Heterogeneous cloud dispatch

Dispatches workloads to the lowest failure-adjusted spot cost across cloud providers.

problem it solves
Prevents lock-in to single cloud vendor pricing spikes and regional capacity limits.

What it does

What it does: Ranks and dispatches requests to the cheapest healthy GPU node across multi-cloud providers.

What it watches: Real-time spot prices, node egress fees, health scores, and data-residency rules.

When it triggers: On every request routed across attached multi-cloud GPU fleets.

The Action: Selects the optimal provider node clearing SLA requirements at all-in minimum cost.

How it recovers: Cleanly falls back to secondary cloud providers if primary tier capacity exhausts.

At least two attached cloud accountsrequired
Credentials for multi-cloud GPU providers (AWS, GCP, CoreWeave, Lambda).
Data residency tags configuredrecommended
Pin sensitive workloads to required geographic regions.

Mode	Effect on your request	What you can see
off	Not consulted. Requests dispatch to static single-cloud endpoints.	No heterogeneous_dispatch stage recorded.
shadow	Ranks candidate nodes and logs counterfactual cost savings without shifting traffic.	Stage with action=would_dispatch_cloud and optimal_vendor logged.
prod	Dispatches requests to the lowest cost healthy cloud provider node.	Selected cloud provider, spot price, and cost savings logged.

Merit order algorithm	Spot price x failure-adjusted throughput	Power-grid style economic dispatch.
Price update frequency	60s polling	Tracks spot price fluctuations.
Residency enforcement	Strict geographic pinning	Respects compliance constraints.

·Cross-cloud data egress fees must be factored into all-in request cost calculations.
·Different cloud providers may have slight variations in GPU node network latency.
·Requires scoping cloud provider API permissions in ACE control plane.