Changelog
Weekly updates and improvements
May 20, 2025
New feature
H200s & simplified reservation billing

NVIDIA H200s now available for custom reservations
We’re excited to bring NVIDIA H200 capacity to Foundry Cloud Platform to support more resource-intensive workloads. In order to scale up capacity thoughtfully and provide the best user experience, we are initially making H200s available via custom short-term and long-term reservations. We plan to make H200 instances available via self-serve reservations and spot as more capacity becomes available.
If you’d like to set up a custom H200 reservation (short-term or long-term) please reach out to your Foundry account team or contact solutions@mlfoundry.com.
Self-serve reservation usage now billed monthly
To simplify making self-serve compute reservations through the Foundry console, we’ve updated reservation billing to occur in conjunction with Spot billing at the end of the month. Now, when making a reservation through the Foundry console, you will no longer be prompted with a checkout screen before your reservation is confirmed. Instead, the cost of the reservation will be included in your organization’s bill at the end of the month, similar to spot usage.
If you have credits available in your account when confirming your reservation, your credits will be applied immediately, and only the remaining balance on the reservation will be billed at the end of the month.
May 14, 2025
New feature
Attach persistent storage to existing spot bids & upgraded authentication flow

You can now update the persistent storage volumes for your spot bids when they are paused–giving you the flexibility to attach additional storage for your workloads without losing your bid configuration or boot disk.To update the persistent storage volumes for an existing bid:
Pause your bid using the ••• menu on the Bids or Instances page in the Foundry console
On your paused bid, select Edit volumes from the ••• menu
Select the volumes you'd like to add or remove
Unpause your bid
Next time your spot instances are allocated, mount the new storage volumes to your instance. You can find detailed instructions for mounting storage volumes in our file share and block storage docs.
We have also upgraded the authentication infrastructure for Foundry Cloud Platform, including minor login flow changes. This infrastructure is intended to support additional upcoming authentication and platform improvements.
May 5, 2025
Fixes
Fileshare bug fixes
We've addressed multiple bugs to improve the performance and stability of fileshare storage drives. In particular, very large fileshares and fileshares used by multiple users were the most impacted and should see the greatest improvement.
Apr 28, 2025
New feature
Pause Spot Bids

You can now pause and resume existing spot bids, allowing you to stop and restart your spot workloads without losing your bid configuration or boot disk(s).
To pause or resume a bid, you can use the ••• menu next to the bid on the Spot Bids tab in the Foundry console, or you can use the /v2/spot/bids/{bid_fid}
API endpoint to pause and resume programmatically.
When a bid is paused, it is immediately removed from the spot auction, shutting down any running instances and wiping ephemeral storage. However, the boot disks for the instances will be preserved, allowing you to maintain your environment and resume your workload quickly the next time your bid is resumed and allocated.
Apr 21, 2025
Improvements
Console updates: flexible sidebar and bug fixes

You can now view more information about your instances & orders without scrolling with our collapsible sidebar! Additionally, we have implemented various bug fixes to improve usability.
Apr 10, 2025
New feature
Monitor Foundry status & set required SSH keys

Foundry Platform Status
To make it easier to monitor the status of Foundry Cloud Platform, we've launched status.mlfoundry.com, and we now show status icons next to region IDs throughout the Foundry Console. Status icons indicate the current health of the region as Operational
, Degraded Performance
, Undergoing Planned Maintenance
, or Unavailable
. You can hover over the status icon for additional information and a link to the status page for full details. You can subscribe to updates from status.mlfoundry.com to be notified about any upcoming maintenance or active incidents.

Required SSH keys
Administrators can now set required SSH keys at the project level for telemetry and management purposes. When one or more SSH keys are marked as required, all new instances spun up within the project will include the required keys by default. When creating a new instance in the Foundry Console, the required key(s) will be preselected. When submitting a spot bid through the Foundry API, the required key(s) must be included in the launch specification for the request to succeed. You can find all required keys via the GET ssh-keys
endpoint.
Mar 11, 2025
Improvements
Foundry API v2

Intuitively manage your compute with the v2 API
The new v2 API makes it more intuitive to manage your spot compute programmatically and is up to 4x more performant. These performance improvements translate to the Foundry Console UI as well, where you can expect to see more consistent and significantly reduced load times.
With the v2 API, you can:
Place and update bids with improved controls
Get availability across regions
Create and manage volumes and SSH keys seamlessly
Coming soon: endpoints for reservations
To get started, check out our updated API quickstart guide and reference
With the launch of API v2, we plan to discontinue support for the v1 API on June 3rd, 2025. If you would like migration support from the Foundry team, please don’t hesitate to reach out.
Ephemeral storage is now auto-mounted to instances
Ephemeral drives will now be automatically mounted at /mnt/local when your instances start up–no manual configuration required. Upon preemption, drives are automatically reformatted and remounted.
Important: If your startup scripts include code to mount ephemeral drives, you should update them to remove this step.
Reservations are now available in us-central-1b
You can now make self-serve reservations in the us-central-1b region for workloads that require uninterrupted access to compute. Reservations guarantee access to capacity for 3 hours to 2 weeks and are proactively healed in the case of hardware failures to mitigate interruptions. If you require a reservation longer than 2 weeks, reach out to our team!