The below terms (the “Preview Terms”) apply to your participation in the Public Preview of, and your use of, the above-identified DigitalOcean Service, GPU Spot Capacity (the “Preview Offering” or “Spot”). The Preview Offering makes GPU Spot Capacity available both (a) as standalone GPU Droplets provisioned on Spot capacity (“Spot GPU Droplets”) and (b) as GPU-backed worker node pools provisioned on Spot capacity within DigitalOcean Kubernetes (“DOKS”) clusters (“Spot GPU Node Pools” or “node pools”). The Preview Terms supplement the terms of the DigitalOcean Terms of Service Agreement (including Service Terms) or such other written agreement between you and DigitalOcean governing your use of our Services (our “Agreement”). The Preview Offering is a Beta Service for purposes of the Agreement, and terms not defined in these Preview Terms have such meaning otherwise defined in the Agreement. If there is a conflict between these Preview Terms and the Agreement, these Preview Terms will control solely with respect to the conflict relating to your use of the Preview Offering.
Subprocessor Disclosure: A current list of DigitalOcean Service subprocessors is available here, including the subprocessors identified under “All DigitalOcean Services.”
Your participation in the Public Preview and use of the Preview Offering is voluntary. If you do not accept these Preview Terms, then you may not participate in the Public Preview.
The Preview Offering is a preliminary, limited-availability release of a DigitalOcean Service. It provides access to GPU compute capacity (“Spot Capacity”) drawn from DigitalOcean’s available fleet and offered on an interruptible, as-available basis at a separate Spot rate. Spot Capacity is made available both as standalone Spot GPU Droplets and as Spot GPU Node Pools within DOKS clusters. Additional terms applicable to Spot GPU Node Pools are set forth in Section 3. This offering is intended for testing, evaluation, and feedback purposes, and for running interruption-tolerant workloads. Access during the Public Preview is limited to eligible accounts, to the GPU configurations (shapes), regions and (for DOKS) cluster configurations that DigitalOcean designates, and to the capacity DigitalOcean makes available from time to time.
2.1 As-Is; As-Available. The Preview Offering is provided “AS-IS” and “AS-AVAILABLE,” whether provisioned as Spot GPU Droplets or as Spot GPU Node Pools. Spot Capacity is designed for fault-tolerant, interruption-tolerant workloads (for example, batch or checkpointed training, batch or asynchronous inference, hyperparameter sweeps, and rendering or simulation jobs). The Preview Offering is not intended or recommended for latency-sensitive, stateful, or production-facing workloads, or for any workload that requires guaranteed uptime, continuity, or availability. You acknowledge the experimental nature of the Preview Offering and that DigitalOcean may modify, suspend, or discontinue any feature of the Preview Offering at any time without notice.
2.2 No Availability Guarantee; Reclamation of Capacity. Spot Capacity is offered on an as-available basis and is not guaranteed. Because Spot Capacity is drawn from capacity that fluctuates with overall demand, DigitalOcean does not guarantee that Spot Capacity will be available at any given time, that any particular quantity or shape will be available, or that a Spot instance or Spot GPU Node Pool, once provisioned, will remain available for any minimum duration. DigitalOcean may reclaim, suspend, or terminate any Spot instance or Spot GPU Node Pool at any time, in its sole discretion, with or without regard to the state of your workload, and may do so in any order it determines (including to return capacity to other DigitalOcean services or uses, or based on operational, capacity, or business considerations). Reclamation of Spot GPU Node Pools occurs as described in Section 3.3, and DigitalOcean does not guarantee that autoscaling will be able to add or replace nodes as described in Section 3.4.
