Saturn Cloud’s token factory platform now supports NVIDIA DSX OS. For neocloud operators, that means you can run a production inference service on NVIDIA infrastructure, and actually sell it, without assembling the stack yourself.
NVIDIA DSX OS is open, modular software with a strong set of NVIDIA’s own components for operating AI infrastructure at scale. It gives you the building blocks for inference, scheduling, and fleet operations: NVIDIA Dynamo, NVIDIA Grove, the KAI Scheduler, NVSentinel, and Fleet Intelligence. What it doesn’t include, by design, is the layer on top, the per-token metering, per-tenant billing, and chargeback that turn served tokens into something you can price and sell. That layer is inherently the operator’s to build, and it’s the part we provide.
How they work together
On the serving side, Saturn Cloud runs NVIDIA Dynamo as the distributed inference and serving layer, with disaggregated prefill and decode. Dynamo is engine-agnostic, so it drives vLLM, SGLang, or TensorRT-LLM underneath. NVIDIA Grove orchestrates the multi-node workloads. Scheduling runs on the NVIDIA KAI Scheduler, which places workloads with GPU awareness, allocates fractional GPUs, and enforces quotas across tenants. NVIDIA Fleet Intelligence and NVSentinel handle fleet-wide health monitoring and failure detection. You can serve NVIDIA Nemotron models alongside other open and partner models.
On top of all of it, Saturn Cloud adds per-token metering, quotas, and billing per tenant. You get a serving stack built on NVIDIA’s tools and a way to sell what it produces, in one platform.
“NVIDIA DSX OS gives AI cloud operators a strong foundation for serving models,” said Sebastian Metti, founder of Saturn Cloud. “We add the piece that lets them turn it into a service they can sell to their own customers, instead of just renting out GPU hours.”
One platform, single- or multi-tenant
Saturn Cloud runs your open and partner models in production on a single, production-grade control plane that’s versioned and managed. Your customers fine-tune on your own infrastructure and serve the results alongside the base models, with no separate training stack to stand up. You serve each model the way the workload demands, dedicated single-tenant for isolation, or shared multi-tenant for density, and Saturn Cloud handles the tenant boundary, quotas, and per-tenant metering either way. Data and inference stay inside your boundary, under the security and governance controls that regulated customers require.
Usage is tracked per tenant against quotas, so faster hardware and better utilization turn into margin rather than a lower hourly rate. The result is infrastructure you can sell, not just operate. That’s the token factory idea in practice.
Runs on your GPUs, wherever they are
Saturn Cloud deploys across public cloud, private cloud, and on-premises environments, and runs on NVIDIA AI infrastructure wherever you have it, including neoclouds adopting the NVIDIA DSX platform stack.
For telcos, sovereign AI clouds, and enterprise operators, the shape is the same. NVIDIA’s tools serve the tokens. Saturn Cloud does the production engineering that turns them into a service you can run and sell.
“AI factories are moving into production, and operators are looking to expand their offerings with value-added services,” said Warren Barkley, vice president of product management at NVIDIA. “Saturn Cloud’s integration with NVIDIA DSX OS, including per-token metering and tenant billing, gives customers AI production-ready services customers want, without the complexity of assembling and integrating the stack themselves.”
Try it
The Saturn Cloud platform integrated with NVIDIA DSX OS is available now for AI cloud operators to evaluate. Get in touch or get started at saturncloud.io.
