vLLM
Indicative from-price range
Indicative from-price range for self-hosting vLLM on 1 server: about $24 to $112 per month (USD), as of 2026-10-19, for the reference workload (Largest model size (B parameters): 7; Peak concurrent requests: 32; Max context length (tokens): 4096). From-price by provider, each the cheapest server in that provider's own shortlist: $24 on DigitalOcean (basic-2vcpu-4gb, 2 vCPU, 4 GB RAM, fra1); $49 on Google Cloud (e2-standard-2, 2 vCPU, 8 GB RAM, us-central1); $63 on Amazon Web Services (c9g.large, 2 vCPU, 4 GB RAM, us-east-1); $64 on OVHcloud (rise-1, 6 vCPU, 32 GB RAM, eu-west); $66 on Microsoft Azure (D2as_v7, 2 vCPU, 8 GB RAM, eastus); $66 on Hetzner (AX41, 6 vCPU, 64 GB RAM, eu-central); $262 on Servers.com (dl-01, 8 vCPU, 64 GB RAM, eu-west).
Reference workload
- Largest model size (B parameters): 7
- Peak concurrent requests: 32
- Max context length (tokens): 4096
Methodology
This range is generated automatically from public provider price lists. It has not been reviewed by a person, it is not a quote, and it may differ from the total the calculator chooses for your own workload with its AI step. Method: the calculator's offer shortlist for the reference workload (the same ranking it builds before the AI step, up to 12 offers across providers; 12 here) is priced at its monthly list price in USD. The low end is the cheapest of the per-provider from-prices below and the high end is the median server of the combined shortlist, each multiplied by 1, the smallest baseline topology (single-gpu-node). The from-price for each provider applies the same method to that provider's own shortlist (the calculator's ranking restricted to that provider), so it is the cheapest server that ranking offers there, not a recommendation. Prices are taken from the price snapshot dated 2026-10-19; prices quoted in EUR are converted with the ECB reference rate dated 2026-10-19. Servers with fewer than 2 vCPU or less than 4 GB RAM are not priced. Only server rental list prices for the product's own servers are counted: dependencies, licences, support, traffic charges, taxes and setup fees are not included.
Price sources
- Amazon Web Services: https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonEC2/current/us-east-1/index.json
- DigitalOcean: https://www.digitalocean.com/pricing/droplets
- Google Cloud: https://cloudbilling.googleapis.com/v1/services/6F81-5844-456A/skus
- Hetzner: https://www.hetzner.com/_resources/app/data/app/live_data_en_EUR.json
- Microsoft Azure: https://prices.azure.com/api/retail/prices
- OVHcloud: https://eu.api.ovh.com/1.0/order/catalog/public/eco?ovhSubsidiary=FR
- Servers.com: https://www.servers.com/products/scalable-bare-metal
Size vLLM for your own workloadMarkdown versionFor AI agents (llms.txt)
vLLM is an open-source inference and serving engine for large language models that uses PagedAttention to manage GPU key-value cache memory efficiently. It enables high-throughput continuous batching and distributed inference behind an OpenAI-compatible API.
Baseline topologies
| Topology | Minimum nodes | Roles |
|---|---|---|
| single-gpu-node | 1 | vllm-server |
| multi-gpu-node | 1 | vllm-server-tensor-parallel |
What drives the cost
- Largest model size (B parameters)
- Largest model being served, in billions of parameters. Primary driver for GPU VRAM requirements and host memory sizing.
- Peak concurrent requests
- Peak simultaneous inference requests. Drives KV cache memory sizing, continuous batching throughput, and host CPU scheduling overhead.
Where it runs
bare_metal, dedicated, cloud_vm
Size vLLM and get a cost estimate