Weights
Model, licence, precision and memory footprint. The customer keeps model rights.
LLM hosting
Dedicated GPU capacity for self-hosted models.
GPU capacity and supported serving-stack deployment for self-hosted language models.
The self-hosting decision
Four layers, each stated before deployment.
Self-hosting a language model is a stack decision, not a card rental. The quote makes each layer explicit.
Runtime
A supported serving stack, qualified for the model and GPU before acceptance.
Host system
GPU, VRAM, host CPU, RAM, NVMe and network stated per configuration.
Region and access
India deployment with the region, access model and data path stated in the quote.
What residency gives you
A stated deployment region, INR billing and technical controls that can support a compliance program.
What it does notHosting location alone is not legal compliance. Model licences, application behaviour and independent review remain the customer's responsibilities.
Common questions.
Does self-hosting make an application compliant?
No. Region and technical controls can support a compliance program, but legal duties, application behaviour, policies and independent review remain separate.
Can you deploy vLLM or another serving stack?
Yes, where the model, GPU and software versions are supported and the deployment acceptance target is agreed.