Together GPU Clusters
Reserved NVIDIA GPU clusters for training and large-scale inference; also takes a trained model to a production endpoint without switching providers.
Find alternatives to Together GPU Clusters
What it does
- GPU cloud
- Model training
Used for
- Model inference
Features 11
- Distributed training clustersReserve from 8 to 4,000+ GPUs connected over InfiniBand for distributed training.
- Resilient infrastructureContinuous health checks, automated remediation and self-serve node repair keep capacity online.
- Kubernetes-native servingLow-latency networking, Kubernetes-native deployment and flexible ingress for serving models.
- Shared filesystemsFully managed shared storage volumes attached to GPU clusters for training data and model weights.
- Project-level RBACCross-team access managed by project-level RBAC via CLI, SDK, API, Terraform or the web console.
- Slurm on KubernetesOffers direct SSH access with Slurm-style simplicity backed by Kubernetes resilience.
- Node autoscalingProvides node autoscaling for elastic compute on GPU clusters.
- Managed GrafanaIncludes managed Grafana for cluster observability.
- HA control planeProvides a highly available control plane with managed upgrades.
- Together Kernel CollectionKernel collection from Tri Dao that improves training and inference performance on the clusters.
- OIDC authenticationLets each team member authenticate to a cluster's Kubernetes API using their own identity via external OpenID Connect.
How it compares
-
CoreWeave AI Cloud
Together AI competes with CoreWeave by offering cloud-based GPU resources for AI and machine learning workloads.
Research report · 22 Oct 2026 -
Lambda Cloud
Lambda, founded in 2012, is a cloud-based GPU company that specializes in GPU workstations, cloud services, and AI infrastructure.
Research report · 19 Sep 2026 -
Pods
RunPod, founded in 2022, is a cloud-based infrastructure service that provides cost-effective and scalable GPU resources.
Research report · 19 Sep 2026
Sources
- Pricing
- Used for
-
Official documentation · 22 Oct 2026
'Go from trained model to production endpoint without switching providers.' (https://www.together.ai/gpu-clusters)
- Constraints
-
Derived from the security facts on this page, from the same source.
- Security
-
Company-wide Together AI trust center (Vanta, script-rendered; read in the in-app browser): SOC 2 Type 2 report, ISO 27001:2022 certificate and a Data Processing Addendum are listed.
- Sold within
-
Together's GPU Clusters page says "Both options are fully self-serve." and lists on-demand rates such as NVIDIA H100 at "$3.99" "/hr per GPU" with a "Create cluster" button; HGX B300 and NVL72 systems are "Contact us for pricing".
- Interfaces
-
Official documentation · 12 Oct 2026
Page: 'Manage cross-team access via project-level RBAC using the CLI, SDK, API, Terraform, or the web console'.
- Intended for
-
"Research labs, enterprises, and AI model creators training or serving at scale"
Also from Together AI
-
Together Inference Model API
Hosted API serving open-weight text, image and audio models, billed per token.
-
Together Fine-Tuning API service
A managed service for fine-tuning open-source models on a customer's own data and serving the result on Together's infrastructure.
-
Together Batch Inference API service
Asynchronous bulk inference for workloads that do not need a real-time response, priced below Together's serverless rate.
-
Together Custom Training Developer tool
Custom model training service covering supervised fine-tuning and direct preference optimization, billed per token.
-
Together Dedicated Container Inference Infrastructure service
Dedicated, reserved GPU containers for model inference with guaranteed performance, billed per GPU-hour.
-
Together Provisioned Throughput API service
Reserved inference capacity for open models, bought in Provisioned Throughput Units that each hold a fixed tokens-per-minute rate, with a 99% uptime SLA and overflow routed to Together's serverless fleet at list price.
-
Together Dedicated Model Inference Infrastructure service
Dedicated single-tenant GPU endpoints for open-weight or custom models pulled from Hugging Face or S3, with autoscaling, canary and rolling rollouts and multi-region failover, billed per GPU-hour.
-
Together Frontier AI Factory Infrastructure service
Private GPU cloud built to order for frontier-scale training, fine-tuning and inference, from 1,000 to 100,000+ NVIDIA GPUs with managed Kubernetes or Slurm and direct access to Together's research team.
Open Together AI in the directory
Something wrong here? Send a correction — quote this product id: together-gpu-clusters.