Skip to content

AI Economics

AI Compute Economics monitors token usage, GPU spend, and model-level costs across AI and LLM services — from AWS Bedrock and Azure OpenAI to GCP Vertex AI and OpenAI's API.

When to use it

  • You want to understand which AI models, teams, or workloads are driving AI spend.
  • You need to allocate AI infrastructure costs to teams or products.
  • You want to compare cost efficiency across models (cost per million tokens).
  • You're tracking the ROI of AI adoption alongside cloud spending.

Open AI Economics

Go to Cost Management > AI Economics from product navigation.

Analyze AI spend

1. Filter and group

Apply filters to narrow your analysis:

  • Provider — AWS, Azure, GCP, or OpenAI
  • AI Provider — specific AI service provider (e.g., OpenAI, Anthropic via Bedrock)
  • Service Family — GPU compute, inference, embedding, etc.
  • Model — specific model name (GPT-4, Claude, Gemini, Llama, etc.)

Choose granularity: Daily, Weekly, or Monthly.

Group trends by Provider, AI Provider, Service Family, or Model.

2. Review the Overview tab

The Overview tab shows total AI spend, token volume, and trend over the selected period.

3. Explore the Models tab

Select a model to open its detail drawer, which includes:

Section What it shows
Spend Gross cost, net cost, credits, discounts, and tax
Usage Token type, total tokens, spend, and cost per 1M tokens
Services Spend by service, region, and account
Allocation Tags Tag key, value, and coverage for cost attribution
Evidence Source signals and row coverage

Supported AI services

CloudVerse AI recognizes spend from: Amazon Bedrock, Amazon SageMaker, Vertex AI (GCP), Gemini API, Azure OpenAI, Azure AI Search, and OpenAI API. See Integrations to connect an AI platform.

Best practices

  • Compare token usage and spend together — a cost increase is not always a volume increase (prices change).
  • Use allocation tags to identify unowned or shared AI spend.
  • Check credits and discounts before comparing gross and net cost across periods.
  • Treat blank model names as a data quality issue to resolve at the source.
  • Set a budget on AI spend to catch runaway model usage early.