Private Cloud Burst

Cloud Burst is the Enterprise path for adding temporary external capacity without transferring the provider account or infrastructure bill to ColabHive. Private Cluster remains the default path.

Path	Where work runs	Commercial model
Private Cluster	Customer-controlled Intel, NVIDIA, AMD and CPU nodes	Pro, Team or Enterprise production license; Community (no license fee for labs, research and personal use under BSL 1.1)
Share Hive	Capacity enabled by an account owner	Explicit opt-in, unselected by default
Private Cloud Burst	Temporary nodes in the customer's DigitalOcean account	Enterprise add-on; paid to the provider, no ColabHive fee

Ownership and billing

1. The customer connects its own DigitalOcean account and credential.
2. DigitalOcean bills the customer directly for provisioned capacity.
3. ColabHive provisions, enrolls, drains and releases eligible burst nodes.
4. ColabHive adds no fee on the provider charges: the capacity is the customer's own.

The add-on is separate from the USD 35,000/year Enterprise Private Cluster package. An agreed spend cap belongs in the customer agreement; Cloud Burst never implies unlimited capacity.

Last resort, by design

Cloud capacity carries a higher marginal cost than already-connected capacity. The planner therefore uses Burst only after local placement cannot satisfy sustained demand:

1. Sustained deficit — a momentary spike does not provision a node.
2. Local options exhausted — compatible private capacity is preferred first.
3. Independent corroboration — pressure or queue age confirms the deficit.
4. One change at a time — the planner observes the effect before further expansion.

Training jobs count as demand. A job that needs more accelerator memory than the private fleet can provide may use an eligible burst node.

Join, drain and release

The lifecycle is provision → install runtime → enroll → heartbeat → active. Time to first useful request depends on provider availability, image availability, model size and warm-up time.

When capacity is no longer needed, the node is cordoned and drained. In-flight work is allowed to finish, compatible private placement is restored, and the provider node is then released. A node running a training job is not destroyed mid-run.

Operator controls

• node-count limits;
• hourly and monthly spend fields when configured;
• rental pacing and a minimum rental window;
• per-node cost tracking and reconciliation against the customer's provider bill;
• an alert stream and manual kill switch;
• a protected-infrastructure deny-list;
• provider-region priorities supported by the connected account.

When capacity cannot be provisioned, excess demand remains subject to admission and queue behavior; the add-on does not turn provider stock into an availability guarantee.

Contract boundary

• Provider capacity, price and geography remain DigitalOcean dependencies.
• ColabHive does not publish a universal residency or provider guarantee.
• Customer-specific residency, maximum spend and workload restrictions must be written into the applicable deployment agreement.
• Support-response commitments do not create an uptime SLA.

See also: Private Agentic Infrastructure · Platform overview · Pricing

---

Authors: J.L. Minich, M. Lucius