# GPU and accelerator capacity by type, count and term

> Source GPU or accelerator capacity: chip type, count, hours or term, region, interconnect and how firmly you would commit, specified so a provider can answer at once.

Nodehour: The cluster you need, specified to the node. Send a requirement.

## What it is for

A team training or serving a model needs a specific accelerator in a specific quantity, often with a fast interconnect between nodes, for a known number of hours or a contract term, sometimes in a region set by data rules. Availability of large clusters is uneven, and smaller buyers struggle to get term pricing. Set out the chip, count, duration, region, interconnect and commitment so a provider can say quickly whether it has the configuration.

Machine-learning engineers and research groups, startups training or fine-tuning models, inference teams planning capacity, and agents provisioning compute for them.

## What happens next

You get a reference number and your configuration as a written specification. We look at every requirement, and when we can find a provider with that configuration or get it quoted, we reply to the email you leave. Nothing is ordered or committed until you agree terms. Contact details are kept apart from the requirement and deleted after 180 days.

## What you get

A reference number, a specification of the compute you need (chip, count, hours or term, region, interconnect, commitment) providers can answer at once, and a reply when we find one.

## Scope

Any region; data-residency requirements are recorded as stated.

- Prices and availability you mention are recorded as your statement, not checked against any provider.
- Export controls and data-residency rules that apply to your workload are yours to determine.

## Send a requirement

Fill in the configuration below and send it. Agents can POST the same fields to /compute-capacity/requests or call the submit_request tool.

POST https://nodehour.forevenza.com/compute-capacity/requests with JSON like:

```json
{
  "action": "lease",
  "need": "64 NVIDIA H100 SXM GPUs (8 nodes) with InfiniBand between nodes, bare metal with Slurm, in the US or EU, for one month from 1 December to train a 30B-parameter model.",
  "specifications": {
    "accelerator": "NVIDIA H100 SXM 80GB",
    "accelerator_count": 64,
    "gpu_hours": 46080,
    "term": "RESERVED_1_MONTH",
    "region": "United States or European Union",
    "interconnect": "INFINIBAND",
    "start_date": "2026-12-01",
    "deployment": "SLURM"
  },
  "quantity": {
    "amount": 46080,
    "unit": "GPU-hours"
  },
  "timing": {
    "earliest": "2026-12-01",
    "latest": "2026-12-31"
  },
  "geography": {
    "text": "United States or European Union"
  },
  "budget": {
    "amount": 2.2,
    "currency": "USD",
    "per": "unit"
  }
}
```

Schema: https://nodehour.forevenza.com/compute-capacity/schema.json

## Who runs it, and what it costs

Nodehour. Operated by Tempered Technologies LLC, 8 The Green STE B, Dover, DE 19901. Privacy questions and deletion requests: privacy@forevenza.com.

Sending a requirement is free.

Privacy notice: https://nodehour.forevenza.com/privacy/