Pricing for compute, storage and models.
Rates are set per customer while the first sites come online. Every line below opens a conversation, not a checkout.
Two ways to buy.
Both include Irish hosting and zero data retention as standard.
GPU instances
Dedicated GPUs and whole nodes we operate in Ireland. On-demand by the hour, no commitment. Contracted on 12 to 36 month terms, billed every hour, at a lower rate for the certainty it gives both sides. Rates on request.
Per GPU per hour, ex VAT. Contracted rates sit below on-demand and apply to 12 to 36 month terms, billed every hour of the term; whole nodes carry a small discount. Capacity is allocated in order of enquiry as each site comes online.
CPU instances
The rest of the pipeline, on the same network as your GPUs. By the hour, or reserved by the month.
Storage
Two tiers, both kept in Ireland: performance storage beside the GPUs (block, shared filesystem, registry) and object storage. Rates on request. Local scratch comes with the GPUs.
No ingress or egress fees between tandemcompute storage and tandemcompute instances.
Model API
contact for pricing →One OpenAI-compatible API. Run a hosted model per token, fine-tune it on your data, or take a dedicated endpoint on a fixed monthly line.
Inference
Hosted open models on a shared endpoint.
per 1M tokensFine-tuning
Customise a hosted model on your own data.
per 1M training tokensDedicated endpoints
GPUs reserved for one model and one customer.
fixed monthlyPrivate AI & managed workloads
Open models on your own documents, or a job we run end to end.
fixed monthlyInference
The same three tiers as the compute page: fast and everyday and mid-size models on the shared API, flagships on dedicated endpoints. Billed per 1M input and output tokens, with cached input discounted. Rates on request.
Rates are quoted per customer and billed in euro; committed monthly volumes are priced at contract rates. Bring your own or a fine-tuned model under private AI & managed workloads. See the model library for capabilities.
Fine-tuning
Customise a hosted model on your proprietary data. Billed per training token; the tuned model then runs on a dedicated endpoint.
Models that train on a single node. Larger models are fine-tuned under private AI & managed workloads.
Dedicated endpoints
An endpoint for an open or fine-tuned model on GPUs no one else touches. Priced as a fixed monthly line, sized to the model.
Private AI & managed workloads
Open models on your own documents, or a rendering, transcription or batch job we run end to end. One fixed monthly line, scoped with you. Start a conversation.