2.3 Reclaim Notice. Except as provided in Section 2.4, DigitalOcean will endeavor to provide at least two (2) hours’ advance notice before reclaiming a Spot instance or Spot GPU Node Pool. This notice is provided on a commercially reasonable, best-effort basis. DigitalOcean does not warrant that notice will be delivered, received, or provided any particular amount of time in advance, and Spot Capacity may be reclaimed with less than two (2) hours’ notice or, as described in Section 2.4, with no notice. Notice will be sent by email to the address associated with your account; for Spot GPU Node Pools, DigitalOcean may also emit Kubernetes-native signals as described in Section 3.3, but email remains the notice mechanism under these Preview Terms. Notice is deemed given when sent by DigitalOcean, whether or not actually received. You are responsible for maintaining a valid notification email address and for monitoring for reclaim notices and signals, and for configuring your workloads and tooling to respond to them. DigitalOcean is not responsible for any failure of delivery, receipt, or processing of a notice or signal that is attributable to your systems, configuration, or third parties.
2.4 Emergency and Immediate Reclamation. Notwithstanding Section 2.3, DigitalOcean may reclaim, suspend, or terminate Spot Capacity immediately and without prior notice where DigitalOcean determines, in its sole discretion, that doing so is necessary or advisable, including for emergency or unplanned maintenance; hardware, network, or facility failure; security, integrity, health, or safety reasons; to comply with law or legal process; to prevent harm to the DigitalOcean platform or other customers; or due to any event beyond DigitalOcean’s reasonable control. The two-hour notice described in Section 2.3 is a target for ordinary-course reclamation only and is not a service commitment.
2.5 Customer Responsibility for Fault Tolerance. You are solely responsible for designing your workloads to tolerate interruption, including by checkpointing progress and persisting state so that work can resume elsewhere. Spot Capacity carries no uptime, continuity, or data-retention guarantee. DigitalOcean will have no liability for any loss of work, data, state, progress, output, or time, or for any interruption, delay, or downstream loss, arising from the reclamation, suspension, termination, or unavailability of Spot Capacity, whether or not notice was provided. Any data on the boot disk or scratch disk associated with a Spot instance, or on the local, ephemeral, boot, or scratch storage of a Spot GPU Node Pool node, is lost upon reclamation.
2.6 Preview Pricing. Spot Capacity is offered as distinct SKUs and on-demand GPU node pools, separate from on-demand GPU Droplets, and appears as its own line item on your bill. You are billed hourly, per-second, at the Spot rate applicable to the relevant SKU. There is no bidding; you pay the Spot rate in effect for the applicable SKU, which DigitalOcean sets and may adjust over time based on the supply of and demand for Spot Capacity. For standalone Spot GPU Droplets, the Spot rate is set and maintained independently of, and is not a discount applied to, any on-demand rate; DigitalOcean makes no representation that Spot pricing reflects any particular level of savings relative to on-demand or other pricing. The Spot rate applicable to an active allocation is the rate in effect at the time that allocation was provisioned, and remains in effect for the lifetime of that allocation. DigitalOcean may change Spot rates for a SKU at any time on a prospective basis; any changed rate applies to allocations and node pools created after the change and does not apply to an already-active allocation or existing node pool. If you provision Spot Capacity again following reclamation, termination, or expiration, the new allocation is priced at the prevailing Spot rate at the time of that new request. Any directional or indicative pricing provided to you in advance of onboarding is non-binding and subject to change; final pricing is confirmed only upon provisioning.
2.7 Support; No SLA. No Technical Support commitments and no Service Level Agreements (SLAs) apply to the Preview Offering. No service credits are available for the Preview Offering, including for any unavailability, reclamation, interruption, autoscaling behavior, failure or timeout of an automated drain, failover behavior, or failure to provide notice. Support, if any, is provided on a reasonable-effort basis during business hours.
This Section 3 applies to Spot Capacity provisioned as Spot GPU Node Pools within a DOKS cluster, and supplements Section 2. If there is a conflict between Section 2 and this Section 3 with respect to Spot GPU Node Pools, this Section 3 controls.
