Lilac partners with Saturn Cloud to expand GPU capacity for Token Factory
By Lucas Ewing
Lilac is partnering with Saturn Cloud to expand the GPU capacity behind Token Factory. The partnership enables Saturn to scale model serving and per-token inference on Lilac infrastructure without requiring customers to reserve new hardware.
Saturn Cloud announced that it selected Lilac as a GPU capacity provider for the platform. Token Factory manages the path from fine-tuning to a production endpoint; Lilac provides the GPU cloud beneath eligible serving workloads.
Why the partnership matters
Token Factory is built for inference demand that can change quickly. A new endpoint may be quiet for weeks and then need to scale immediately. Holding enough reserved hardware for every possible spike raises costs for Saturn's customers and for the operators serving them.
Adding Lilac gives Saturn another source of production GPU capacity while preserving the per-token experience Token Factory customers already use. For Lilac, the partnership brings production inference demand into the same capacity layer that supports its dedicated and serverless products.
The partnership also extends Lilac's support for NVIDIA-based workloads. NVIDIA named Saturn among the AI and MLOps providers embedding NVIDIA NIM in their platforms. Through Saturn, eligible serving workloads built with NIM, backed by TensorRT-LLM, or using Nemotron models can run on Lilac GPU capacity.
It also creates a path toward broader support for the AI factory stack NVIDIA describes through DSX. Lilac is evaluating DSX OS components such as NVIDIA Dynamo as its inference fleet grows.
What this unlocks for customers
A Saturn customer can move from a fine-tuned checkpoint to a production endpoint without separately procuring the GPUs needed for a traffic spike. Saturn continues to manage the model workflow, endpoint, and per-token billing. Lilac handles capacity qualification, scheduling, metering, and infrastructure support behind eligible workloads.
Lilac's dedicated customers can opt to release unused reservation windows through Lilac Flex. When Lilac matches one of those windows with demand, the customer receives bill credits. As the integration expands, Saturn inference workloads are expected to become one source of demand for eligible windows.
GPU cloud operators and enterprises can also use Saturn to offer model serving without carrying enough reserved capacity for every tenant's peak. The infrastructure may change as demand rises and falls; the Token Factory endpoint does not.
Why Saturn selected Lilac
Token Factory demand changes by model, customer, and launch. A cluster sized for peak traffic can sit underused in quieter periods; one sized for the baseline can run short when demand spikes.
Lilac gives Saturn dedicated capacity for predictable demand and a way to add capacity as traffic changes. Lilac Flex is designed to let Saturn release eligible windows from reserved clusters when those GPUs are not needed, reducing the cost of carrying capacity through quieter periods.
That is the operating model Saturn founder Sebastian Metti described in the announcement: route Token Factory workloads onto usable GPU capacity instead of waiting for new infrastructure.
Available now
Lilac capacity is now available to Saturn Cloud customers through Token Factory. Operators and enterprises can contact Saturn Cloud about using Lilac for model serving and per-token inference.
Startups looking for dedicated GPU VMs, bare-metal nodes, or clusters, as well as serverless inference, can explore Lilac GPU capacity.