How Vast.ai billing works and what stopped instances cost

What stopped instances actually cost

When you stop an instance in the Vast.ai console, the system detaches your container from its assigned GPU. You stop paying the base active rental cost for the compute hardware right at that second, and the GPU becomes available for someone else to rent.

The disk allocation does not disappear when the instance stops. Your files stay on the host machine so you can start the instance back up later, and your data can still be copied off the machine while it is stopped. Because that disk space remains reserved on the host drive, Vast.ai continues billing you for storage for every single second the instance exists and is online. Stopping an instance does not avoid storage costs.

If you leave a stopped instance sitting in your dashboard for days or weeks, that storage charge keeps eating into your credit balance. Instances showing in your account are never free, even if your balance is zero or negative.

To completely stop all charges from an instance, you have to destroy it. Once an instance is destroyed, its storage allocation is wiped from the host, and disk billing stops immediately. The only time an existing instance does not bill you for active rental or disk storage is when the host machine itself is currently offline. You are also not charged when an instance status reads "Loading".

The three separate billing components

Pricing on Vast.ai is not a single bundled rate. Every host sets their own pricing for their hardware, so the exact numbers vary from offer to offer on the marketplace. When you look at an offer, the price is split into three distinct items:

• Active rental (GPU): Quoted in $/hr and billed for every second your instance is in an active or connected state.
• Storage costs: Quoted in $/GB/hr and calculated directly from the size of the disk allocation you configure. This runs every second the instance exists, in all states other than offline.
• Bandwidth costs: Quoted in $/TB and billed for every byte sent or received by the machine, regardless of whether the instance is active, idle, or stopped.

You can inspect the full breakdown before launching by hovering your cursor over the listed price on the Search page or on your Instance page in the console.

Bandwidth is where unexpected charges usually hide. Vast.ai bills upload and download bandwidth on a raw usage basis. Because network transfer is not a constant hourly burn, it is not included in the main $/hr or $/day rates shown in the search filters. If your workload downloads large model weights or streams training datasets, your credit balance will drop faster than the hourly price suggests. You can verify the host's transfer rates by checking the "Internet:" line inside the pricing tooltip.

Instance pricing tiers

Vast.ai rents compute under three different contract tiers: On-Demand, Interruptible, and Reserved. Each tier targets a different type of workload and risk tolerance.

On-Demand gives you guaranteed uptime with no interruptions and per-second billing. You can spin machines up or down whenever you want, which makes it the standard choice for interactive sessions and production serving.

Interruptible instances run on spare host capacity at a discount of 50%+ compared to on-demand pricing. The tradeoff is that these instances are preemptible. When capacity is needed elsewhere, your instance can be reclaimed. They work best for batch training runs, rendering, or fault-tolerant jobs where your code regularly checkpoints progress to disk or remote storage.

Reserved instances are built for sustained, predictable usage. They offer up to 50% off standard pricing in exchange for a 1, 3, or 6 month commitment term, giving you guaranteed capacity and volume discounts on steady hardware.

Across all tiers, Vast.ai bills per second without rounding up to the nearest hour. There are no minimum hours required to spin up an offer.

Balances, autobilling, and the $0.00 cutoff

Vast.ai operates strictly on prepaid credits. You must add funds before you can launch an instance, and your current balance appears in your account dashboard. The platform recalculates balances about once every few seconds based on your running instances and storage allocations.

You can purchase credits using a credit card through Stripe, or with cryptocurrency through BitPay and Crypto.com. Vast.ai passes credit card data directly to Stripe and does not see, store, or process your raw card numbers.

To keep long-running jobs from failing unexpectedly, you have two balance management options:

1. One-time manual top-ups: Click the add credit button in your billing dashboard to add funds as needed.
2. Autobilling: Enable autobilling with a saved credit card to automatically charge the card whenever your account balance drops below a set threshold. Autobilling is only available for credit card payments; it cannot be used with crypto.

The official documentation recommends setting your autobilling threshold to roughly match your average daily or weekly spend. In addition, you should configure a low-balance email notification at roughly 75% of your autobilling threshold. That 75% mark gives you an early warning if an automated card transaction declines, leaving you time to fix payment details before the balance completely drains.

Running out of credit carries immediate consequences. When your account balance reaches $0.00 or goes negative:

• Running instances are stopped automatically, cutting off active GPU charges.
• Your instances, storage volumes, and stored data are scheduled for deletion.
• Disk storage charges continue accumulating against your balance while instances remain in the stopped state.

If your balance hits zero and you have a credit card on file, the system will periodically and automatically charge the card to cover the negative balance, saving your instances from being wiped. For accounts with an established usage history, Vast.ai also provides a short grace period where your balance can dip negative before deletion starts. The platform calculates the length of this grace buffer based on your account's average daily spend.

If you do not have a saved credit card on file when the balance reaches zero or below, your instances and stored data will be destroyed to prevent unpaid storage use.

Once credits are spent on active compute or storage, Vast.ai provides no refunds. If you have unspent credits that you deposited via credit card, you can request a refund by contacting support through the website chat. Unspent credits purchased via cryptocurrency or legacy Coinbase Commerce cannot be refunded.

Invoices, charge tracking, and team credits

You can manage payments and review past burn by navigating to the Account section in the sidebar and selecting the Billing page.

At the bottom of the page, the Transaction History area opens by default to the Invoices tab. This view logs every manual deposit, autobilling charge, credit transfer, and refund on the account. To pull records for accounting or tax purposes, click the Export button in the top right corner of the table to generate an invoice across a custom time span.

Next to that tab is the Charges tab, which lists an itemized breakdown of costs across your active instances, storage volumes, and serverless workers. Clicking the dropdown arrow on any individual charge line reveals the exact split between active GPU usage, reserved disk storage allocations, and raw upload or download network transfer.

If you work with colleagues, you can reallocate funds using the Transfer Credits button on the billing page:

• Personal transfers: You can send credits to another Vast.ai user using their account email address. Transfers between personal accounts are permanent and irreversible, so verify the email address before confirming.
• Team transfers: You can move credits into a team you belong to. If you are the team owner, you can transfer credits from the team pool back to your personal balance by switching into the team context and opening the Billing page from there.

Tracking your Vast.ai burn with Costbrake

If you run instances across multiple providers, tracking individual balances and hourly storage burn gets messy. Costbrake watches what you spend across 23 cloud, AI, and SaaS platforms, including Vast.ai, RunPod, AWS, and GCP, alerting you before small instances turn into large billing surprises.

Costbrake tracks your Vast.ai spend by reading your account balance drop between balance checks and alerting on anomalies. A spike alert triggers if an hour costs more than 3x your 7-day hourly baseline and at least $5 more than usual. You can also configure budget alerts at 50%, 80%, and 100% of your target, plus month-end forecasts based on your recent 72-hour burn rate.

For hands-on control, Costbrake provides an opt-in killswitch called brakes. The Vast.ai brake can stop a running instance straight from an alert without opening the Vast.ai dashboard. Stopping an instance ends its active GPU rental costs immediately, though the disk is retained on the host and storage charges will continue to bill until you destroy the machine. Costbrake never fires a brake automatically; an admin must confirm the action.

You can set up monitoring via read-only API keys by connecting your accounts manually or by instructing your coding assistant to run the agent setup.