3.1 Overview. A Spot GPU Node Pool is a node pool whose worker nodes are provisioned on interruptible Spot Capacity and attached to a DOKS cluster that you manage. Each node in a Spot GPU Node Pool is a Kubernetes node object and is labeled and tainted as Spot capacity (for example, doks.digitalocean.com/capacity-te=spot) so that workloads must explicitly tolerate or target Spot nodes. You are solely responsible for your cluster and workload configuration, including scheduling, labels, taints, tolerations, node affinity, PodDisruptionBudgets, priority configuration, and autoscaling settings.
3.2 Cluster Configuration Requirements. A DOKS cluster that contains a GPU node pool (including a Spot GPU Node Pool) must also contain at least one CPU node pool, which is required to run in-cluster system components. DigitalOcean may enforce this requirement through validation and may decline to create or maintain a cluster or node pool that does not meet it. DigitalOcean may designate the GPU shapes, regions, and cluster configurations eligible for Spot GPU Node Pools and may change them at any time.
3.3 Node Pool Reclamation; Automated Drain. Reclamation of Spot GPU Node Pools occurs at the node-pool level: when DigitalOcean reclaims Spot Capacity associated with a Spot GPU Node Pool, the entire node pool is reclaimed, and DigitalOcean does not reclaim individual nodes within a pool on a partial basis. Upon reclamation, DigitalOcean will initiate an automated sequence that cordons and drains the nodes in the affected pool and then deletes those nodes. The drain sequence is designed to observe configured PodDisruptionBudgets on a commercially reasonable, best-effort basis and is subject to a drain timeout, after which the nodes are deleted regardless of whether pods have been successfully evicted, rescheduled, or relocated. DigitalOcean will endeavor to emit native Kubernetes signals (such as Events and node conditions) when a reclaim is initiated, so that your schedulers and tooling may respond; DigitalOcean does not warrant that any such signal will be delivered, observed, or acted upon in time to prevent interruption or loss. The reclaim-notice, notice-timing, and emergency provisions in Sections 2.3 and 2.4 apply to Spot GPU Node Pools; the target two-hour notice is best-effort only, and reclamation may occur with less or no notice.
3.4 Autoscaling. If you enable autoscaling for a Spot GPU Node Pool, the DOKS cluster autoscaler may attempt to add Spot nodes up to the maximum you configure. Scale-up is subject to, and contingent upon, available Spot Capacity, and may fail or be paused when sufficient Spot Capacity is not available. DigitalOcean does not guarantee that any scale-up request will succeed, that any minimum or maximum number of nodes will be achieved or maintained, or that capacity will be available to add or replace nodes at any time. Nodes added by autoscaling are billed as provided in Section 3.5.
3.5 Node Pool Pricing; Price Lock. For a Spot GPU Node Pool, the applicable Spot rate is fixed at the time the node pool is created and applies to all Spot nodes in that pool for the life of the pool, including nodes subsequently added by autoscaling or manual scaling. Changes to the Spot rate for the relevant SKU after node pool creation do not apply to that node pool. Each Spot GPU Node Pool is billed at its locked Spot rate for the GPU-hours accrued by its nodes and appears as a line item on your bill reflecting that rate, the GPU-hours accrued, and the total charges for the applicable period. If a Spot GPU Node Pool is reclaimed, terminated, or deleted and you create a new Spot GPU Node Pool, the new node pool locks in the Spot rate in effect at the time of its creation, which may be higher or lower than the rate of the prior node pool. Except for the timing of the price lock described in this Section 3.5, the pricing terms in Section 2.6 apply to Spot GPU Node Pools.
3.6 Node Count Restrictions. After a Spot GPU Node Pool is created, you may maintain or reduce the number of nodes in the pool, but you may not increase it. Without limitation, DigitalOcean may prevent you from (a) increasing the node count of a fixed-size pool; (b) switching a fixed-size pool to autoscaling with a maximum node count above the pool’s current node count; © increasing the maximum node count of an autoscaling pool; or (d) switching an autoscaling pool to a fixed size with a node count above the pool’s current node count. To obtain Spot Capacity beyond a pool’s current size, you must create a new Spot GPU Node Pool, which will be priced at the then-current Spot rate under Section 3.5.
