New B200 spot capacity is live in US East from $1.69 per GPU-hour. See availability
Documentation

The limits, written down.

Everything the platform refuses to do, and the number behind the refusal. If a limit blocks something reasonable, it is a conversation, not a wall.

Last reviewed 15 September 2026 · something wrong or missing?

Instances and nodes#

LimitValueNote
GPUs per instance1 to 8Depends on the model; the exact sizes are on each model page
GPU models47NVIDIA and AMD, data-center, workstation and consumer
Regions3US East, US West, EU Central
Instances per accountLimited by capacity and balanceNo hard count; a deployment needs free capacity of that model in that region
Instance name60 charactersOptional; a reference like i-7f3a2c is always assigned
Multi-node clustersOn requestAnything wider than one node is arranged as reserved capacity

Volumes#

LimitValueNote
Volume size10 GB to 4,000 GBIn steps of 10 GB, chosen at creation
Resize in placeNot availableCreate a larger volume and copy across while both are attached
AttachmentsOne instance at a timeA volume cannot be shared between two running instances
RegionFixed at creationA volume cannot move between regions; copy through an instance
Volumes per accountNo fixed limitEach is billed at $0.08 per GB-month while it exists
DeletionOnly by youAn attached volume must be detached first; nothing is deleted on our initiative

Balance and spending#

LimitValueNote
Minimum top-up$40No top-up fee; credit does not expire
Balance needed to deployAt least one hour of the instance rateThe platform will not start what it cannot bill for an hour
Billing granularityOne started minuteHourly price ÷ 60, no minimum duration
Spend capOptional, monthlyInstances are stopped before the cap is crossed; volumes are kept
Balance at zeroInstances stopVolumes survive and keep being billed; nothing is deleted without contact
Payment confirmationOn-chain confirmationsA payment already sent is credited even if you close the page

Account and security#

LimitValueNote
Failed sign-insThrottledRepeated failures slow down, then block, attempts from that address
Two-factor authenticationTOTP, 8 recovery codesRecovery codes are shown once; store them outside the account
SessionsListed per deviceRevoke one device or all the others from Settings
SSH keysSeveral per accountInstalled on every instance you launch
API keysCreated in the consoleShown once at creation; revoke and rotate at any time
Support ticketsAccounts with credit or a live resourceEveryone can read the help centre and use the contact form

API and webhooks#

The API and the CLI are in private beta.

Rate limits are set per account when access is enabled, and the defaults below are what a beta account gets. Ask for access and we will tell you where the queue stands.

LimitValueNote
Requests120 per minute per keyRead endpoints; bursts above it answer 429 with Retry-After
Write requests20 per minute per keyCreating and stopping instances and volumes
Page size100 itemsCursor pagination on every list endpoint
Webhook delivery5-second timeoutRetried with a backoff for up to an hour
Metadata endpointNo limitLink-local, inside the instance; poll every 5 to 15 seconds

Raising a limit#

Capacity limits are physical, and we will tell you plainly when something is not available. Everything else — a larger volume, more instances at once, a higher API rate, a cluster wider than one node — is a message away. Tell us what you are running, the shape you need and the dates, through the contact form or a ticket in the console if you already have an account.