Nodus Compute is a GPU compute platform for running training, agents and inference on GPUs you already own or on reserved capacity. It appeared on YC Launches on 9 November 2026. The idea is to make the graphics cards sitting in offices, labs and data centers useful again: connect them to Nodus Compute, run your AI work on them, and lend out idle compute to earn credits you can spend later.
Buying GPU time from a cloud provider can be expensive and inflexible, especially for teams that already own hardware. Nodus Compute takes a different route. It starts with the machines you have, adds reserved capacity when you need more, and gives you one place to manage training runs, agents and inference, instead of juggling separate setups for each kind of work.
Details below reflect the official site as checked on 10 November 2026.
What is Nodus Compute?
Nodus Compute is built around two ideas. The first is BYOCompute, short for bring your own compute, which puts your own machines first. You connect GPUs you already own, and Nodus Compute helps you run work on them. The second is Liquid Capacity, a way to share idle compute with others and earn credits in return. Together, they turn hardware that might sit unused between jobs into a resource you can use or trade.
For work that needs guaranteed access, the platform also offers Reserved Compute: dedicated GPUs and clusters you can reserve, so a training run or a production agent does not have to wait for spare capacity.
Key features
- BYOCompute: run AI work on GPUs you already own, with your own machines first.
- Liquid Capacity: lend idle compute to others and earn credits for it.
- Reserved Compute: dedicated GPUs and clusters for work that needs guaranteed access.
- One place for training, agents and inference, rather than separate setups for each.
- Published GPU-systems research from the team behind the platform.
- Support for running agents as well as training and inference workloads.
Who is it for?
Nodus Compute fits teams and individuals who already own GPUs and want to get more value from them, whether that is a lab with cards that sit idle overnight, a company with on-premise hardware, or a developer with a powerful machine at home. It also fits teams that need GPU capacity for training or inference but would rather reserve dedicated hardware than compete for shared cloud capacity. If your GPUs are idle part of the time, lending that capacity for credits can help pay for the reserved compute you need when your own work peaks.
Pricing
Nodus Compute lists hourly rates in US dollars on its official pricing page. The listed rates are: H100 SXM 80GB at $2.600 per hour, A100 80GB at $1.090 per hour, H200 at $3.593 per hour, B200 at $5.500 per hour, RTX 4090 at $0.290 per hour, RTX 3090 at $0.180 per hour, and L40S at $0.793 per hour. Pricing may change; see the official website for current pricing.
Sources and verification: information in this article is based on the official Nodus Compute website and its pricing page, checked on 10 November 2026.
Image: nodus-compute.ai (official visual)