3.7 On-Demand Fallback. DigitalOcean may recommend that you provision an on-demand GPU node pool to serve as a fallback for workloads running on a Spot GPU Node Pool, but DigitalOcean does not require you to do so and does not provision one on your behalf. Configuring failover (for example, through labels, taints, tolerations, node affinity, priority configuration, and the cluster autoscaler) is your responsibility. If a Spot GPU Node Pool is reclaimed and no eligible fallback capacity is available, affected pods may remain pending until Spot or other eligible capacity becomes available. Any on-demand GPU node pool, and any other non-Spot resource, is not Spot Capacity, is not offered under these Preview Terms, and is billed at the applicable on-demand or other standard rate, which is separate from and not locked to any Spot rate. DigitalOcean does not guarantee that any failover or fallback will occur, succeed, or preserve workload state.
3.8 Customer Responsibility. You are solely responsible for designing and configuring your DOKS workloads to tolerate interruption and node-pool reclamation, including by checkpointing progress and persisting state to durable storage located outside the Spot GPU Node Pool. Local, ephemeral, boot, and scratch storage on the nodes of a Spot GPU Node Pool is lost when those nodes are reclaimed and deleted. The provisions of Section 2.5 apply to Spot GPU Node Pools, and DigitalOcean has no liability for any loss described in Section 2.5 arising from node-pool reclamation, automated drain, drain timeout, autoscaling behavior, failover configuration, or the unavailability of Spot Capacity.
4.1 Input Data. You are responsible for the data, models, code, and content you use with, or store on, the Preview Offering, and for compliance with the Agreement and applicable law with respect to that data. Given the interruptible and pre-production nature of the Preview Offering, you should not store the sole or authoritative copy of any data on a Spot instance or on a node of a Spot GPU Node Pool, and should maintain your own backups and checkpoints.
4.2 Improvement. By using the Preview Offering, you authorize DigitalOcean to use technical metadata (e.g., utilization, latency, error rates, reclaim events, autoscaling events, and usage frequency) and any Feedback you provide to operate, evaluate, and improve our infrastructure, the Preview Offering, and our Services.
4.3 Feedback. You may, but are not required to, provide DigitalOcean with suggestions, comments, or other feedback regarding the Preview Offering (“Feedback”). You grant DigitalOcean a perpetual, irrevocable, worldwide, royalty-free license to use and exploit Feedback for any purpose without restriction or obligation to you.
5.1 DigitalOcean may terminate or conclude this Preview Offering as a whole at any time without advance notice. For the avoidance of doubt, this Section 4.1 does not apply to the reclamation, suspension, or termination of individual Spot instances or allocations, which is governed by Sections 2.2 through 2.4.
5.2 If the Preview Offering advances to a subsequent release cycle (General Availability), DigitalOcean will endeavor to assist you with transitioning your use. Notwithstanding this, DigitalOcean reserves the right, in its sole discretion, to reclaim capacity, purge data associated with the Preview Offering, and terminate Preview Offering instances at the conclusion of the Preview Offering or at any time thereafter, regardless of any planned transition or migration.
DigitalOcean may suspend or terminate your access to the Preview Offering at any time (in our sole discretion), with or without notice, if: (a) we choose to discontinue the Preview Offering or transition it to General Availability; or (b) we believe your use violates these Preview Terms or the Agreement. Suspension or termination of your access to the Preview Offering does not affect the reclamation rights described in Section 2.
The Preview Offering is a Beta Service. The warranty disclaimers, exclusions of damages, and limitation of liability applicable to Beta Services under the Agreement (including Section 8.6) apply to the Preview Offering and are incorporated by reference. Sections 2.5 and 2.7 and Section 3 of these Preview Terms supplement, and do not limit, those provisions. Nothing in these Preview Terms expands DigitalOcean’s liability beyond, or limits any disclaimer, exclusion, or cap set forth in the Agreement.