AI / LLM Spend Tracking

AI workloads are among the fastest-growing cost centers in the cloud, yet they are often spread across multiple providers and billing systems. CostBeacon consolidates AI spend into a single dashboard so you can track total cost, compare providers, and spot trends without switching between portals.

CostBeacon tracks four categories of AI spend:

• Azure OpenAI — Inference and fine-tuning costs from Azure's hosted OpenAI models.
• GPU Virtual Machines — Compute costs from GPU-accelerated VM series used for training and inference.
• OpenAI — API spend from your OpenAI organization account.
• Anthropic — API spend from your Anthropic account.

Azure OpenAI costs are automatically detected from your Azure cost data — no additional configuration is required. CostBeacon identifies resources with the Microsoft.CognitiveServices provider type and the OpenAI kind, then maps meter names to friendly model names (e.g. gpt-4o, gpt-4o-mini, text-embedding-ada-002).

This mapping lets the AI Spend dashboard break down costs by model, so you can see exactly how much you are spending on GPT-4o inference vs. embedding generation vs. fine-tuning, even if they share the same Azure resource.

CostBeacon automatically identifies GPU virtual machines by matching the VM size against known GPU-accelerated series: NC, ND, and NV families. Any VM whose size starts with one of these prefixes (e.g. Standard_NC24ads_A100_v4) is classified as a GPU VM and included in the AI Spend totals.

This is especially useful for teams running self-hosted models, training jobs, or batch inference on dedicated GPU compute. These costs often hide inside general “Virtual Machines” line items in Azure billing, making them easy to overlook.

The dashboard groups GPU VMs by series and size, so you can compare the cost-efficiency of different hardware tiers side by side.

To track spend from OpenAI or Anthropic, add your API keys in Admin → Integrations. CostBeacon uses these keys to pull usage and cost data from each provider's billing API on a 12-hour sync cycle.

For OpenAI, CostBeacon calls the /v1/organization/costs endpoint to retrieve per-model cost breakdowns. For Anthropic, it uses the /v1/usage API to fetch token-level usage, then calculates costs based on published per-token pricing.

API keys are stored securely in Azure Key Vault and are never exposed in the CostBeacon UI after initial entry. See the Integrations documentation for setup details.

The AI Spend page brings all four categories together into a single view. At the top, KPI cards display:

• Total AI Spend — Combined cost across all providers for the selected period.
• Azure OpenAI — Cost from Azure-hosted OpenAI models.
• GPU Compute — Cost from GPU VM series.
• Third-Party — Combined OpenAI and Anthropic API spend.

Below the KPI cards, a model breakdown table lists every model with its total cost, token count (where available), and percentage of overall AI spend. A trend chart shows daily AI spend over time, stacked by provider, so you can see how the mix is shifting.

Use the time-period toggle (7d / 30d / 90d) to adjust the window. All cards, tables, and charts update in real time when the period changes.