reserving the entire device for one job. Teams can right-size GPU memory allocations while compute capacity is shared across active workloads, making unused resources available for other work. This allows inference, fine-tuning, training and interactive workloads to run more efficiently across a fixed GPU pool, providing greater flexibility as demand changes.
Why we integrated NVIDIA Run:ai into PaletteAI
NVIDIA Run:ai solves an important GPU orchestration problem. But production AI infrastructure is a stack, and every layer of that stack has dependencies.
NVIDIA Run:ai runs on Kubernetes, which runs on NVIDIA GPU Operator and NVIDIA Network Operator, which runs on GPU-configured bare metal. Every layer must stay version-compatible with the layers above and below, and updates ripple through the stack in both directions. A new NVIDIA Run:ai release may need matching GPU Operator versions. A Kubernetes CVE patch can shift what drivers are supported. Every change opens a compatibility question that platform teams must resolve, and every upgrade risks breaking something further down. When every team building AI infrastructure inherits that same problem, standardization becomes impossible; every team ends up with a different stack.
PaletteAI addresses that operational challenge by managing the AI stack as a single unit. Our Featured Profile Bundles package the whole thing (OS, Kubernetes, GPU Operator, storage, networking, observability) into tested version combinations for every component, reducing the dependency validation work customers must perform. There’s now a Featured Profile Bundle for NVIDIA Run:ai.
PaletteAI Studio is where we publish tested components for the AI stack. Beyond NVIDIA Run:ai, the catalog includes NVIDIA Dynamo, KServe, Kubeflow, Confidential Mind, and others, all versioned and tested against the underlying platform.
When a customer runs the same profile across public cloud, on-prem data centers, edge locations, and air-gapped facilities, they get identical tested stacks in every one. NVIDIA Run:ai stops being an integration project and becomes a component of a tested AI platform, deployed through PaletteAI Studio and kept current by the same automation that manages the rest of the stack.
What the compatibility testing confirms
As part of compatibility testing, PaletteAI ran the NVIDIA Run:ai compatibility suite covering cluster provisioning, NVIDIA GPU Operator and NVIDIA Network Operator behavior, NVIDIA Run:ai deployment, GPU-accelerated workload scheduling, and upgrade paths. This compatibility testing established the integration pattern and testing discipline that carry directly into how we are delivering open-source NVIDIA Run:ai in PaletteAI. The published compatibility matrix helps customers identify which PaletteAI, Kubernetes, NVIDIA GPU Operator, and NVIDIA Run:ai versions were tested together and plan upgrades against the stated version combination.
Where this fits
Every AI cloud operator we work with is solving the same equation: build a platform that scales with demand and stays profitable while serving many tenants. GPUs are the expensive part, and idle GPUs erode margins — as do weak tenant isolation and slow stack buildout.
NVIDIA Run:ai on PaletteAI can help address those pressures with dynamic scheduling and policy-based resource allocation, tenant isolation at both the Kubernetes and NVIDIA Run:ai layers, one stack profile deployed consistently across every cluster, and PaletteAI’s blueprint-driven provisioning that can help shorten AI factory buildout timelines.
Regulated and sovereign AI environments can use this configuration as part of a controlled, repeatable deployment model. With NVIDIA Run:ai available as open source and integrated into PaletteAI, PaletteAI customers can deploy the same tested version combinations through the managed platform.
Try it in your environment
If you are on PaletteAI today and open source NVIDIA Run:ai is available in your PaletteAI Studio environment, deploy it against a new cluster or an existing one, and use the same lifecycle automation to keep it current as NVIDIA Run:ai and the rest of the stack evolve.
Get in touch to see it running on your infrastructure.