> ## Documentation Index
> Fetch the complete documentation index at: https://docs.capital.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Deploy your GPUs

Click **Compute** in the rail. This is where contracts become machines.

![The Compute page](https://mintcdn.com/capital-ai/2r_nz6EQ_1nQxtlD/images/compute.jpg?fit=max\&auto=format\&n=2r_nz6EQ_1nQxtlD\&q=85\&s=0a399cdcc98e8260e499eb22694bbae3)

## What the page shows

* **Free Capacity** has one tile per contract you hold for the month shown, with the **GPU Contract** it pairs with and the **Available Contracts**: held, less any in a resting sell or already deployed. Step through months with the arrows, or switch **Month** to **Year** to see every month at once.
* **Balance** on the right lists every contract you hold. Filter it to **BASIS** or **GPU**.
* **Active Capacity** at the bottom is your deployments: the BASIS and GPU contracts, identifier, region, GPUs, term and **Status**.

![Free Capacity by year](https://mintcdn.com/capital-ai/2r_nz6EQ_1nQxtlD/images/compute-year.jpg?fit=max\&auto=format\&n=2r_nz6EQ_1nQxtlD\&q=85\&s=0e1b36a27909e7133a4972747b3d0306)

## What you need

A deployment pairs a provider's **BASIS** contract with the **GPU** contract for the same GPU and month, one for one. To deploy 8 GPUs of `BBA-10-26` you need 8 free `BBA-10-26` and 8 free `H100SXM-10-26`. Buy the BASIS on [Offtake](https://docs.capital.ai/app/offtake.md) and the GPU contract on the [markets](https://docs.capital.ai/app/trade.md), or the other way round.

Only the **current month** deploys. Later months show on the page but their tiles wait.

## Add a deployment

1. Click **Activate** on a tile, or **Add Deployment** in Active Capacity.
2. Check the **BASIS Contract**, its **Identifier**, **Region** and **Term**, and how many **Free BASIS contracts** and **Free GPU contracts** you have.
3. Enter **GPUs to deploy**, up to the **Max** shown.
4. Click **Continue**, then confirm.

![Add Deployment](https://mintcdn.com/capital-ai/2r_nz6EQ_1nQxtlD/images/deployment-add.jpg?fit=max\&auto=format\&n=2r_nz6EQ_1nQxtlD\&q=85\&s=c25a52b90ad1afea9ebb7241f522d8df)

The deployment appears in Active Capacity as **requested**. From here the contracts are locked: they stay yours, but can't be sold until the deployment ends.

## What happens next

| Status | Meaning |
| - | - |
| **requested** | Waiting for ops. You can still **Cancel request**. |
| **provisioning** | Ops has claimed it and the machines are being provisioned on Northflank, our hosting partner. |
| **active** | Your GPUs are running. **Open** on the row takes you to them, and the row's drawer has the Northflank CLI commands for your project. Where ops provisioned by hand, access details come by email. |
| **ending** | You stopped it, or the month ended, and ops is tearing it down. |
| **ended** | Done. The contracts are free again. |

## Stop a deployment

Click **Stop** on the row. You can stop all of its GPUs or some of them; stopping some splits them into their own row so ops tears down exactly those machines. A deployment that ops hasn't claimed yet ends at once; a running one shows **ending** until it is torn down.

At the end of the month every deployment ends and its contracts are released. The contracts stay yours: a deployment locks them, it does not use them up.

From code: [Deployments](https://docs.capital.ai/compute/deployments.md).


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.