# AI Spend

> Track the cost and consumption of AI and LLM workloads across your connected cloud accounts

---

Site [llms.txt](/llms.txt)

---

The AI Spend page provides teams with dedicated visibility into the cost and consumption of generative AI and large language model (LLM) workloads across their connected cloud accounts, covering spend trends, token economics, cost attribution, and anomaly detection in one place.

<div class="alert alert-info" role="alert">


AI Spend is enabled per tenant. AI usage is currently detected in AWS billing data: Anthropic and other marketplace LLMs, native Amazon Bedrock, and AWS AI and machine-learning services such as SageMaker.
</div>


## Access the AI Spend page

1. Log in to the CloudBolt Platform.
1. In the sidebar, select **AI Spend**.

## AI Spend overview

| Section | Description |
| --- | --- |
| **Spend Trend** | Displays total AI spend over the selected period alongside the percentage change versus the prior period, charted over time with one series per model.<br>- Use the **Time grain** control to switch between **Daily**, **Weekly**, and **Monthly** bars.<br>- Use the **Account** filter to focus on a single cloud account (shown when multiple accounts have AI spend).<br>- Hover over a bar to see detailed spend for that interval. |
| **Token Economics** | Summarizes token consumption and efficiency for the selected period.<br>- Shows estimated savings versus uncached input pricing.<br>- Breaks down token volume (Mtok) by input, output, cache read, and cache write.<br>- Displays cost per Mtok by model, so different models' prices sit side by side rather than blending. |
| **Attribution** | Breaks down AI spend by cloud account for the selected period.<br>- Shows each account's share of total spend.<br>- Attribution is at the cloud-account grain; finer-grained keys such as inference profiles and cost-allocation tags appear as they become available in billing data. |
| **Anomalies & Run Rate** | Projects month-end AI spend and highlights unusual usage.<br>- Displays a projected total for the current month once enough complete days are available, with a comparison against last month.<br>- Flags spike days: days where spend exceeds twice the trailing 7-day average. |

Each card has its own period selector: **30 Days** (default), **60 Days**, **90 Days**, **6 Months**, or **12 Months**. Periods end at the most recent complete day of billing data.

## How AI spending is measured

- **Spend** figures are the billed cost of all detected AI usage, including marketplace LLMs (Anthropic and others), native Amazon Bedrock, and AWS AI and machine learning services.
- **Token metrics**, such as token volumes, cost per Mtok, and cache savings, currently reflect usage in the LLM marketplace (Anthropic models on Amazon Bedrock). Other AI spend appears in the spend, attribution, and anomaly cards but not in Token Economics.
- **Cache savings** are estimated from your own billing data: for each cache-read token, the difference between the uncached input price and the cache-read price actually paid, computed per model, account, and region. No external rate card is used.
- **Run rate** projects the current month from complete days only. Because cloud billing data lands one to two days behind, the newest day or two are treated as partial and excluded from the projection and spike detection.