GPU Compute for Every Scale
From your first model to enterprise production workloads — Tensormesh grows with you. All plans include access to our distributed training infrastructure and inference optimization layer.
- Up to 8 GPUs (A100 equivalent)
- 500GB model storage
- Training jobs: 10 per month
- Inference: 1M requests/day
- Standard support (business hours)
- 3 user seats included
- Up to 64 GPUs (A100 / H100 mix)
- 5TB model storage
- Training jobs: unlimited
- Inference: 50M requests/day
- Priority support — 4h response SLA
- 10 user seats included
- Custom model registry
- Unlimited GPU allocation
- Unlimited model storage
- Dedicated cluster option
- Inference: unlimited requests
- 24/7 dedicated support — 30min SLA
- Unlimited users + SSO / SAML
- On-premise deployment option
Frequently Asked Questions
What GPU types are available?
Tensormesh clusters run NVIDIA A100 and H100 GPUs across our distributed data centers. Scale and Enterprise plans have priority access to H100 nodes for the most demanding workloads.
Can I upgrade or downgrade my plan?
Yes. You can upgrade at any time with immediate effect. Downgrades take effect at the start of the next billing cycle. No penalties or lock-in periods on any plan.
How is inference usage measured?
Inference requests are measured per API call to your deployed model endpoint. Requests are counted regardless of model size or response length. Burst capacity is available on Scale and Enterprise plans.
Is my data secure and isolated?
All plans run in isolated compute namespaces. Model weights, training data, and inference inputs are never shared across tenants. Enterprise customers can opt for dedicated physical clusters.
Do you offer annual billing discounts?
Yes — annual billing is available on all plans at a 15% discount versus monthly billing. Enterprise contracts can also include volume-based discounts negotiated at signing.
What frameworks and runtimes are supported?
Tensormesh supports PyTorch, TensorFlow, JAX, Hugging Face Transformers, and vLLM out of the box. Custom runtime containers are supported on Scale and Enterprise plans.