Neocloud and AI Factory operators can now turn bare-metal GPU infrastructure into a fully managed, white-label AI platform with per-token billing and production inference.
NEW YORK, June 9, 2026 /PRNewswire/ — Saturn Cloud, the AI development platform for GPU cloud operators, today announced its Token Factory platform, which gives enterprise AI teams a complete fine-tuning and model serving workflow on top of the operator’s GPU infrastructure. The platform lets neocloud operators, AI Factory builders, and enterprises offer their customers managed fine-tuning jobs, dataset management, and OpenAI-compatible inference endpoints, all metered per token and delivered under the operator’s own brand, without having to build or maintain any of it in-house.
The Gap Between AI Infrastructure and Enterprise AI
GPU cloud operators have invested heavily in accelerated infrastructure. NVIDIA Grace Blackwell, NVIDIA Blackwell, and NVIDIA Hopper systems are being deployed at scale, and neocloud revenue is growing rapidly. For many operators, the business model has been overly simple: rent GPU hours at a per-hour rate.
However, the needs of enterprise organizations extend beyond raw compute. Increasingly, enterprise organizations need managed development environments, distributed training orchestration, model fine-tuning pipelines, SSO and RBAC, usage tracking, and compliance tooling. Building this platform infrastructure in-house takes months of engineering and ongoing maintenance that most GPU cloud operators are not staffed to deliver.
“Operators shouldn’t have to build an AI development platform from scratch just to make their GPU infrastructure useful to enterprise teams. Saturn Cloud gives them managed environments, training orchestration, fine-tuning, OpenAI-compatible inference endpoints, and per-token billing on day one,” said Sebastian Metti, Founder, Saturn Cloud.
How Saturn Cloud Works
Saturn Cloud’s Token Factory platform gives AI teams a way to fine-tune and serve open models without managing infrastructure. The workflow is straightforward: upload a dataset, configure a fine-tuning job, and deploy the resulting model to an inference endpoint, all within the operator’s branded environment.
Fine-tuning jobs support supervised fine-tuning (full-weight and LoRA) on open models, with automatic DeepSpeed multi-GPU configuration when the selected instance has multiple GPUs. Users specify a base model, dataset, and a small set of hyperparameters; Saturn Cloud renders a complete training config and handles orchestration, retry, and checkpoint output. Supported training frameworks include Axolotl, vLLM, Unsloth, TRL, PEFT, and DeepSpeed.
Datasets are typed, validated collections of training data in conversational, instruction, text, or pretokenized formats. Users can upload datasets directly, import from external sources (S3, NFS), or curate data in a managed workspace and then register it as a Token Factory dataset. All dataset storage uses high-performance parallel filesystems rather than object storage, eliminating the cold-start tax that degrades GPU utilization during training jobs.
Checkpoints and artifact lineage are managed automatically. When a fine-tuning job completes, the resulting checkpoint is registered in Saturn Cloud’s artifact registry, preserving the full lineage from training run to model weights. Checkpoints are immediately available as inputs to inference endpoint deployments.
Inference endpoints deploy fine-tuned or base models as persistent serving deployments backed by vLLM, with per-deployment subdomains, health monitoring, and per-token metering. The serving configuration (dtype, max context length, quantization) is rendered at deployment time and no custom serving scripts are required.
The entire workflow is isolated per organization. Token Factory resources are scoped to the tenant, so datasets, checkpoints, and endpoints from one customer are never visible to another.
What Operators Get
Saturn Cloud gives GPU cloud operators a turnkey path from bare-metal infrastructure to a revenue-generating AI platform. The operator-facing layer includes white-label branding, per-token and per-GPU-hour billing infrastructure, tenant onboarding and self-service provisioning, usage dashboards and chargeback reporting, and enterprise security tooling including SSO, RBAC, and SOC 2 compliance.
Without a platform layer, operators sell compute hours and compete on price. With Saturn Cloud, they sell a platform and compete on developer experience, security posture, and time to production. Saturn Cloud enables operators to pass enterprise security reviews because the compliance tooling is already in place. It lets operators show tenants usage dashboards, cost controls, and team management. It gives the operator’s sales team a product demo, not a spec sheet.
What Engineers Get
AI teams and developers working on the operator’s infrastructure get managed development environments with JupyterLab, VS Code, RStudio, and SSH access; distributed multi-GPU training with orchestration, retry, and logging; Token Factory for fine-tuning and deploying open models; and pre-configured NVIDIA CUDA, GPU drivers, and AI framework support. Engineers have access to the operator’s full GPU fleet, including NVIDIA Hopper, Blackwell, and Blackwell Ultra systems, including NVIDIA GB200 NVL72 rack-scale systems. Saturn Cloud is a member of the NVIDIA Inception program for startups.
Saturn Cloud integrates with infrastructure automation partners across the ecosystem, including Mirantis k0rdent AI, Spectro Cloud, OpenNebula, and Rafay. Operators managing Kubernetes directly on cloud backends can also deploy Saturn Cloud on top of their existing stack without changing their infrastructure layer.
Availability
Saturn Cloud’s Token Factory capabilities are now available to GPU cloud operators, neoclouds, and enterprises operating their own GPU infrastructure. Organizations interested in deploying the platform can contact Saturn Cloud for an evaluation.
About Saturn Cloud
Saturn Cloud is the AI token factory platform for neoclouds, AI Factory operators, and enterprises. The platform provides managed fine-tuning, OpenAI-compatible model serving with per-token billing, managed environments, distributed training, and enterprise security and governance. Saturn Cloud supports GPU architectures and deploys across public cloud, private cloud, and on-premises environments. Learn more at saturncloud.io.
SOURCE Saturn Cloud