GPU cloud providers
Committed capacity to grow your footprint into new regions.
Capacity and offtake
Clean AI capacity, sold the way you buy.
We design each project's commercial model with its buyers, across the layers of the AI stack and across the regions demand comes from. Offtake is contracted during development, so the capacity you commit to is backed by a funded project.
Commercial layers.
Buy raw infrastructure, hosted models or finished workloads, whichever fits the way you build.
| Layer | Priced as | Suited to |
|---|---|---|
| Infrastructure | $/GPU-hour, reserved or committed | Neoclouds, AI labs, hyperscaler overflow and resellers |
| Model hosting | Dedicated or managed capacity | Model providers and enterprises serving their own models |
| Agentic workloads | $/million tokens | Enterprises and platforms running agentic AI at scale |
Two kinds of workload.
Demand is global, but not all workloads can travel. We plan for both.
Latency-insensitive
Global buyers
Workloads that can run wherever clean power and capacity are available.
- Training and fine-tuning
- Batch inference and evaluation
- Buyers in North America, Europe, the UK and Asia-Pacific
Latency-sensitive
Regional buyers
Workloads that need to run close to the people and systems they serve.
- Real-time inference and agentic applications
- Enterprises, financial institutions and the public sector
- Data kept in the region
Who buys from a Caldera project.
Neoclouds
Enterprises
Regional and global
Dedicated capacity for your own models and agentic workloads.
Resellers
Channel and marketplaces
Wholesale capacity to package and resell.
Model providers
Labs and platforms
Hosting and inference capacity on clean power.
Capacity
Tell us what you need to run.
Share your workload, scale, timing and preferred commercial terms, and we'll show you where it fits in our pipeline.