Control AI and GPU spend
Track AI cost by model, team, and token across OpenAI, Anthropic, and Bedrock, and keep GPU spend in check.
Connect every AI source
| Source | How |
|---|---|
| OpenAI | Admin usage key |
| Anthropic | Admin usage key |
| Amazon Bedrock | Included with AWS |
| Azure OpenAI, Vertex AI | Included with Azure and Google Cloud |
| GPUs | Compute instances in each cloud, and Kubernetes nodes |
Build an AI cost canvas
Ask Iris to build a canvas:
- Monthly AI cost by model and account
- GPU & AI spend analysis multicloud
- Bedrock input vs output tokens by model, last 30 days
Catch new models early
AI spend often starts as New spend anomalies, when a team tries a new model. Keep High-severity anomalies → ticket on, and filter Anomalies to the Data & AI category.
Reduce AI cost
- Model opportunities flag workloads on a more expensive model than they need.
- Compare input and output token cost. Long prompts and verbose outputs are often the cheapest fix.
- Bring token counts in as a dataset (such as
token_usage) to build cost per 1K tokens by team or feature.
GPU spend
- Group compute by instance type and filter to GPU families.
- Check utilisation in the resource panel before right-sizing.
- Schedule GPU instances used for experiments.